1. Chromosomes and Gene Expression
  2. Genetics and Genomics

Controlling gene activation by enhancers through a drug-inducible topological insulator

  1. Taro Tsujimura  Is a corresponding author
  2. Osamu Takase
  3. Masahiro Yoshikawa
  4. Etsuko Sano
  5. Matsuhiko Hayashi
  6. Kazuto Hoshi
  7. Tsuyoshi Takato
  8. Atsushi Toyoda
  9. Hideyuki Okano
  10. Keiichi Hishikawa  Is a corresponding author
  1. Department of iPS Cell Research & Epigenetic Medicine, Keio University School of Medicine, Japan
  2. Department of Physiology, Keio University School of Medicine, Japan
  3. Apheresis and Dialysis Center, Keio University School of Medicine, Japan
  4. Division of Tissue Engineering, University of Tokyo Hospital, Japan
  5. Department of Oral and Maxillofacial Surgery, University of Tokyo Hospital, Japan
  6. Department of Genomics and Evolutionary Biology, National Institute of Genetics, Japan
https://doi.org/10.7554/eLife.47980
Share this article
Cite this article
  1. Taro Tsujimura
  2. Osamu Takase
  3. Masahiro Yoshikawa
  4. Etsuko Sano
  5. Matsuhiko Hayashi
  6. Kazuto Hoshi
  7. Tsuyoshi Takato
  8. Atsushi Toyoda
  9. Hideyuki Okano
  10. Keiichi Hishikawa
(2020)
Controlling gene activation by enhancers through a drug-inducible topological insulator
eLife 9:e47980.
https://doi.org/10.7554/eLife.47980
  2. Download BibTeX
  3. Download .RIS

Abstract

While regulation of gene-enhancer interaction is intensively studied, its application remains limited. Here, we reconstituted arrays of CTCF-binding sites and devised a synthetic topological insulator with tetO for chromatin-engineering (STITCH). By coupling STITCH with tetR linked to the KRAB domain to induce heterochromatin and disable the insulation, we developed a drug-inducible system to control gene activation by enhancers. In human induced pluripotent stem cells, STITCH inserted between MYC and the enhancer down-regulated MYC. Progressive mutagenesis of STITCH led to a preferential escalation of the gene-enhancer interaction, corroborating the strong insulation ability of STITCH. STITCH also altered epigenetic states around MYC. Time-course analysis by drug induction uncovered deposition and removal of H3K27me3 repressive marks follows and reflects, but does not precede and determine, the expression change. Finally, STITCH inserted near NEUROG2 impaired the gene activation in differentiating neural progenitor cells. Thus, STITCH should be broadly useful for functional genetic studies.

Introduction

Interaction of genes and enhancers is greatly affected by architectural proteins that bind to chromatin and organize folding of the genome (Dekker et al., 2017). Most notably, CTCF mediates loop formation of chromatin in association with a cohesin complex, which physically bundles two distant loci of the genomic DNA (Parelho et al., 2008; Phillips-Cremins et al., 2013; Wendt et al., 2008). The genome-wide contact maps of chromatin show that the CTCF-binding sites often demarcate boundaries of so-called contact domains or topologically associating domains (TADs), where chromatin association takes place more preferentially inside than outside (Dixon et al., 2012; Phillips-Cremins et al., 2013; Rao et al., 2014). The looping between two CTCF-binding sites is mostly established where they are in the converging orientations with each other (de Wit et al., 2015; Guo et al., 2015; Rao et al., 2014; Vietri Rudan et al., 2015). Loss of cohesin or CTCF resulted in disappearance of contact domains (Gassler et al., 2017; Nora et al., 2017; Rao et al., 2017; Schwarzer et al., 2017; Wutz et al., 2017). According to the extrusion model, the cohesin ring extrudes the chromatin fiber from a site of loading and pauses at a CTCF-binding site that is oriented towards the ring (Fudenberg et al., 2016; Sanborn et al., 2015). This model is widely accepted as the underlying mechanism for the formation of the loops and contact domains.

On the other hand, several studies have shown that the CTCF boundaries limit the action ranges of enhancers and thus restrict the enhancer targets to genes within the same contact domains as the enhancers reside in Dowen et al. (2014); Lupiáñez et al. (2015); Symmons et al. (2014); Tsujimura et al. (2015); Tsujimura et al. (2018). These results are interpreted that CTCF demarcates contact domains, which then serve as entity to restrict or facilitate gene-enhancer interaction within themselves (Schoenfelder and Fraser, 2019). In the above studies, however, the gene-enhancer regulation was investigated primarily with respect to CTCF/cohesin and their binding sites in the genome, but not directly to the contact domains. Therefore, it remains elusive if contact domains per se have instructive roles in gene-enhancer interaction, or CTCF/cohesin directly regulates the interaction separately from creating contact domains.

Nonetheless, considering the apparent importance of CTCF, engineering the genome based on the CTCF function can add a new layer to the techniques of artificially controlling gene expression. The classical insulator element identified in the chicken β-globin locus (cHS4), which harbors a CTCF-binding site (Bell et al., 1999), has been utilized in heterologous systems (Bessa et al., 2014). However, the mechanistic investigation of these elements was limited. Therefore, the general utility of these elements as a tool was not very evident. In this respect, re-examining synthetic CTCF binding elements in light of the current understanding of chromatin regulation is desired to explore the utility of CTCF for genome engineering.

Also, a recent study showed that the SETDB1 repressive complex negatively regulates CTCF binding probably through heterochromatin formation involving KRAB zinc-finger proteins around the binding sites at the clustered protocadherin locus (Jiang et al., 2017). Currently, the generality of CTCF regulation by heterochromatin formation is unclear. Besides, it is not shown how such epigenetic change would affect the enhancer blocking activity of CTCF binding regions. Nonetheless, the possibility of artificially controlling CTCF binding is quite attractive in terms of genome engineering.

The Tfap2c-Bmp7 locus in mice is partitioned into two contact domains by a region termed TZ (Tsujimura et al., 2015; Tsujimura et al., 2018). The TZ also limits target ranges of enhancers at the locus (Tsujimura et al., 2015). The TZ consists of two arrays of CTCF-binding sites in divergent orientations with each other. Serial mutagenesis has shown that this configuration underlies the strong ability of the TZ to block chromatin contacts (Tsujimura et al., 2018). Taking advantage of the well-characterized nature of the TZ, in this study, we developed a new system to control the interaction between a gene and an enhancer. We first reconstituted the CTCF-binding sites of the TZ as a short DNA cassette, which successfully functioned as an enhancer blocker. Further, we added a feature that enables epigenetically controlling the blocking activity of the cassette in a drug-inducible manner. Here we describe the system, demonstrate its utility to study gene regulation by enhancers, and discuss the future potential of the system.

Results

STITCH blocks the interaction of MYC with the enhancer when inserted in between

To newly develop an artificial genomic insulator cassette to switch on and off the gene-enhancer interaction, we reconstituted arrays of binding sites of CTCF derived from the TZ present at the mouse Tfap2c-Bmp7 locus. The TZ consists of seven binding sites of CTCF: they are L1, L2, L3, L4, R1, R2, and R3, arrayed in this order from the Tfap2c side to the Bmp7 side (Figure 1A; Tsujimura et al., 2018). L1-L4 are oriented towards Tfap2c and collectively referred to as L, while R1-R3 are towards Bmp7 and referred to as R. The seven sites are constantly called as peaks of CTCF binding in different cell types by ChIP-seq (Chromatin immunoprecipitation followed by sequencing) with cross-linking. However, native-ChIP (nChIP) failed to detect CTCF binding at L1 and L4, suggesting the binding there is weak or indirect (Tsujimura et al., 2018). We extracted the 178 or 179 bp DNA sequences carrying the motif sequences for CTCF binding and concatenated them as a short DNA cassette. We embedded the core sequence of the tetracycline operator (tetO) at four different positions within the cassette. tetO is bound by the tetracycline repressor (tetR), but not in the presence of doxycycline (DOX), and thus allows recruitment of a linked effector protein to the cassette in a drug-dependent manner (Gossen and Bujard, 1992). We also put a puromycin-resistant gene (PUROr) sandwiched by two loxP sites for the sake of efficient targeting (Figure 1A, Supplementary file 1B). We expected that the CTCF-binding sites of the cassette would recruit CTCF and function as a topological insulator and that the tetO/tetR system would enable epigenetically modifying the insulation activity. We named the cassette as Synthetic Topological Insulator with TetO for Chromatin-engineering (STITCH) (Figure 1A).

Figure 1 with 2 supplements see all
Download asset Open asset
Serial insertion of STITCH around MYC localized the enhancer.

(A) Design of STITCH and scheme of inserting the cassette. After recombination of the two loxP sites (rectangles), the puromycin resistant gene is removed. The orientations of the CTCF binding motifs are represented by the orientations and colors of the triangles. Note that binding of CTCF at L1 and L4 was detected by nChIP neither in the endogenous locus of the mouse genome nor at STITCH in the MYC locus, as represented by the paled color (see Figure 1—figure supplement 1C). The ovals represent tetO. The sequences of these elements are shown in Supplementary file 1B. (B) The H3K27ac profile and the insertion sites of STITCH around MYC in the human iPS cells. The Hi-C map and the contact domains in human ESCs are shown at the top (Dixon et al., 2015). The Hi-C contact map was generated with the 3D Genome Browser (http://3dgenome.org) (Wang et al., 2018). The ChIP-STARR-seq profiles and annotated super-enhancer regions in human naïve and primed ES cells (Barakat et al., 2018) are also depicted. The triangle flags indicate the positions and orientations of the CTCF binding sequences identified in this study. Note that the algorithm that we used could not determine the binding motif of one site represented by a rectangle flag. The 3 Mb region deleted from one of the two alleles to make ‘Hap’ is indicated by the dashed line, flanked by scissors that indicate the target sites of CRISPR/Cas9. The numbers in the insertion names indicate the distance from MYC. (C) The 4C-seq profiles from VP-MYC2 of the wild type (Hap) and STITCH-30kb, +30kb, and +440kb alleles. (D) Relative MYC expression levels normalized with ACTB expression in the different alleles. Each dot represents replicate clones (see Materials and methods for details). The bars represent their means. (E) The 4C-seq profile of del(30-440) from VP-MYC2. The numbers indicate the ratios of sequence reads mapped to given intervals within the locally haploid 3 Mb region around MYC except for the 10 kb region from the viewpoint fragment (C, E).

Figure 1—source data 1

4C-seq read counts in the given intervals.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig1-data1-v1.xlsx

MYC is highly expressed in human pluripotent stem cells (Knoepfler, 2008). As MYC expression is regulated by long-range enhancers in various cell types, we thought that MYC expression in the stem cells should also be dependent on long-range enhancers (Bahr et al., 2018; Cho et al., 2018; Dave et al., 2017; Herranz et al., 2014; Hnisz et al., 2013; Lovén et al., 2013; Pulikkan et al., 2018; Shi et al., 2013; Sur et al., 2012; Uslu et al., 2014; Zhang et al., 2016). Hence, we used the human induced pluripotent stem cell (iPSC) line 253G1, which was generated via retroviral transduction of OCT4, KLF4, and SOX2 but without MYC, to test the functionality of STITCH (Nakagawa et al., 2008). A previous study called a large contact domain around MYC in human embryonic stem cells (ESCs) spanning almost 3 Mb (Dixon et al., 2015; Figure 1B). A super-enhancer region is annotated within the neighboring long non-coding RNA (lncRNA) gene PVT1 in ESCs based on ChIP-seq for histone H3 lysine 27 acetylation (H3K27ac), the enhancer associated histone modification, and ChIP-STARR-seq (self-transcribing active regulatory region sequencing, after chromatin immunoprecipitation) (Barakat et al., 2018; Figure 1B). Similarly, a super-enhancer was annotated within the same region in the mouse ESCs (Witte et al., 2015). We also confirmed the broad deposition of H3K27ac around there in the human iPSC line (Figure 1B).

Since the diploidy would hamper the following genome editing procedures, we first deleted one allele of the 3 Mb region around MYC as described before (Tsujimura et al., 2018) to make the locus locally haploid, and termed the clone as ‘Hap’ (Figure 1B). Then we inserted STITCH into five different positions of the remaining allele of the locus: ‘STITCH+30kb’, ‘STITCH+440kb’, ‘STITCH+1760kb’ and ‘STITCH+1790kb’ have the STITCH insertions away from the MYC promoter for the indicated distances to the telomeric side of the q arm of the chromosome (the right side on the map, Figure 1B); ‘STITCH-30kb’, at the 30 kb upstream from MYC (the left side, Figure 1B). STITCH+30kb and STITCH+440kb flank the super-enhancer and PVT1. STITCH+1760kb and STITCH+1790kb flank a peak of H3K27ac (Figure 1B).

We first performed 4C-seq (Circular chromatin conformation capture assay followed by deep sequencing) from the MYC promoter as a viewpoint to see how STITCH impacts on the chromatin conformation. We designed two sets of primers around the MYC promoter as viewpoints of 4C-seq (VP-MYC1 and VP-MYC2, see Figure 1—figure supplement 1A). In the wild type allele of Hap, MYC mainly contacts with the PVT1 region and around (Figure 1C, Figure 1—figure supplement 1B). In STITCH+30kb, STITCH+440kb, and STITCH-30kb, the contacts were blocked at the inserted positions of STITCH as expected (Figure 1C, Figure 1—figure supplement 1B). We then extracted RNA from the cells and measured the MYC expression levels with quantitative PCR (qPCR). We found that only STITCH+30kb strongly down-regulated the MYC expression, while the others did not (Figure 1D). These results suggest that the region between STITCH+30kb and STITCH+440kb (+(30-440)kb region) possesses the enhancer for the MYC expression. We made a deletion clone of the region, termed del(30-440) (Figure 1E, Figure 1—figure supplement 2). While the 4C contact profile of MYC extended further away from the deleted region (Figure 1E, Figure 1—figure supplement 1B), MYC was strongly down-regulated by the deletion (Figure 1D), showing that the region contains the responsible enhancer. nChIP-seq in STITCH+30kb confirmed that each of the binding sites of STITCH, except L1 and L4, was bound by CTCF as in the endogenous mouse genome, showing that these CTCF bindings are recapitulated regardless of the genomic context in human iPSCs (Figure 1—figure supplement 1C). Thus, STITCH recruits CTCF and blocks the gene-enhancer interaction when located in between as an insulator.

Of note, the nChIP-seq also identified endogenous sites directly bound by CTCF. In this study, we performed in total six nChIP-seq, including the two replicates from STITCH+30kb (Figure 1—figure supplement 1C) and the other following four that are two replicates from two different conditions (see Figure 5F, Figure 5—figure supplement 1). We collected peaks that are called at least in two out of the six experiments as reliable binding sites of CTCF for this study. Then we mapped the sites and orientations (Figure 1B,C). As indicated, MYC carries two CTCF-binding sites directed to the right side near the promoter region. These sites may account for the directional bias of the MYC contact towards the right side in WT(Hap) (Figure 1C). At the left side border of the large contact domain of the locus, a CTCF-binding site oriented to the right side was detected. The contact of MYC in STITCH+30kb appears to extend up to this boundary (Figure 1C).

Insulation and deletion of the enhancer resulted in similar transcriptome profiles

We employed RNA-seq to understand how the insulation (STITCH+30kb) and deletion (del(30-440)) of the enhancer affect the transcriptome of the cells through the down-regulation of MYC (Figure 2, Figure 2—figure supplement 1). Of note, deletion and duplication of the whole PVT1 genic region, as well as the knockdown experiment via RNAi, has suggested a role for the PVT1 lncRNA in MYC activation (Tseng et al., 2014). However, clearly distinguishing if it is the transcribed RNA or the associated enhancer regions that regulate MYC could be complicated (Bassett et al., 2014). Indeed, a recent study shows that inhibition of the PVT1 transcription does not impact on MYC expression in a cancer cell line (Cho et al., 2018). This study instead showed that the PVT1 promoter modulates MYC expression as a competitor for enhancer activity, which may indicate that the transcribed RNA is a byproduct. Comparing the two mutations in this study might also clarify the role of PVT1 as lncRNA or a cis-regulator.

Figure 2 with 1 supplement see all
Download asset Open asset
Transcriptome analysis of Hap, STITCH+30kb and del(30-440).

(A) Tracks of RNA-seq from Hap, STITCH+30kb, and del(30-440) around the MYC locus. (B–D) MA-plots of RNA-seq to compare STITCH+30kb vs. Hap (B), del(30-440) vs. Hap (C) and STITCH+30kb vs. del(30-440) (D). Differentially expressed genes (adjusted p-values<0.05, log2 fold changes > 0.5) are marked by colors (orange for up-regulated genes and dark blue for down-regulated ones). (E and F) Enriched categories among HALLMARK50 (Liberzon et al., 2015) by GSEA (Liao et al., 2019) (left) and the enrichment plots against the categories MYC targets variant 1 and 2 (right) in STITCH+30kb (E) and del(30-440) (F).

Figure 2—source data 1

RNA-seq read counts and the results of the DESeq2 analyses.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig2-data1-v1.xlsx

We prepared libraries from three replicates (as for Hap the parental clone and two derived subclones; as for STITCH+30kb and del(30-440), three different clones isolated upon the Cre recombination, respectively) for each configuration. Consistently with the qPCR assay (Figure 1D), strong down-regulation of MYC was confirmed in both STITCH+30kb and del(30-440) (Figure 2A, Figure 2—figure supplement 1A). PVT1 expression was not altered in STITCH+30kb (Figure 2A, Figure 2—figure supplement 1A). We did not observe other detectable expression changes around the MYC locus in either STITCH+30kb or del(30-440) (Figure 2A). We computed the log2 fold changes of the transcriptome with the shrinking algorithm implemented in DESeq2 (Love et al., 2014; Figure 2B–D) and applied the results to Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) against the hallmark gene sets (HALLMARK50) in the Molecular Signatures Database (MSigDB) (Liberzon et al., 2015). Strikingly, the down-regulated genes in both STITCH+30kb and del(30-440) are highly enriched with known MYC target genes, showing that the down-regulation of MYC by both mutations is large enough to affect its target transcriptome. The other enriched categories are also well shared by the two mutations, highlighting the similarity in the transcriptomic change.

With threshold of log2 fold change <0.5 and p-adjusted <0.05, STITCH+30kb and del(30-440) had 218 and 68 down-regulated genes, and 494 and 137 up-regulated genes, respectively (Figure 2B,C). Among those, large fractions (36 and 92 genes, for down- and up-regulation, respectively) were common between the two alleles (Figure 2—figure supplement 1B,C). Importantly, the comparison between STITCH+30kb and del(30-440) called much less number (64) of differentially expressed genes (Figure 2D). Moreover, del(30-440) exhibited a rather milder effect on the transcriptome than STITCH+30kb (Figure 2B and C, Figure 2—figure supplement 1B,C). These data suggest that PVT1 has little impact on the transcriptome as trans-acting lncRNA if any.

It should be noted that the insulation showed a stronger effect than the deletion of the enhancer. STITCH+440kb did not show almost any effect on the MYC expression level (Figure 1C,D), indicating that the region beyond +440 kb does not contribute to the activation of MYC when the locus is intact. However, upon the deletion of the enhancer, contacts of MYC greatly extended beyond +440 kb (Figure 1C,E, and Figure 1—figure supplement 1B). Therefore it might be possible that MYC can be slightly activated by regions with some enhancer activity located beyond +440 kb that do not associate with MYC in the normal context, which may account for the milder outcome of del(30-440) than STITCH+30kb.

The results of the GSEA indicate the possible functional roles of MYC. The categories enriched in down-regulated genes include those in which MYC has been implicated by previous studies such as cell cycle progression (Bretones et al., 2015), unfolded protein response (Shajahan-Haq et al., 2014), TCA cycle (Anderson et al., 2018), mTORC1 signaling (Liu et al., 2017; Yue et al., 2017), and cholesterol synthesis (Hofmann et al., 2015; Figure 2E). Also, gene ontology (GO) enrichment analysis shows that the commonly down-regulated genes in STITCH+30kb and del(30-440) are highly enriched with genes encoding regulators of ribosome assembly, which are known target groups of MYC in various systems (Hofmann et al., 2015; Uslu et al., 2014; van Riggelen et al., 2010; Zeller et al., 2006), as well as those involved in cholesterol metabolism similarly as above (Figure 2—figure supplement 1B). Our results strengthen the link between MYC and these biological processes.

Titrating blocking activity of STITCH by serial mutations of the CTCF-binding sites

The divergent configuration of CTCF-binding sites establishes boundaries of contact domains in the genome, while those directed to only one side are also capable of partitioning the chromatin into two domains, namely as loop and exclusion domains (Guo et al., 2015; Sanborn et al., 2015). In fact, deletion and inversion of either of the two CTCF binding arrays, L or R, impaired, but still kept, the blocking activity of the TZ at the endogenous locus in the mouse ESCs (Tsujimura et al., 2018). However, it has been unclear how these differences in the CTCF configuration would impinge on gene activation by enhancers. In the present study, to understand how important the arrangement of the CTCF-binding sites is for STITCH to block the chromatin contact and the gene activation, we made deletion of each CTCF array, L (delL) and R (delR), inversion of R (invR), deletion of the middle five binding sites from L2 to R2 (del(L2-R2)), and deletion of the six sites but for R3 (del(L1-R2)) in STITCH+30kb. We also obtained deletion and inversion of the whole of STITCH (del(L1-R3) and inv(L1-R3), respectively) (Figure 3A, Figure 3—figure supplement 1).

Figure 3 with 4 supplements see all
Download asset Open asset
MYC expression and 4C-seq profiles in serially mutated STITCH alleles.

(A) Configurations of CTCF-binding sites of mutated STITCH alleles and a plot showing their MYC expression levels. Each dot represents replicate clones (see Materials and methods for details). Note that the data of Hap and STITCH+30kb are the same as Figure 1D. Bars indicate means of the replicates. (B) 4C-seq profiles from VP-MYC2 in the different alleles. The numbers indicate the ratios of the mapped reads to the indicated regions within the 3 Mb region, except for the 10 kb region from the viewpoint. Below the coordinate map, blue bars indicate bins (each 30 kb) for PCA in (C–H) and Figure 3—figure supplement 4. (C) PCA plot of all the clones using the normalized counts in all the bins of the whole locus. (D) Component loadings of PC1 in the PCA in (C) are plotted along the coordinate for each bin. (E, F) The PCA plot only with the non-blocking alleles, the original STITCH, and inv(L1-R3) using the bins of the whole locus (E), and the corresponding PC1 component loading plots (F). (G, H) The PCA plot with the same subset clones as (E) (left), and the corresponding PC1 component-loading plots (right) using the re-normalized counts in the bins of the left 900 kb region (G) or the right 600 kb region (H). Below the component-loading plot in (H), tracks of the super-enhancers and ChIP-STARR-seq plots reported in Barakat et al. (2018) are depicted along with the 30 kb bins of the right 600 kb region. The six bins with the lowest values of component loadings in (H) are depicted with pink. (I) A log-log plot of the MYC expression levels against the 4C contact frequencies of VP-MYC2 in the +(30-440)kb region (orange) and the PVT1 region (dark blue) for each clone. Note the difference between the two slopes. (J) A log-log plot of the 4C contact frequencies of VP-MYC2 in the PVT1 region against the +(30-440)kb region.

Figure 3—source code 1

Source Code File.

The R code for the PCA in Figure 3.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig3-code1-v1.r
Figure 3—source code 2

Source Code File_4CMYCcount.txt.

The file containing the 4C-seq read counts used in Figure 3-Source Code File.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig3-code2-v1.txt
Figure 3—source code 3

Source Code File_4CMYCcolor.txt.

The file used in Figure 3-Source Code File to specify the dot colors in the PCA plots.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig3-code3-v1.txt
Figure 3—source code 4

Source Code File_4CMYCshape.txt.

The file used in Figure 3-Source Code File to specify the dot shapes in the PCA plots.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig3-code4-v1.txt
Figure 3—source data 1

4C-seq read counts in the given intervals.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig3-data1-v1.xlsx

The MYC expression levels in delL and delR were slightly increased from the original STITCH allele (Figure 3A). invR also increased it but to a lesser extent (Figure 3A). del(L2-R2) and del(L1-R2) up-regulated the expression even more, but much less than the wild type Hap allele (Figure 3A). The MYC expression in del(L1-R3) was comparable to that of Hap, showing that the gene activation could be safely recovered upon removal of the CTCF-binding sites (Figure 3A). inv(L1-R3) exhibited the same degree of repression as STITCH+30kb, showing that STITCH blocks enhancer activation regardless of the orientation of the insertion as a whole (Figure 3A).

We next examined the 4C contact profiles of the MYC promoter in these mutation alleles (Figure 3B–J, Figure 3—figure supplements 2–4). The contact frequency with the +(30-440)kb enhancer region was changed depending on the configuration (Figure 3B, Figure 3—figure supplement 2A). The original STITCH and inv(L1-R3) most strongly reduced the contacts. invR showed slightly more of contacts there, but not as much as delL and delR. These results indicate firstly that the divergent configuration is the strongest way to block contacts, and secondly that the more CTCF binds there, the more strongly it blocks contacts (Figure 3B, Figure 3—figure supplement 2A). This observation is very consistent with the previous study about the endogenous TZ in the mouse ESCs (Tsujimura et al., 2018). del(L2-R2) and del(L1-R2) further recovered the contact frequency (Figure 3B). Thus, the gene expression level and the contact frequency are well correlated. The Spearman's rank correlation coefficients were 0.92 and 0.90 for VP-MYC1 (Figure 3—figure supplement 2B) and VP-MYC2 (Figure 3I), respectively. We noted that the expression level fits with a power-law model with the contact frequency of the +(30-440)kb region with a scaling exponent of 4.1–4.3 (Figure 3I, Figure 3—figure supplement 2B). It is particularly notable that even the del(L1-R2) efficiently blocks the gene activation only with the remaining one CTCF-binding site R3, but not much the contact.

To investigate into how the inserted STITCH impacts on chromatin conformation of the locus, we next performed 4C-seq from viewpoints flanking the insertion site (VP-STITCH-left and VP-STITCH-right) (Figure 3—figure supplement 3) in WT(Hap), STITCH+30kb, delR, and delL. The different compositions of the CTCF-binding sites in these mutants may affect the folding directionality of the flanking sites locally, as shown in previous studies (de Wit et al., 2015; Guo et al., 2015; Tsujimura et al., 2018). The flanking regions of the mouse TZ exhibit diverging directionality of chromatin folding (Tsujimura et al., 2018). This divergence is a typical hallmark feature of boundaries of contact domains (Dixon et al., 2012). We, therefore, calculated folding directionalities at each viewpoint (VP-MYC1, -MYC2, -STITCH-left, and -STITCH-right) as difference of read counts between the left and right intervals for given distances (1 Mb, 500 kb, or 100 kb) normalized by the sum of them (Figure 3—figure supplement 3B).

The Hap allele without the STITCH insertion exhibits overall rightward directionality from VP-MYC1/2 till VP-STITCH-left/right (Figure 3—figure supplement 3B). This tendency might be associated with the presence of the two CTCF-binding sites directed to the right side near the MYC promoter (Figure 1, Figure 3—figure supplement 3A). The insertion of STITCH introduced a skewed change of the directionality across the insertion site. The rightward directionality at VP-STITCH-right was even more enhanced, while those at VP-MYC1/2 and VP-STITCH-left were decreased to neutral (Figure 3—figure supplement 3B). In delR, the directionality at VP-STITCH-right became less prominent than the intact STITCH allele, while the directionality at both VP-MYC1/2 and VP-STITCH-left was again neutral (Figure 3—figure supplement 3B). By contrast, in delL, the rightward directionality was kept or slightly enhanced at both VP-STITCH-left and -right, while the directionality at VP-MYC1/2 was marginally reduced from the wild type allele (Figure 3—figure supplement 3B). These results suggest that the array L mainly orients the folding directionality at VP-MYC1/2 and VP-STITCH-left relatively towards the left side, while the array R enhances the rightward directionality at VP-STITCH-right. These relative transition patterns of the directionality across the insertion site are consistent with the case of the endogenous TZ (Tsujimura et al., 2018). However, it should be noted that the absolute divergence of folding directionality was not very evident around STITCH. Notably, the delL allele keeps the overall rightward directionality of chromatin folding across the region (Figure 3—figure supplement 3B). These results suggest that neither the diverging configuration of CTCF-binding sites nor the diverging directionality of chromatin folding is a prerequisite for enhancer blocking.

We note that VP-STITCH-left and VP-STITCH-right appear to have enhanced contacts with the left- and the right-side border of the large contact domains, respectively, which might represent the formation of loops by STITCH (Figure 3—figure supplement 3A). However, these contacts are not very striking compared to other recognizable contacting regions for both viewpoints (Figure 3—figure supplement 3A). Therefore, without deleting these regions, it is hard to specify loops, if any, that might be engaged in the STITCH functionality in this study. Also, more comprehensive analysis methods such as 5C (Chromatin Conformation Capture Carbon Copy) (Dostie et al., 2006) or Hi-C (Lieberman-Aiden et al., 2009) are required to fully describe the locus-wide conformational change induced by STITCH.

Preferential association with the enhancer over non-enhancer regions upon CTCF removal

To understand more quantitatively and unbiasedly how the various configurations of the CTCF-binding sites at STITCH reshape the contact pattern of MYC along the locus, we performed the principal component analysis (PCA) for the 4C contact frequencies of 30 kb bins within a given region (Figure 3B–H, Figure 3—figure supplement 4), as inspired by its application in the Hi-C analysis to find the compartment domains (Lieberman-Aiden et al., 2009). We first analyzed the frequencies within the whole MYC locus for all of the alleles above (Figure 3C,D). The PCA plot well segregated the non-blocking alleles (Hap and del(L1-R3)) from the other blocking ones, especially the original STITCH, inv(L1-R3) along the PC1 axis (Figure 3C). To understand which bins of the locus contribute to this segregation, we plotted the component loadings for each bin along the genomic coordinate (Figure 3D). Component loadings are calculated as the product of the eigenvector and the square root of the eigenvalue of the component. They correspond to the correlative coefficients of the original values of the bins and the component values. Therefore, component loadings indicate how much the values of each bin are reflected by the component. The component loadings of PC1 show that the segregation is mostly explained by lower and higher contact frequencies in the left side region, and higher and lower frequencies in the 570 kb region from the +30 kb site to the right side, of the non-blocking and the blocking alleles, respectively (Figure 3D).

We note that the different alleles are also arranged on the PCA plot according to the orientations of the CTCF binding motifs (leftward vs. rightward in Figure 3C). Both blocking effects of the mere presence of CTCF and directionality bias due to the orientations of the CTCF motifs seem to account for the segregation. To uncouple the two different effects, we performed PCA against subsets of the alleles. We first removed from the analysis the non-blocking alleles, Hap and del(L1-R3), to reduce the simple blocking effects and to enhance the directionality effect (Figure 3—figure supplement 4). Then, the alleles with leftward motifs were placed at the top, and the rest were at the bottom along PC1 on the plot (Figure 3—figure supplement 4). The component-loading plot indicates that the leftward alleles are more associated with the left side regions (Figure 3—figure supplement 4). These patterns are consistent with the above analysis showing that the array L reduces the rightward directionality of VP-MYC1/2 more than the array R (Figure 3—figure supplement 3B).

Next, to reduce the directionality effect, we used only the Hap/del(L1-R3), the original STITCH, and inv(L1-R3) clones for PCA. The PCA plot showed segregation between the non-blocking and blocking alleles along the PC1 axis (Figure 3E). The component-loading plot shows a clear split between the left 900 kb region and the right-side region at the STITCH insertion site (Figure 3F). The former region associates more with the blocking alleles, and the latter associates with the non-blocking alleles (Figure 3E,F).

To investigate if the left- or right-side regions contain sub-regions that specifically change contact patterns with MYC depending on the presence of STITCH, we then performed PCA for each of the left 900 kb region and the right 600 kb region with the subset clones (Figure 3G,H). The PCA plot for the left side did not show apparent segregation according to the CTCF composition (Figure 3G). By contrast, PCA for the right 600 kb region showed segregation between the blocking and non-blocking alleles (Figure 3H). These results indicate that the right 600 kb region contains bins that characteristically alter contact tendency with MYC depending on the presence of STITCH, while the left 900 kb region does not.

The pattern of the PC1 component loadings for the right side PCA was notable (Figure 3H). The association with the PVT1 region, especially with the super-enhancer region (Barakat et al., 2018), accounts for the lower PC1 values of the Hap/del(L1-R3) clones, while that with the other remaining non-active regions accounts for STITCH/inv(L1-R3) (Figure 3H). These results suggest that MYC has preferential contacts with the super-enhancer/PVT1 region more than with the other non-active regions in the absence of the CTCF insulation. We found that the power-law scaling of the MYC expression with the contact frequency with the PVT1 region has a scaling exponent of 3.6–3.7, which is slightly less than with the +(30-440)kb region (Figure 3I, Figure 3—figure supplement 2B). Consistently, the contact with PVT1 scales with that with the +(30-440)kb region, with an exponent factor 1.14–1.15, which is slightly higher than the linear correlation (Figure 3J, Figure 3—figure supplement 2C). Thus, titration of STITCH insulation revealed that the contact of MYC with the super-enhancer/PVT1 region is enhanced more than the other non-active regions when the insulation is absent. In other words, the presence of CTCF insulation effectively impairs the gene-enhancer contact more than the contacts with neutral regions. A similar observation was also reported by a previous study (Hou et al., 2008). We think this kind of selective disruption of the gene-enhancer interaction may, at least in part, account for the discrepancy between the relatively small changes of the overall contact frequency and the drastic reduction of gene expression by the CTCF insulators here and in other genomic contexts.

Epigenetic states of MYC well correlate with the gene activation by the enhancer

We next investigated how the STITCH insulation of the enhancer impinges on the epigenetic modifications of histones around MYC (Figure 4). Active transcription is associated with H3K4me3 at gene promoters, while repressed genes are often marked by H3K27me3. In the wild type allele, MYC is exclusively marked by H3K4me3, but not by H3K27me3. Upon the STITCH insulation, the H3K4me3 deposition remained, but was markedly decreased. Instead, H3K27me3 was enriched. By contrast, the neighboring PVT1 gene was strongly marked by H3K4me3 at the promoter in both conditions (Figure 4A, Figure 4—figure supplement 1A,B). Some typically active (ACTB, NANOG, DPPA4) and repressed (T, HOXD13) genes were constantly marked by either H3K4me3 or H3K27me3, respectively (Figure 4A, Figure 4—figure supplement 1C). Also, the H3K27ac mark around the super-enhancer region was similarly observed in both alleles (Figure 4B, Figure 4—figure supplement 1A,B). Among the peaks that were called in at least two out of the total four experiments (two from Hap and the other two from STITCH+30kb), the peaks at MYC were ranked as one of the top peaks exhibiting the largest fold change for both H3K4me3 (Figure 4C) and H3K27me3 (Figure 4D), while the H3K27ac peaks around PVT1 did not change much (Figure 4E). These results show that the epigenetic change only occurred at MYC upon isolation from the enhancer by STITCH. We performed nChIP-qPCR to quantify the H3K4me4 and H3K27me3 levels at MYC in the alleles with the STITCH mutations (Figure 4F,G). We normalized the enrichment by that at ACTB and T (Figure 4A) to better compare different experiments for H3K4me3 and H3K27me3, respectively (Figure 4F,G). We found that MYC in the mutant alleles were epigenetically intermediate between the active and repressive states (Figure 4F,G). These results show that the histone marks around MYC vary depending on the association levels with the enhancer or the gene expression level.

Figure 4 with 1 supplement see all
Download asset Open asset
Epigenetic profile around MYC with and without STITCH.

(A) nChIP-seq for H3K4me3 (green) and H3K27me3 (red) in the wild type (Hap) and STITCH+30kb clones. The magnified view around MYC is shown below, together with the typical repressive (T) and the active (ACTB) regions. (B) nChIP-seq for H3K27ac in Hap and STITCH+30kb. (C–E) The peaks of H3K4me3 (C), H3K27me3 (D), and H3K27ac (E) are ordered according to the normalized log2 fold changes in STITCH+30kb. The H3K4me3 and H3K27me3 peaks at MYC are depicted with red, and peaks at other representative genes are depicted with green, in C and D, respectively. Similarly, H3K27ac peaks within the PVT1 genic region are depicted with red in E. (F and G) nChIP-qPCR for H3K4me3 (F) and H3K27me3 (G) in Hap, STITCH+30kb and the indicated mutant alleles of STITCH. The enrichment at MYC was normalized with those at ACTB (F) and T (G). We also quantified the relative enrichment at DPPA4 and T for H3K4me3 (F), and HOXD13 and ACTB for H3K27me3 (G), as positive and negative controls, respectively. The dots represent data from replicate experiments. The bars and the error bars indicate their means and the standard deviations (SD), respectively.

Figure 4—source data 1

nChIP-seq read counts in the peaks for H3K4me3, H3K27me3, and H3K27ac.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig4-data1-v1.xlsx

Induction of a heterochromatic state by tetR-KRAB impairs the STITCH insulation

The KRAB domain can induce heterochromatin formation around the tetO when linked to tetR (tetR-KRAB) and recruited there (Deuschle et al., 1995; Groner et al., 2010; Sripathy et al., 2006). If this leads to impairment of CTCF bindings as implicated in a previous study (Jiang et al., 2017), it would be possible to control the insulation ability of STITCH by DOX (Figure 5A). To test this, we integrated a transgene consisting of tetR-KRAB followed by DNA encoding the 2A peptide and the puromycin resistant gene (2A-PUROr) with piggyBac transposition into the genome of a STITCH+30kb clone, and established several cell lines that stably express it (Figure 5B). The expression levels of the transgene varied much among them (Figure 5—figure supplement 1A). Nonetheless, in all the cell lines tested, MYC expression was repressed in the presence of DOX but became activated after the removal of DOX (Figure 5C). Titration of the DOX concentration showed that 1 ng/ml is enough to achieve STITCH insulation in the tested clones with different expression levels of the transgene (Figure 5D, Figure 5—figure supplement 1D).

Figure 5 with 1 supplement see all
Download asset Open asset
Drug-inducible control of STITCH insulation with tetR-KRAB.

(A) DOX dependent binding to and dissociation from STITCH of tetR-KRAB. (B) The piggyBac transposon with the tetR-KRAB transgene followed by a sequence encoding 2A peptide and puromycin resistant gene. (C) The relative expression levels of MYC normalized to ACTB in five independent clones of STITCH/KRAB with and without DOX were compared to the expression levels of the ancestral Hap and STITCH+30kb clones from which the STITCH/KRAB clones were derived. (D) The MYC expression level in the clone 1 of STITCH/KRAB with different concentrations of DOX. The dots represent data from replicate experiments, and the bars indicate the means. (E, F) nChIP-seq tracks for H3K9me3 (E) and CTCF (F) of the clone one with and without DOX. The reads were mapped to a synthetic genomic DNA sequence around the MYC locus carrying the STITCH insert. (G) The 4C-seq tracks with and without DOX from VP-MYC2. The numbers indicate the ratios of sequence reads mapped to given intervals within the locally haploid 3 Mb region except for the 10 kb region from the viewpoint fragment.

Figure 5—source data 1

4C-seq read counts in the given intervals, and CTCF nChIP-seq read counts in the peaks.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig5-data1-v1.xlsx

We performed nChIP-seq for H3K9me3, a mark representing the heterochromatin state, and for CTCF. When DOX was present, no H3K9me3 peak appeared around the inserted STITCH (Figure 5E, Figure 5—figure supplement 1E–G); instead, CTCF was strongly bound there (Figure 5F, Figure 5—figure supplement 1E–I). Accordingly, STITCH kept blocking the contacts of MYC towards PVT1 (Figure 5G, Figure 5—figure supplement 1M). In the absence of DOX, however, H3K9me3 became highly enriched around STITCH (Figure 5E, Figure 5—figure supplement 1E–G). Concomitantly, the CTCF binding was strongly reduced, and the contact of MYC well extended to the enhancer region (Figure 5F,G, Figure 5—figure supplement 1E–I,M). We calculated the normalized fold changes of the read counts of the CTCF nChIP-seq mapped to each peak throughout the genome. Then, the arrays L and R of STITCH were the most significantly altered peaks by the removal of DOX (Figure 5—figure supplement 1F).

By contrast, induction of tetR linked to 3xFLAG with HA tag followed by 2A-PUROr neither affected CTCF binding at STITCH nor activated MYC in the STITCH+30kb clone (Figure 5—figure supplement 1B,J), showing that the KRAB domain is required to expel CTCF binding. We also confirmed that the STITCH before the Cre/loxP recombination harboring the PUROr cassette, which should be bound by some transcription factors around the promoter for the expression, recruits CTCF and blocks MYC activation (Figure 5—figure supplement 1A,K,L), further arguing that binding of transcription factors does not impair CTCF binding. Also, integration of tetR-KRAB into a del(30-440) clone, which keeps two tetO sites at the +30 kb position, did not up-regulate MYC in the absence of DOX (Figure 5—figure supplement 1C). These results show that the re-association with the enhancer upon KRAB-dependent displacement of CTCF led to the MYC activation by tetR-KRAB in the absence of DOX. Thus, the STITCH/KRAB system functions as a drug-inducible topological insulator to control gene activation by enhancers.

We next followed temporal changes of the system upon the addition and removal of DOX (Figure 6). The nChIP-qPCR for H3K9me3, the 4C-seq assays, and gene expression assays show that 16–24 hr, but not 8 hr, are sufficient to almost completely switch the STITCH insulation and MYC expression upon both removal and addition of DOX (Figure 6A–E). We tested how the switching of MYC expression would affect the cell proliferation and found that the addition of DOX (i.e., repression of MYC) for five days resulted in about 40% reduction of proliferated cells (Figure 5—figure supplement 1N).

Figure 6
Download asset Open asset
Temporal changes of STITCH insulation upon removal and addition of DOX.

(A, B) The 4C-seq profiles in 0, 4, 8, 16, and 24 hr after removal (A) and addition (B) of DOX. The numbers indicate the ratios of sequence reads mapped to given intervals within the locally haploid 3 Mb region except for the 10 kb region from the viewpoint fragment. (C–E) Temporal changes of nChIP-qPCR for H3K9me3 at STITCH (C), 4C contact frequency with +(30-440)kb region from VP-MYC2 (D), the relative MYC expression level normalized to ACTB (E). (F, G) Temporal changes of relative enrichment of H3K4me3 at MYC normalized with that at ACTB (F), and relative enrichment of H3K27me3 at MYC normalized with that at T (G), up to 48 hr after removal and addition of DOX. We did not perform replicate experiments in (A–E). The nChIP-qPCR for H3K4me3 and H3K27me3 were performed for three replicate samples. The means and SDs are represented in the plots (F, G).

Figure 6—source data 1

4C-seq read counts in the given intervals, and the results of nChIP-qPCR for H3K4me3 and H3K27me3.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig6-data1-v1.xlsx

The epigenetic state of MYC follows and reflects the gene expression level

The H3K4me3 and H3K27me3 histone marks correlate well to the gene expression level (Figure 4, Figure 4—figure supplement 1). The rapid control of STITCH insulation with KRAB offers us an opportunity to investigate if the epigenetic changes precede the gene expression changes or not. Therefore, we also profiled the H3K4me3 and H3K27me3 levels around MYC at different time points up to 48 hr after the inductions. Interestingly, while the H3K4me3 mark returned to the levels expected from the gene expression levels within 24 hr after both removal and addition of DOX (Figure 6F), the H3K27me3 did not (Figure 6G). This result suggests that the change of the repressive histone mark follows, but does not precede, the gene expression change.

To test the hypothesis and confirm the reproducibility, we again sampled cells at time points of 24 and 72 hr after the addition/removal of DOX as well as cells that were kept either with or without DOX for more than one passage as the controls (Figure 7A–F). First, we confirmed that the MYC expression was up- and down-regulated within one day after removal and addition of DOX to the levels of the controls, respectively (Figure 7A,B). Then we performed nChIP-qPCR for both histone marks. Consistently to above, the deposition of H3K27me3 was significantly higher and lower in 24 hr than 72 hr and the controls after removal and addition of DOX, respectively (Figure 7C,D). By contrast, we did not see such significant differences for H3K4me3, suggesting that the active mark is more rapidly turned over than the repressive mark (Figure 7E,F).

Figure 7
Download asset Open asset
Delayed turnover of H3K27me3 enrichment after the gene expression change.

(A–F) Relative MYC expression levels normalized to ACTB (A and B), relative H3K27me3 enrichment at MYC normalized to the enrichment at T (C and D) and relative H3K4me3 level at MYC normalized to that at ACTB (E and F) were measured at 24 hr (1 day) and 72 hr (3 days) after removal (A, C, E) or addition (B, D, F) of DOX in the STITCH/KRAB. The controls are the cells kept without (A, C, E) or with (B, D, F) DOX without switching for a few passages. The dots represent data from replicate experiments, the bars indicate their means, and the error bars indicate the SDs. *, *** and n.s. indicate p<0.05, p<0.001 and p>0.05, respectively, by one-way ANOVA. The p-values with Tukey’s multiple-comparison post hoc test are indicated. (G, H) Enrichment of H3K27me3 (G) and H3K4me3 (H) at MYC after two days treatment with EPZ or DMSO in STITCH+30kb. The dots represent replicates, the bars indicate their means, and the error bars indicate the SDs. * and n.s. indicate p<0.05 and>0.05, respectively, by two-sided Welch's two-sample t-test. (I) Relative MYC expression levels in the Hap, STITCH+30kb, and the mutants of STITCH after three-days treatment of EPZ or DMSO. The dots represent replicates, and the bars indicate their means. (J) Temporal changes of relative MYC expression levels after DOX removal in the STITCH/KRAB. Before DOX was removed, cells were exposed to EPZ or DMSO for two days. Means and SDs of three replicate experiments were plotted. (I, J) n.s. indicates p>0.05, by one-sided Welch's two-sample t-test, in which the alternative hypothesis was that the mean of EPZ was greater than DMSO.

Figure 7—source data 1

MYC expression levels upon removal of DOX with DMSO or EPZ.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig7-data1-v1.xlsx

These results suggest that the H3K27me3 mark per se only reflects, but does not determine the gene expression level. To test this, we treated the cells with EPZ-6438 (EPZ), an inhibitor of Enhancer of zeste homolog 2 (EZH2), an enzymatic subunit of Polycomb Repressive Complex 2 (PRC2), which catalyzes methylation of H3K27 (Knutson et al., 2013). The addition of the inhibitor at 200 nM for two days was enough to mostly diminish the H3K27me3 mark at MYC (Figure 7G). This reduction of H3K27me3 did not result in significantly higher enrichment of the active H3K4me3 mark (Figure 7H). We compared the MYC expression levels in Hap, STITCH+30kb, and the mutant alleles of STITCH treated with EPZ or DMSO for three days (Figure 7I). The difference between the two treatments was not significant in any of the alleles. We next treated the STITCH/KRAB cells with EPZ or DMSO for two days, then removed DOX, and compared the MYC expression at different time points up to 24 hr after removal of DOX. The expression profiles showed no significant difference between the two, suggesting that the H3K27me3 mark does not affect the gene activation by the enhancer (Figure 7J).

Blocking NEUROG2 activation in differentiating neural progenitor cells with STITCH

We next tested the applicability of the STITCH/KRAB system to a different locus in a different cell-type. NEUROG2 is a proneural gene expressed in neural progenitor cells (NPCs) (Bertrand et al., 2002). In the mouse embryonic brain, a stretch of the tissue-specific peaks of H3K27ac is present over the neighboring gene Alpk1 (ENCODE Project Consortium, 2012), suggesting that these are the neural enhancers for Neurog2 (Figure 8—figure supplement 1). NPCs can be efficiently derived from the human pluripotent stem cells by the dual SMAD inhibition (Chambers et al., 2009). A reported data shows that the differentiated NPCs with this method also exhibit prominent H3K27ac marks over ALPK1 (Xie et al., 2013; Figure 8A), suggesting that these enhancers activate NEUROG2 in vitro.

Figure 8 with 2 supplements see all
Download asset Open asset
Testing STITCH/KRAB at the NEUROG2 locus in the neural progenitor cells.

(A) STITCH was inserted into the 65 kb downstream of NEUROG2 near ALPK1. There seem active enhancers over the ALPK1 genic region, as indicated by the previously reported H3K27ac profile in the NPCs (Xie et al., 2013). (B) The schematic illustration of the NEUROG2/KRAB cells (left) and the differentiation experiments (right). (C) The relative expression levels of NEUROG2, PAX6, and NANOG normalized by the ACTB expression levels during the differentiation process. Day 0 indicates iPSCs with or without DOX collected just before the start of the neural induction. The numbers of replicates were three for day 0 and four for days 2, 4, and 6. The means and the SDs of the replicate experiments are represented. The indicated p-values are obtained by one-sided Welch's two-sample t-test, where the alternative hypothesis was that the mean of DOX minus was greater than DOX plus. * and n.s. indicate p<0.05 and>0.05, respectively. (D and E) nChIP-qPCR for H3K9me3 (D) and CTCF (E) enrichment at STITCH (the L2 motif region) in iPSCs and NPCs (day6) with and without DOX. We also quantified the enrichment at ZNF544 (D) and RAE1 (E) regions as the positive controls for H3K9me3 and CTCF, respectively. We used the region around T as the negative control for both assays. (F) 4C-seq from VP-NEUROG2 in iPSCs and NPCs with and without DOX. The numbers indicate the ratios of the mapped reads in the given intervals.

Figure 8—source code 1

Source Code File_4CNGN2color.txt.

The file used in Figure 3-Source Code File to specify the dot colors in the PCA plots.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig8-code1-v1.txt
Figure 8—source code 2

Source Code File_4CNGN2count.txt.

The file containing the 4C-seq read counts used in Figure 3-Source Code File.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig8-code2-v1.txt
Figure 8—source code 3

Source Code File_4CNGN2shape.txt.

The file used in Figure 3-Source Code File to specify the dot shapes in the PCA plots.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig8-code3-v1.txt
Figure 8—source data 1

Relative gene expression levels of NEUROG2, PAX6, NANOG, and tetR in differentiating NPCs, and 4C-seq read counts in the given intervals.

https://cdn.elifesciences.org/articles/47980/elife-47980-fig8-data1-v1.xlsx

We inserted STITCH into the 65 kb downstream of NEUROG2 near ALPK1 in the iPS cells (the Hap clone), removed the PUROr cassette with Cre, and then integrated the tetR-KRAB-2A-PUROr with the piggyBac transposon. We term the resultant cells as NEUROG2/KRAB (Figure 8A,B). Here, STITCH was inserted only into one allele with the other one remaining intact. Also, after the piggyBac transposition, we did not clone single colonies, but just expanded the survived cells as a bulk for several passages in the presence of puromycin. Of note, the MYC expression levels in this cell population did not change by the absence and presence of DOX (Figure 8—figure supplement 2A), confirming again that tetR-KRAB controls MYC expression only through the STITCH+30kb insertion (Figure 5).

We split the NEUROG2/KRAB cells derived from a single dish equivalently to different dishes, and then either did or did not add DOX upon the start of the differentiation into NPCs. The neural differentiation was achieved by the dual SMAD inhibition (Chambers et al., 2009) with SB-431542, an inhibitor for the SMAD2/3 pathway (Inman et al., 2002), and LDN-193189, an inhibitor for the SMAD1/5/8 pathway (Cuny et al., 2008; Figure 8B). We then compared the expression levels of NEUROG2, as well as PAX6 (NPCs marker) (Chambers et al., 2009) and NANOG (iPSCs marker) on days 2, 4 and 6 (Figure 8C). The induction diminished the NANOG expression already on day2 (Figure 8C). PAX6 was strongly activated from day 4 (Figure 8C), showing that the cells were efficiently differentiated. NEUROG2 was also activated from day 4 (Figure 8C). We tested if DOX treatment would decrease NEUROG2 expression and found that NEUROG2 was significantly less expressed in the cells with DOX than those without on day 4 (Figure 8C). On day 6, the tendency that the DOX treated cells express less NEUROG2 was kept, though the difference was milder than day 4 and not statistically significant (Figure 8C). We realized that the expression level of tetR-KRAB was progressively decreased during the differentiation (Figure 8—figure supplement 2B), suggesting a part of the cells in culture might have experienced silencing of the transgene possibly due to the complete alteration of the epigenomic state. This silencing effect might be a reason why the difference between DOX plus and minus became smaller on day 6 (Figure 8C).

We compared the heterochromatin formation and CTCF binding at STITCH in the iPSCs and NPCs on day 6 between with and without DOX. In both cell types, tetR-KRAB induced H3K9me3 and expelled CTCF binding in the absence, but not in the presence of DOX (Figure 8D,E).

We next performed 4C-seq from a viewpoint at the NEUROG2 promoter (VP-NEUROG2). Though the intact allele seems to mask the difference between the conditions a lot, contacts of NEUROG2 with the ALPK1 region beyond the STITCH insertion were constantly reduced by the addition of DOX in both iPSCs and NPCs (Figure 8F). PCA against the 4C-seq data segregated NPCs and iPSCs along PC1, and DOX plus and minus along PC2 (Figure 8—figure supplement 2C). Notably, the PC2 component loading plot unbiasedly exhibited the changing point exactly at the STITCH insertion site: the segregation between DOX plus and minus well correlates with contacts with the right- and the left-side regions from the insertion site, respectively (Figure 8—figure supplement 2C). We further performed 4C-seq using a viewpoint designed at the right edge of the inserted STITCH cassette (VP-R3) in the NPCs (Figure 8—figure supplement 2D–G). The addition of DOX strongly extended the contacts to further distances, as indicated by the ratio of reads between 100 and 200 kb distance region against those immediately within 100 kb region (Figure 8—figure supplement 2E,F). This change of chromatin conformation should reflect the extrusion-mediated contacts of the CTCF-binding sites at STITCH in the presence, but not the absence, of DOX (Haarhuis et al., 2017). Of note, the leftward extension of the contact indicates that the 4C-seq captures the effect of the leftward CTCF array L1-L4 in the very close vicinity of VP-R3 (Figure 8—figure supplement 2D). Therefore, it was not surprising that the directionality of the chromatin folding of VP-R3 was not drastically biased towards the right side (Figure 8—figure supplement 2E,G). Overall, these results are consistent to the above observation that STITCH blocks the chromatin contacts of NEUROG2. Thus, the STITCH/KRAB system can be used in different loci in different cell types, strengthening its generality and robustness as a tool.

Discussion

STITCH blocks the interaction of genes and enhancers when inserted in between as an insulator element (Figure 9A,B). Further combining this with the DOX control of tetR-KRAB achieved drug-inducible switching of the insulation (Figure 9C). Thus, the system adds a new layer to the toolkits for manipulating gene expression. Here, we first discuss the mechanism of the STITCH system and then the applicability of the system as a tool.

Figure 9
Download asset Open asset
Summary of the STITCH system and models for the CTCF insulation.

(A, B) Schematic illustration of how STITCH blocks the gene-enhancer interaction. STITCH insertion efficiently blocks the interaction, while it also alters the contact tendency of the locus though less prominently. (C) Upon the tetR-KRAB induction, the contact frequency becomes normal, and the gene-enhancer interaction is restored. (D–E) Models of how CTCF efficiently impairs the gene-enhancer interaction. There might be a mechanism that a slight increase/decrease of contact frequency leads to a drastic increase/decrease of the gene-enhancer interaction (D). Also. CTCF might actively disentangle the gene-enhancer interaction through loop extrusion (E).

Mechanism of the STITCH insulation and its control by heterochromatin induction

Though CTCF binding to L1 and L4 was not confirmed by the nChIP, the other five binding sites at STITCH were directly bound by CTCF (Figure 1—figure supplement 1C and Figure 5—figure supplement 1G). The delL and del(L1-R2) alleles only keep the direct binding sites of CTCF, and still show substantial insulation activity (Figure 3). Further, the insulation activity and the folding property is dependent on the orientation of the binding motifs (Figure 3), as in the endogenous TZ region (Tsujimura et al., 2018). Therefore, it should be safe to attribute the STITCH insulation primarily to the bindings of CTCF. Blocking of enhancer activity by heterologously inserted CTCF-binding sites is consistent with previous studies (Bell et al., 1999; Hou et al., 2008; Liu et al., 2015).

CTCF establishes a boundary of contact domains through the function of blocking the extrusion of cohesin (Fudenberg et al., 2016; Guo et al., 2015; Sanborn et al., 2015). Numerous studies have shown that the domain boundaries with CTCF-binding sites insulate enhancer activation (Dowen et al., 2014; Lupiáñez et al., 2015; Narendra et al., 2015; Symmons et al., 2014; Tsujimura et al., 2015; Tsujimura et al., 2018). From these observations, it is vaguely accepted that CTCF limits the action range of enhancers through the formation of contact domains. However, whether the contact domains are the entity that regulates the gene-enhancer interaction in the context of CTCF insulation is elusive, because their direct causality was not shown so far. Recent comprehensive imaging of chromatin structure showed that the domain-like structures are frequently present across the boundary positions (Bintu et al., 2018), showing that the contact domains might be a mere averaged projection of highly variable chromatin structures. Our analysis based on the folding directionality, particularly of the delL allele, may indicate that CTCF can prevent the gene-enhancer interaction without demarcating contact domains (Figure 3—figure supplement 3). However, there are various ways to define contact domains (Zufferey et al., 2018). Moreover, applying 5C or Hi-C might be more appropriate to describe formation of contact domains than the present 4C-based analyses. Therefore, our study cannot conclude about the causality of the contact domains for the enhancer blocking activity of STITCH.

We instead compared contact profiles of MYC between different alleles deeply. First, we showed that the small changes (at most by half) of the contact frequency with the enhancer region lead to the drastic reduction of the MYC expression level by up to 20 folds (Figures 1 and 3, Figure 1—figure supplement 1, Figure 3—figure supplement 2). This fact may indicate that the disruption of gene activation should not only be attributed to the simple reduction of the contacts beyond the CTCF-binding sites.

Next, we compared the contact distribution only within the region beyond the STITCH insertion site (Figure 3H–J). Then we found that the contact of MYC with the super-enhancer/PVT1 region was enhanced upon the stepwise loss of CTCF-binding sites of STITCH more than with the other non-active regions (Figure 3H–J). We think this result well explains, at least in part, how the gradual changes of contact frequency are translated into the skewed expression changes (Figure 9A–C).

The genome tends to be compartmentalized into two parts, active and repressive domains (Lieberman-Aiden et al., 2009). The depletion of CTCF or cohesin was shown to enhance the compartmentalization (Nora et al., 2017; Rao et al., 2017; Schwarzer et al., 2017). Along with this line, our observation can be interpreted that the preferential association of MYC with the super-enhancer obeys the same compartmentalization principle and that the CTCF binding interrupts this process. Then how does CTCF do so? Possibly there might be a mechanism that enhances aggregation of the active regions upon an increase of contact frequencies (Figure 9D). For example, the recently proposed phase separation model may explain it well (Hnisz et al., 2017). The increase of the overall contact frequency in the absence of CTCF-binding sites in between may boost the compartmentalization. Whether the loop extrusion process by cohesin would further help the association of MYC with the enhancer or not is unclear. A previous report has shown that the compartmentalization among super-enhancers is established even between different chromosomes upon depletion of cohesin, suggesting that the loop extrusion is not required for this process (Rao et al., 2017). In addition, or alternatively, CTCF per se, probably through anchoring the stabilized or dynamically extruding cohesin loops, might actively disrupt the compartmentalized association of MYC with the enhancer, when present in between (Figure 9E). Distinguishing the boost effect of the compartmentalization by the increase of contact frequency in the absence of CTCF (Figure 9D) and the interference effect of the compartmentalization by the loop extrusion in the presence of CTCF (Figure 9E) would be challenging.

The induction of tetR-KRAB impaired binding of CTCF at STITCH and restored the contacts of MYC with the enhancer over STITCH. This could be due to the formation of the heterochromatic states that were represented by the H3K9me3 deposition. The heterochromatic regions form dense nucleosomes, which may exclude binding of transcription factors (Machida et al., 2018). A previous study indirectly suggested that KRAB induction reduces the binding of CTCF (Jiang et al., 2017). Notably, our study shows that the formation of heterochromatin does not prevent the association between genes and enhancers. At the same MYC locus, recruitment of the KRAB domain to the PVT1 promoter did not block the enhancer activation (Cho et al., 2018). It should be emphasized that the KRAB protein is generally considered as a repressor protein, and has also been widely used to repress gene expression artificially. Our study clearly shows that in certain contexts, KRAB might be able to activate gene expression. The prevalence of this kind of regulation in the endogenous genomes needs to be studied in the future.

STITCH as a novel tool for manipulating gene expression

In this study, we mainly applied the STITCH/KRAB to dissect gene regulation by long-range enhancers. The system has a unique advantage that it can target specifically only one locus without affecting much of the cellular and epigenetic states even around the enhancer region (Figure 4, Figure 4—figure supplement 1). This is in contrast to many other studies that depleted genes and proteins or induced cellular differentiation and signaling cascades. We think coupling the STITCH/KRAB system with live-imaging techniques and others should further contribute to understanding gene regulation by enhancers.

We also anticipate that STITCH can be a useful tool to disrupt gene function in a tissue-specific manner. Currently, this is predominantly achieved by the Cre-loxP system, which inevitably needs a suitable driver for Cre expression. However, STITCH disruption needs just one insertion between a gene and an enhancer. Our work exemplified that even the enhancers stretching over a vast region could be blocked. Controlling the insulation by KRAB can repeatedly switch on and off gene expression as desired and thus adds another degree of control.

The functionality of STITCH primarily relies on the binding of CTCF, as discussed above. Therefore, its generality should mostly depend on how robustly CTCF binds to STITCH and blocks the gene-enhancer interaction, and on how robustly the KRAB induction controls the binding of CTCF. The motif sequences recognized by CTCF at STITCH are derived from the TZ, to which CTCF consistently binds in various cell types in mice (Tsujimura et al., 2018). It was also shown that the TZ blocks chromatin contacts and gene-enhancer interactions in different contexts upon several balanced inversions in the mouse embryos (Tsujimura et al., 2015). We show in this study that the motif sequences robustly recruit CTCF in the same way as the TZ does even as a reconstituted DNA cassette in the genome of a different species, human, in both iPSCs and NPCs. We also confirmed that the tetR-KRAB induction expelled CTCF binding from STITCH regardless of the insertion sites and the cell types examined. Moreover, STITCH blocked chromatin contacts and the enhancer activities in these different contexts, as the endogenous TZ does in the mouse genome. These facts well argue that the STITCH system should be applicable robustly to various genomic and cellular contexts. As discussed above, however, there is still uncertainty in how CTCF interrupts the gene-enhancer interaction. Therefore, it cannot be excluded that STITCH might encounter cases where it does not affect the gene-enhancer interaction as expected, which might instead lead to uncovering yet unknown modes of genome regulation by CTCF.

The MYC regulation

MYC is one of the four factors of the original cocktail to induce pluripotent stem cells (Takahashi and Yamanaka, 2006; Takahashi et al., 2007). In the STITCH/KRAB cells, the decrease of MYC expression led to a decreased proliferation rate (Figure 5—figure supplement 1N). It is well known that MYC accelerates cell proliferation in various systems, including cancers (Bretones et al., 2015). Further, our transcriptome analysis revealed that down-regulation of MYC leads to a decrease of genes involved in several cellular and metabolic processes that are also known to be targets of MYC in various cell types (Figure 2). These results suggest that in the iPSCs, MYC regulates cellular metabolism and proliferation through up-regulation of a specific set of target genes that are also shared by different types of cells, including cancer cells. Further digging into the function of MYC in our system should be fruitful in this sense.

The H3K27me3 mark reflects the gene expression

The STITCH insulation not only down-regulated the gene expression but also affected the epigenetic states of MYC (Figure 4). We further investigated the temporal change and showed that the deposition of H3K27me3 only follows and reflects, but does not precede and affect gene expression changes. Perhaps this might seem contradictory to the prevailing notion of the histone mark as a repressor. However, the delayed change of the histone modification after the transcriptional change is consistent with previous reports showing the same relationship upon the global induction of cellular stimuli (Hosogane et al., 2013; Kashyap et al., 2011), and with the mechanical property of the repressive state as an epigenetic memory (Reinberg and Vales, 2018). Also, accumulative evidence has shown that PRC2 has almost no effect on gene expression in a particular context (Riising et al., 2014). Yet, mutations in genes encoding PRC2 components have indicated that PRC2 has diverse and critical roles in organisms (Schuettengruber and Cavalli, 2009). Also, it was shown that PRC2 maintains gene silencing during the differentiation of mouse ES cells (Riising et al., 2014). To explain these observations, it has been proposed that the deposition of H3K27me3 raises the threshold for gene activation (Comet et al., 2016). However, the studies involving gene activation so far were carried out under the global induction of cellular stimuli. Therefore, it has not been clear if the H3K27me3 marks regulate gene activation locally as a resistance in cis or rather globally through effects on the cellular and epigenomic states. Our experiment showed that the presence of H3K27me3 makes no significant difference in MYC activation upon the local induction by the enhancer (Figure 7J), and thus challenged the above hypothesis. The role of this repressive histone mark needs to be further studied in future.

Materials and methods

Key resources table
Reagent type
(species) or
resource		DesignationSource or referenceIdentifiersAdditional
information
Cell line (Homo-sapiens)253G1 induced pluripotent stem cellsRIKEN BRCHPS0002: 253G1, RRID:CVCL_B518
Cell line (Homo-sapiens)HapThis paper3 Mb deletion of an allele around MYC, in 253G1 cells
Cell line (Homo-sapiens)STITCH-30kbThis paperSTITCH insertion into 30 kb upstream of MYC, in Hap cells
Cell line (Homo-sapiens)STITCH+30kbThis paperSTITCH insertion into 30 kb downstream of MYC, in Hap cells
Cell line (Homo-sapiens)STITCH+440kbThis paperSTITCH insertion into 440 kb downstream of MYC, in Hap cells
Cell line (Homo-sapiens)STITCH+1760kbThis paperSTITCH insertion into 1760 kb downstream of MYC, in Hap cells
Cell line (Homo-sapiens)STITCH+1790kbThis paperSTITCH insertion into 1790 kb downstream of MYC, in Hap cells
Cell line (Homo-sapiens)del(30-440)This paperDeletion of +(30.440)kb region in Hap cells
Cell line (Homo-sapiens)delLThis paperDeletion of the CTCF binding sites L1-L4 of STITCH in STITCH+30kb
Cell line (Homo-sapiens)delRThis paperDeletion of the CTCF binding sites R1-R3 of STITCH in STITCH+30kb
Cell line (Homo-sapiens)invRThis paperInversion of the CTCF binding sites R1-R3 of STITCH in STITCH+30kb
Cell line (Homo-sapiens)inv(L1-R3)This paperInversion of the whole STITCH in STITCH+30kb
Cell line (Homo-sapiens)del(L1-R3)This paperDeletion of the whole STITCH in STITCH+30kb
Cell line (Homo-sapiens)del(L2-R2)This paperDeletion of the CTCF binding sites L2-R2 of STITCH in STITCH+30kb
Cell line (Homo-sapiens)del(L1-R2)This paperDeletion of the CTCF binding sites L1-R2 of STITCH in STITCH+30kb
Cell line (Homo-sapiens)STITCH+30kb/KRABThis paperSTITCH+30kb with piggyBac integration of tetR-KRAB-2A-Puror
Cell line (Homo-sapiens)STITCH+30kb/tetR-3xFLAG-HAThis paperSTITCH+30kb with piggyBac integration of tetR-3xFLAG-HA-2A-Puror
Cell line (Homo-sapiens)STITCH+30kb with PurorThis paperSTITCH+30kb with Purorinside STITCH
Cell line (Homo-sapiens)del(30-440)/KRABThis paperdel(30-440) with piggyBac integration of tetR-KRAB-2A-Puror
Cell line (Homo-sapiens)NEUROG2/KRABThis paperSTITCH insertion into
the 65 kb downstream of NEUROG2 in Hap cells, with piggyBac integration of tetR-KRAB-2A-Puror
Transfected construct
(Escherichia virus P1)		Cre Recombinase encoding mRNAOZ BiosciencesCat#MRNA32-20synthetic mRNA encoding Cre recombinase
Antibodyanti-CTCF (Rabbit polyclonal)MilliporeCat#07–729, RRID:AB_441965ChIP (1:88)
Antibodyanti-H3K4me3 (mouse monoclonal)MAB InstituteCat#MABI0304S, RRID:AB_11123891ChIP (1:147)
Antibodyanti-H3K27me3 (mouse monoclonal)MAB InstituteCat#MABI0323S, RRID:AB_11123929ChIP (1:220)
Antibodyanti-H3K9me3 (mouse monoclonal)MAB InstituteCat#MABI0318SChIP (1:176)
Antibodyanti-H3K27ac (mouse monoclonal)MAB InstituteCat#MABI0309S, RRID:AB_11126964ChIP (1:220)
Recombinant DNA reagentpUC-STITCH (plasmid)This paperAddGene 129535A plasmid carrying STITCH with the homology arms with the MYC+30kb integreation site. Supplementary file 1B
Recombinant DNA reagentpUC57-PB-PGK-tetR-KRAB-2A-Puro (plasmid)This paperAddGene 129536A piggyBac transposon vector encoding tetR-KRAB-2A-Puror under the PGK promoter.
Recombinant DNA reagentpUC57-PB-PGK-tetR-3xFLAG-HA-2A-Puro (plasmid)This paperAddGene 129537A piggyBac transposon vector encoding tetR-3xFLAG-HA-2A-Purorunder the PGK promoter.
Recombinant DNA reagentSuper PiggyBac Transposase Expression VectorSystem BiosciencesCat#PB210PA-1
Sequence-based reagentAlt-R CRISPR tracrRNAIntegrated DNA TechnologiesCat#1072532
Sequence-
based reagent		Alt-R CRISPR crRNAIntegrated DNA TechnologiesSupplementary file 1A
Sequence-based reagentPCR primersThis paperSupplementary file 1C-G
Peptide, recombinant proteinAlt-R S.p. Cas9 Nuclease 3NLSIntegrated DNA TechnologiesCat#1074181
Peptide, recombinant proteinDynabeads Protein GThermo Fisher ScientificCat# 10003D
Peptide, recombinant proteinmicrococcal nucleaseNew England BiolabsCat#M0247S
Peptide, recombinant proteinNlaIII restriction enzymeNew England BiolabsCat#R01254C-seq Library Prep
Peptide, recombinant proteinDpnII restriction enzymeNew England BiolabsCat#R05434C-seq Library Prep
Peptide, recombinant proteinT4 DNA ligaseThermo Fisher ScientificCat#EL00144C-seq Library Prep
Peptide, recombinant proteinTks Gflex DNA
Polymerase		TakaraCat#R060A4C-seq Library Prep
Commercial assay or kitNEBNext Poly(A) mRNA Magnetic IsolationNew England BiolabsCat#E7490S
Commercial assay or kitNEXTflex Rapid RNA-Seq KitBioo ScientificCat#NOVA-5238–01
Commercial assay or kitNEBNext Ultra II DNA Library Prep with Sample Purification BeadsNew England BiolabsCat#E7103S
Chemical compound, drugDoxycyclineSigma AldrichCat#D9891
Chemical compound, drugEPZ-6438Adipogen Life SciencesCat#SYN-3045-M001
Chemical compound, drugLDN-193189StemRD
Chemical compound, drugSB-431542TocrisCat#1614
Software, algorithmWebGestaltPMID:31114916http://www.webgestalt.org
Software, algorithmDESeq2PMID:25516281
Software, algorithmtopGOPMID:16606683
Software, algorithmBowtie2PMID:22388286
Software, algorithmFourCSeqPMID:26034064
Software, algorithmHISAT2PMID:31375807
Software, algorithmHOMERPMID:20513432
Software, algorithmHTSeqPMID:25260700
Software, algorithmIntegrated
Genome Viewer		PMID:21221095
Software, algorithmGimmeMotifsPMID:21081511
Software, algorithmSAMtoolsPMID:19505943
Software, algorithmBEDtoolsPMID:20110278
Software, algorithmRCRAN

Cell culture

Request a detailed protocol

The human iPSC line 253G1 (Nakagawa et al., 2008) was kindly provided by Prof. Shinya Yamanaka through RIKEN BRC. We cultured the cells in the StemFit AK02N medium (ReproCELL, Cat#RCAK02N) on dish coated with iMatrix-511 (ReproCELL, Cat#NP892-012) without feeder cells. We added Y-27632 (FUJIFILM Wako, Cat#036–24023) at the final concentration of 10 μM when seeding the cells on a dish. We used the 0.5x of TrypLE Select (Thermo Fisher Scientific K.K., Cat#12563–011) to dissociate the cells for passaging. The iPSCs were sampled for assays in their growth phase, well before the color of the medium turns yellow and cells reach near confluency.

To differentiate the iPSCs to NPCs, we let the iPSCs become almost confluent and then switch the medium to the neural induction medium consisting of 1:1 of DMEM/Ham's F-12 (FUJIFILM Wako, Cat#042–30795) and Neurobasal Plus Medium (Thermo Fisher Scientific K.K., Cat#A3582901), 1X GlutaMAX Supplement (Thermo Fisher Scientific K.K., Cat#35050061), 1X MEM Non-Essential Amino Acids Solution (Thermo Fisher Scientific K.K., Cat#11140050), 1X Penicillin-Streptomycin (Thermo Fisher Scientific K.K., Cat#15140122), 1X N-2 Supplement (Thermo Fisher Scientific K.K., Cat#17502048), 1X B-27 supplement (Thermo Fisher Scientific K.K., Cat#17504044), 0.1 mM 2-Mercaptoethanol (Sigma, Cat#M7522), 250 nM LDN-193189 (StemRD), and 10 μM SB-431542 (Tocris, Cat#1614). The medium was changed every or every other day up to day 6.

DOX (Sigma, Cat#D9891) was basically added at the final concentration of 10 ng/ml unless specifically indicated. When DOX was removed for time-course analysis, the concentration was first changed to 1 ng/ml one day before the start of removal. Then at the start of the removal, the cells were first washed with PBS (Thermo Fisher Scientific K.K., Cat#10010–049), and then fresh medium without DOX was supplied. Further two hours later, wash with PBS and replacement of medium was repeated to ensure the removal of DOX. EPZ (Adipogen Life Sciences, Cat#SYN-3045-M001) was used at the final concentration of 200 nM. For the DMSO controls, the same volume of DMSO as EPZ was added.

We verified the authenticity of the cells by confirming the presence of the pMX-KLF4 transgene in the 253G1 cells (Nakagawa et al., 2008; Takahashi et al., 2007) with PCR using a primer pair of 5'-CCCTCAAAGTAGACGGCATC-3' and 5'-GGTCTCTCTCCGAGGTAGGG-3'. We tested infection of mycoplasma with HiSense Mycoplasma PCR Detection Kit (CellSafe, Cat#HD-25), and confirmed they were negative.

Genome editing

Request a detailed protocol

To delete the 3 Mb region of the MYC locus, we co-transfected the RNP complex of CRISPR/Cas9 targeting both edges of the deletion interval with Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific K.K., Cat#13778030) (Figure 1B). We assembled the RNP from Alt-R CRISPR crRNA (Integrated DNA Technologies, listed in Supplementary file 1A), Alt-R CRISPR tracrRNA, and Alt-R S.p. Cas9 Nuclease 3NLS (Integrated DNA Technologies, Cat#1072532 and Cat#1074181, respectively), following the manufacturer's protocol. The target sequences of the guide RNAs are described in Supplementary file 1A. After the transfection, cells were sparsely re-plated on a dish. Grown colonies were picked up and expanded. The clones were screened for the correctly edited allele by PCR genotyping (see Supplementary file 1D for the primer sequences). We then confirmed the deletion by direct Sanger sequencing.

The STITCH vector targeting into the +30 kb position with the homology arm of 150 bp length at each side was synthesized by Integrated DNA Technologies (see Supplementary file 1B for the DNA sequences). We amplified the fragment by PCR (see Supplementary file 1C for the primer sequences) with Tks Gflex DNA Polymerase (Takara, Cat#R060A) and purified it. Then we transfected it into the cells with Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific K.K., Cat#L3000001) together with the RNP complex of CRISPR/Cas9 targeting the insertion site as described above and transfected with Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific K.K., Cat#13778030). See Supplementary file 1A for the target sequences of the guide RNAs. The positive cells were first selected in the culture medium containing 0.2 mg/L puromycin. Then survived colonies were picked up and expanded. The correct insertion was confirmed by PCR and direct sequencing. We found a single nucleotide mutation within the R3 sequence in the clone that we obtained, which was far away from the core motif for CTCF binding for more than 30 bp. To insert STITCH into the other four sites at the MYC locus and the one at the NEUROG2 locus, we attached 50 bp homology arms by PCR using the STITCH vector as the template (see Supplementary file 1C for the primer sequences) and performed the transfection as the same way as above. We screened puromycin resistant clones and then confirmed the insertion by PCR (see Supplementary file 1D for the primer sequences). These targeted cells were further transfected with Cre Recombinase encoding mRNA (OZ Biosciences, Cat#MRNA32-20) using Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific K.K., Cat#13778030) to remove the puromycin resistant cassette (Figure 1A). After the transfection, the cells were sparsely plated on a dish, and colonies were picked up after they formed. We screened positive clones by PCR (see Supplementary file 1D for the primer sequences).

To delete or invert the CTCF-binding sites within STITCH, we transfected CRISPR/Cas9 RNPs targeting the edges of the intervals of the deletion/inversion as described above (see Figure 3—figure supplement 1A–D). The target sequences of the guide RNAs are described in Supplementary file 1A. After transfection with the RNPs, the cells were sparsely seeded, and grown colonies were picked up. The mutations were first screened by PCR (see Supplementary file 1D for the primer sequences). Then the DNA sequences were confirmed by direct sequencing. While we tried to obtain the del(L2-R2) clones, we obtained the del(L1-R2) clone, probably due to the excessive excision at the cutting site (Figure 3—figure supplement 1C).

To make the del(30-440) allele, we inserted the selection cassette only (i.e., the two loxP sites sandwiching the Puromycin resistant gene inside) of the STITCH vector into the +440 kb position of a delL clone in the same way as above (Figure 1—figure supplement 2). The targeting fragment was prepared by two rounds of PCR from the STITCH vector (see Supplementary file 1C for the primer sequences). After correct integration, Cre Recombinase encoding mRNA (OZ Biosciences, Cat#MRNA32-20) was transfected, and the deletion allele was selected by PCR screening (see Supplementary file 1D for the primer sequences).

To obtain cells that stably express tetR-KRAB, we designed a plasmid vector of a piggyBac transposon carrying coding sequence for tetR-KRAB followed by that of the 2A peptide and the puromycin resistant gene (PUROr) under the promoter of human PGK gene. The plasmid was synthesized by GenScript. We also designed the piggyBac vector containing tetR-3xFlag-HA instead of the KRAB fragment, followed by the same 2A-PUROr. This plasmid was also synthesized by GenScript. We transfected the plasmids with Super PiggyBac Transposase Expression Vector (System Biosciences, Cat#PB210PA-1) using Lipofectamine 3000 Transfection Reagent, and screened positive clones under puromycin selection, as described above. We obtained and characterized several clones, but picked one (the clone one in Figure 5A) for the subsequent analysis of STITCH/KRAB. We did not isolate single colonies for the NEUROG2/KRAB cells after the piggyBac integration, but only expanded all the cells that survived in the presence of puromycin. Therefore the NEUROG2/KRAB cells should be composed of heterogeneous populations with different integration sites of the piggyBac cassette. The positive cells were expanded and maintained in the presence of puromycin at 0.1 mg/L.

RNA extraction, cDNA synthesis, qPCR and library preparation for RNA-seq

Request a detailed protocol

RNA was extracted using the High-pure RNA isolation kit (Roche, Cat#11828665001) in the presence of the DNase I included in the kit for most of the study. We subsequently synthesized the cDNA with the High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific K.K., Cat#4368813). We used KAPA SYBR Fast qPCR Kit (Kapa Biosystems, Cat#KK4621) as the reagent and the Applied Biosystems 7500 Fast Real-Time PCR System (Thermo Fisher Scientific K.K.) for the qPCR reaction for most of this study. We used RNeasy mini kit for the RNA extraction and the Viia 7 Real-Time PCR System (Thermo Fisher Scientific) with TB Green Premix Ex Taq II (Takara Bio, Cat#RR820A) for the qPCR reaction for the analysis presented in Figures 7A–B and 8C, and Figure 8—figure supplement 2A–B. The primers used for qPCR assays are listed in Supplementary file 1E. To prepare libraries for RNA-seq, we first enriched mRNA using NEBNext Poly(A) mRNA Magnetic Isolation (New England Biolabs, Cat#E7490S). Then subsequently, we used NEXTflex Rapid RNA-Seq Kit (Bioo Scientific, Cat#NOVA-5238–01) for the library preparation with the oligo DNAs designed by ourselves (listed in Supplementary file 1G) as primers for the PCR reaction. The libraries were sequenced with HiSeq2500 System (Illumina) using HiSeq SR Rapid Cluster Kit v2-HS (Illumina, Cat#GD-402–4002) and HiSeq Rapid SBS Kit v2-HS 50 Cycle (Illumina, Cat#FC-402–4022).

4C-seq library preparation and sequencing

Request a detailed protocol

For a 4C-seq library prep, we collected c.a. 1 million cells and fixed them in 2% paraformaldehyde for 10 min at room temperature. Then the cells were lysed in lysis buffer (50 mM Tris (pH7.5), 150 mM NaCl, 5 mM EDTA, 0.5% NP-40, 1% Triton X-100, 1x complete proteinase inhibitors (Roche, Cat#11697498001); 1 ml), passed through a 23-gauge needle, pelleted and frozen in liquid nitrogen. After the cells were resuspended in H2O and CutSmart Buffer (New England Biolabs, Cat#B7204) and treated with 0.3% SDS and 2.5% Triton X100 at 37°C for 1 hr, respectively, we performed first digestion of the chromatin with 25 units of NlaIII restriction enzyme (New England Biolabs, Cat#R0125) on a rotator at 37°C for overnight. After heat inactivation of the enzyme, 12.5 units of T4 DNA ligase (Thermo Fisher Scientific, Cat#EL0014) were applied for self-ligation of the digested chromatin. After de-crosslinking and purification, we carried out second digestion with 20 units of DpnII restriction enzyme (New England Biolabs, Cat#R0543). Then the chromatin was again self-ligated with 12.5 units of T4 DNA ligase (Thermo Fisher Scientific, Cat#EL0014). We then performed the inverse PCR from the chromatin of the c.a. 1 million cells as the template to amplify the 4C library from a given viewpoint for 25 cycles with Tks Gflex DNA Polymerase (Takara, Cat#R060A). The primer sequences used for the 1st round of PCR are listed in Supplementary file 1F. We purified the DNA with High-pure PCR Product Purification Kit (Roche, Cat#11732676001) and performed the 2nd round of PCR to attach to the libraries adaptor and index sequences for the NGS analysis for eight cycles again with Tks Gflex DNA Polymerase (Takara, Cat#R060A). The DNA sequences of the adaptor/index primers are listed in Supplementary file 1G. The DNA was purified with High-pure PCR Product Purification Kit (Roche, Cat#11732676001). The final libraries were pooled and sequenced with the HiSeq2500 system, as described above, except for VP-R3. Note that the sequences were read from the side of NlaIII for VP-MYC1, -STITCH-left, -STITCH-right, and -NEUROG2 and the DpnII side for VP-MYC2. The sequencing for VP-R3 was performed with the iSeq 100 system using iSeq 100 i1 Reagent (Illumina, Cat# 20021533) in the paired-end mode. The reads from both NlaIII and DpnII sides were independently used for the subsequent analyses.

nChIP for histone modifications and CTCF binding, qPCR, and library preparation for sequencing

Request a detailed protocol

For nChIP for histone modifications, cells were dissociated from the dish with TrypLE Select (Thermo Fisher Scientific K.K., Cat#12563–011), washed with PBS, and frozen as pellets. After resuspension in ChIP dilution buffer (20 mM Tris-HCl pH8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100), supplemented with 0.05% SDS, 3 mM CaCl2 and protease inhibitors, they were incubated on ice for 10 min, and incubated at 37°C for 2 min. We added 0.48 μl of micrococcal nuclease (NEB, Cat#M0247S) per 1.0 million cells, and incubated them at 37°C for 10 min. To stop the digestion reaction, EDTA and EGTA were added, so the final concentration was 10 mM and 20 mM, respectively. To solubilize the chromatin, we applied sonication with Ultrasonic Homogenizer UH-50 (SMT Co., Ltd.) for three times of 20 s pulse and incubated them at 4°C for 1 hr. The solubilized chromatin after removal of the cell debris by centrifugation was incubated with antibodies at 4°C for overnight. We used 0.6, 0.4, 0.5 and, 0.6 μl of antibodies per 400,000 cells for H3K4me3, H3K27me3, H3K9me3, and H3K27ac (MAB Institute, Cat#MABI0304S, Cat#MABI0323S, Cat#MABI0318S, and Cat#MABI0309S), respectively. The chromatin with the antibodies was incubated with 6 μl of Dynabeads Protein G (Thermo Fisher Scientific, Cat# 10003D) for one hour. Then the beads were washed three times with ChIP dilution buffer supplemented with 0.05% SDS and subsequently twice with high-salt wash buffer (20 mM Tris-HCl pH8.0, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.05% SDS). The chromatin was treated with RNase A (50 ng/μl) at 37°C for 15 min and then with Proteinase K (100 ng/μl) at 55°C for 1 hr in ChIP extraction buffer (20 mM Tris-HCl pH 8.0, 300 mM NaCl, 10 mM EDTA, 5 mM EGTA, 0.1% SDS). The DNA was precipitated with ethanol and eluted in 10 mM Tris-HCl pH 8.0 after removal of the beads. We performed the CTCF nChIP exactly as described before with the same polyclonal anti-CTCF antibody (Millipore, Cat#07–729) (Tsujimura et al., 2018). For qPCR assays, we used KAPA SYBR Fast qPCR Kit (Kapa Biosystems, Cat#KK4621) as the reagent and the Applied Biosystems 7500 Fast Real-Time PCR System (Thermo Fisher Scientific K.K.) as the platform for the most of this study. We also used the Viia 7 Real-Time PCR System (Thermo Fisher Scientific) with TB Green Premix Ex Taq II (Takara Bio, Cat#RR820A). To prepare nChIP-seq libraries, we used the NEBNext Ultra II DNA Library Prep with Sample Purification Beads (NEB, Cat#E7103S). We basically followed the protocol from the manufacturer but used partly oligo DNAs that we designed by ourselves for the PCR reaction as listed in Supplementary file 1G,H. The libraries were sequenced with the HiSeq2500, as described above.

Data analysis of qPCR assay for gene expression levels

Request a detailed protocol

We first confirmed that the amplification efficiency is nearly 100% for all the primer pairs. Therefore, we used the ΔΔCt method to obtain the relative expression levels normalized to ACTB. As a reference sample, we used a large stock of cDNA prepared from the same iPSC line (253G1), which were cultured in a different condition from the present study (with feeder cells in a different medium), and always placed the reference sample in duplicates or triplicates in the same PCR plates, when measuring the Ct values of samples.

Replicates were defined differently for different experimental purposes. For STITCH insertions and del(30-440), replicates mean independent clones that were segregated after Cre transfection. The relative expression levels were measured for each clone and plotted in Figure 1D. For mutant clones of STITCH, replicates mean independent clones after CRISPR/Cas9 genome editing. The relative expression levels were measured for each clone and plotted in Figure 3A. We also obtained sub-clones from Hap and treated them as replicates in Figure 1D and Figure 3A. In Figures 1D and 3A, the mean values of the replicates were also represented as bars. The relative expression levels of the STITCH mutants and the Hap clone in Figure 3I and J, and Figure 3—figure supplement 2B and C were the mean values of the replicates. We obtained five and three clones after the transfection of tetR-KRAB and tetR-3xFlag-HA transposons, respectively. The relative expression levels of MYC and the puromycin resistant gene were assayed for all of these clones in Figure 5C and Figure 5—figure supplement 1A and B. For the treatment of the STITCH/KRAB cells with DOX and EPZ, we used only one representative clone (the clone 1), and performed replicate experiments, which mean samples separately treated with drugs in different dishes (Figures 5D and 7A,B,I,J). For the NEUROG2/KRAB cells, we obtained only one group of cells and performed replicate experiments for each condition. We performed one-way ANOVA with Tukey’s multiple-comparison post hoc test to infer statistical significance between different conditions in Figure 7A and B, and one-sided Welch's two-sample t-test in Figure 7I,J for the statistical significance between the DMSO and EPZ treatments and in Figure 8C for the statistical significance between with and without DOX. The data were represented as graphs with the ggplot2 package in R.

Cell proliferation assay

Request a detailed protocol

To compare cell proliferation rates between conditions with and without DOX, we first seeded equal volumes of cells in three replicates for each from a single population in the same medium without DOX. On the next day, we replaced the medium with a fresh one with or without DOX. After five days, the cell numbers were counted using a hemocytometer. We represented the relative proliferation rates as normalized cell numbers divided by the mean number of cells in the DOX minus condition. The assay was performed for both the Hap and STITCH/KRAB clones. We performed two-sided Welch's two-sample t-test to infer the statistical significance between the two conditions.

Data analysis of RNA-seq

Request a detailed protocol

We prepared and sequenced libraries from three replicate clones (see above) for each of Hap, STITCH+30kb, and del(30-440). We first combined separately sequenced reads of the same libraries from different lanes as fastq files. We mapped the sequences to the human genome (hg19) with HISAT2 (Kim et al., 2019). We made BedGraph tracks with HOMER (Heinz et al., 2010) and visualized them in Integrative Genomics Viewer (version 2.4.6) (IGV) (Robinson et al., 2011). The data ranges are indicated by counts per 10 million. We assigned the mapped reads to annotated genes with HTSeq (Anders et al., 2015). We normalized the counts and calculated log two fold changes between different conditions with the ‘normal’ shrining algorithm in DESeq2 (Love et al., 2014). To perform GSEA, we input the shrunken log2 fold change values into WebGestalt (http://www.webgestalt.org) (Liao et al., 2019), selecting GSEA (Subramanian et al., 2005) as the method and HALLMARK50 (Liberzon et al., 2015) as the functional database. To call differentially expressed genes, we set the threshold as the adjusted p-value<0.05 and the shrunken log2 fold change >0.5 with DESeq2. We visualized the shrunken log2 fold changes and the base means as the MA-plots using the ggplot2 package in R. The Venn diagram was drawn with the VennDiagram package in R (Chen and Boutros, 2011). The GO term enrichment analysis was performed with the topGO package in R (Alexa et al., 2006), where Fisher's exact test was employed for the statistical test. The data were visualized with the ggplot2 package in R.

Data analysis of 4C-seq

Request a detailed protocol

We only employed a representative clone for each genomic configuration for the 4C-seq assays. However, for each viewpoint in the most cases, we prepared a couple of replicate libraries that were separately prepared from different dishes, to confirm the reproducibility of the experiment (Figure 1—figure supplement 1B and Figure 3—figure supplements 2A and 3A).

We first combined separately sequenced reads of the same libraries from different lanes as fastq files. The sequences of the viewpoint fragment up to the restriction sites were removed with FASTX-Toolkit. Then we mapped the rest of the sequences to the human genome (hg19) using Bowtie2 mostly with the default settings except that the –score-min option was set as ‘L,−0.1,–0.1’ (Langmead and Salzberg, 2012). The generated SAM files were converted to BAM files, indexed and sorted with SAMtools (Li et al., 2009). We used the FourCSeq package to normalize the counts as reads per million (RPM), smooth them with the window size of seven fragments, and produce BedGraph files (Klein et al., 2015). We visualized the tracks in Integrative Genomics Viewer (version 2.4.6) (Robinson et al., 2011). The data ranges are indicated by counts per million. Counting the number of reads mapped to given regions was performed with BEDTools (version 2.26.0) (Quinlan and Hall, 2010). To calculate contact frequencies, we divided the read numbers in a given region by the total read numbers mapped to the defined locus except for the 10 kb region from the viewpoint fragment. When analyzing the directionality of chromatin folding, we combined the read numbers of replicates from the same viewpoints. To perform PCA, we first counted reads in defined bins. We took 30 kb and 10 kb as the sizes of the bins for VP-MYC1/2 and VP-NEUROG2, respectively. We combined the read numbers of replicates from the same viewpoints (either VP-MYC1 or VP-MYC2). Then we calculated ratios of reads in each bin within the region of interest (whole locus, the left 900 kb region, or the right 600 kb region for VP-MYC1/2). Then we performed PCA using the data sets with the prcomp function in R. The component loadings were calculated using the sweep function in R. The R codes used for the analyses are shown in Figure 3—source code 1 and Figure 8—source code 1. To perform the correlative analysis between the 4C-seq counts and gene expression levels, we also combined reads of replicates and calculated contact frequencies first. Then, the linear regression was performed against the log-log plot to obtain the slope in R. The Spearman's rank correlation coefficients were also calculated using a function in R. The log-log plots were visualized using the ggplot2 package in R.

Data analysis of nChIP-qPCR assay

Request a detailed protocol

We always took input samples for every nChIP and calculated enrichment as ratios to the input samples. Our replicates mean different nChIP samples derived from separately cultured cells in different dishes. In order to cancel the inevitable variance in the total enrichment efficiency of nChIP experiments, we normalized the enrichment at MYC to those at control regions, which were the ACTB region for the active H3K4me3 mark and the T region for the repressive H3K27me3 mark. As the treatment with EPZ causes an epigenetic change in genome-wide, we did not do the normalization in Figure 7G and H. To test statistical significance between different conditions in Figure 7C–F, we performed one-way ANOVA with Tukey’s multiple-comparison post hoc test. To assess statistical significance between treatments with DMSO and EPZ, we performed Welch's two-sample t-test in Figure 7G and H.

Data analysis of nChIP-seq

Request a detailed protocol

The reads from the same libraries were first combined as a fastq file when they were sequenced in different lanes. We mapped the data to the human genome (hg19) using Bowtie2 with the same options as the 4C-seq (Langmead and Salzberg, 2012). Then, we generated BedGraph files for visual inspection with HOMER (Heinz et al., 2010). Peak calling was also performed with HOMER. We also mapped reads to a synthetic genomic DNA carrying the STITCH sequence inside. For this purpose, we first retrieved unmapped reads and reads that are likely to be unique from the mapped BAM file with SAMtools, with scripts of ‘samtools view -b -f 4’ and ‘samtools view -b -q 10’, respectively, and combined them together, in order to remove reads that can be potentially mapped to repeat sequences. Then we re-mapped the reads against the custom reference genome. The subsequent generation of BedGraph files was carried out as above with HOMER (Heinz et al., 2010). We visualized the BedGraph tracks in IGV (Robinson et al., 2011). The data ranges are indicated by counts per 10 million.

To calculate the log2 fold change for H3K4me3, H3K27me3, H3K27ac, we first obtained a list of peaks that are called at least two among the four experiments (two replicates from the Hap and two from the STITCH+30kb). Next, we counted the read counts mapped to the peaks for each experiment. Then, we calculated the log2 fold change for each peak normalized by the size factors determined by the read counts in all the peaks, using the framework of DESeq2, without the shrinking algorithm. We ranked the peaks according to the values and plotted with the ggplot2 package in R.

Similarly, for CTCF nChIP-seq, we obtained a list of peaks that are called at least two among the six experiments (two replicates from STITCH+30kb, STITCH/KRAB with DOX, and STITCH/KRAB without DOX). We determined the orientations of the CTCF binding using GimmeMotifs (van Heeringen and Veenstra, 2011) with the position weight matrix from the HOCOMOCO database (Kulakovskiy et al., 2018), with the threshold of false discovery rate <0.1. To calculate the log2 fold change between plus and minus of DOX, we counted the read counts mapped to the peaks for each experiment. For the binding at STITCH, we separately count the reads against the synthetic genome. Then, we calculated the log2 fold change and the base means for each peak normalized by the counts in all the peaks, as described above. We ranked the peaks according to the log2 fold changes. The rank plot and the MA-plot were generated with the ggplot2 package in R.

Data availability

Allthe deep sequencing data of the 4C-seq, RNA-seq and nChIP-seqlibraries analyzed in this study were deposited in ArrayExpress:E-MTAB-7668, E-MTAB-7669, E-MTAB-7670, E-MTAB-8492, andE-MTAB-8957.

The following data sets were generated
    1. Tsujimura T
    (2019) ArrayExpress
    ID E-MTAB-7669. RNA-seq of wild type (Hap), insulation (STITCH+30kb) and deletion (del(30-440)) of the MYC enhancer in human iPS cells.
    https://www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-7669
    1. Tsujimura T
    (2019) ArrayExpress
    ID E-MTAB-7668. 4C-seq from viewpoint at MYC promoter (VP-MYC1 and VP-MYC2), in wild type (Hap) and variously modified alleles around the locus in human iPS cells.
    https://www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-7668
    1. Tsujimura T
    (2019) ArrayExpress
    ID E-MTAB-7670. nChIP-seq for CTCF, H3K4me3, H3K27me3 and H3K9me3, in wild type (Hap), STITCH+30kb and STITCH/KRAB clones of human iPS cells.
    https://www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-7670
    1. Tsujimura T
    (2019) ArrayExpress
    ID E-MTAB-8492. 4C-seq to show the effects of insertion of STITCH into MYC+30kb and NEUROG2-65kb positions on the chromatin conformation in human iPSCs and differentiated neural progenitor cells.
    https://www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-8492
    1. Tsujimura T
    (2020) ArrayExpress
    ID E-MTAB-8957. 4C-seq to show the effects of insertion of STITCH into NEUROG2-65kb positions on the chromatin conformation in neural progenitor cells differentiated from human iPSCs.
    https://www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-8957
The following previously published data sets were used
    1. Barakat TS
    2. Halbritter F
    3. Zhang M
    4. Rendeiro AF
    5. Bock C
    6. Chambers I
    (2016) NCBI Gene Expression Omnibus
    ID GSE99631. Functional dissection of the enhancer repertoire in human embryonic stem cells.
    https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE99631
    1. Dixon JR
    2. Jung I
    3. Selvaraj S
    4. Ren B
    (2015) NCBI Gene Expression Omnibus
    ID GSE52457. Global Reorganization of Chromatin Architecture during Embronic Stem Cell Differentiation.
    https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE52457

References

    1. Bretones G
    2. Delgado MD
    3. León J
    (2015) Myc and cell cycle control
    Biochimica Et Biophysica Acta (BBA) - Gene Regulatory Mechanisms 1849:506–516.
    https://doi.org/10.1016/j.bbagrm.2014.03.013

Decision letter

  1. Job Dekker
    Reviewing Editor; University of Massachusetts Medical School, United States
  2. Jessica K Tyler
    Senior Editor; Weill Cornell Medicine, United States
  3. Job Dekker
    Reviewer; University of Massachusetts Medical School, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This study describes development of a inducible insulator cassette, STITCH, that can act as boundary element that blocks long-range chromatin interactions. This can be a very valuable tool to dissect rules of long-range promoter – enhancer communication, and chromosome folding through mechanisms such as loop extrusion.

Decision letter after peer review:

Thank you for submitting your article "Controlling gene activation by enhancers through a drug-inducible topological insulator" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Job Dekker as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Jessica Tyler as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

In this paper, the authors present a new tool to modify chromosome structure and enhancer-promoter interactions. The major advance over previously identified insulators is that this tool (STITCH) is designed to include not only divergent tandem arrays of CTCF sites to create a boundary, but also interspersed tetO arrays to allow inducible regulation of CTCF binding. The tetO arrays are added between the CTCF sites such that with the addition of a tetR-KRAB transgene to the cell line, CTCF binding can be disrupted by inducing heterochromatin formation at STITCH in a doxycycline repressible manner. Disruption of CTCF binding to STITCH leads to increased interactions and enhancer-promoter contacts across the STITCH insertion site, and can modify gene expression. Therefore, STITCH is an inducible insulating element, which would be a broadly useful tool in the field of chromosome structure and beyond, such as in genome editing applications.

Essential revisions:

1) Given that STITCH is presented as a tool, all reviewers felt that the approach should be used for 2-3 case studies, ideally in at least 2 different cell types. This will help show how generally applicable the tool is and how robust the results are.

2) All reviewers felt that the PCA analysis was confusing and did not make a critical contribution. Please thoroughly revise the corresponding text and figures, or remove this analysis from the manuscript.

3) It is critical to show whether TetR binding alone, without the KRAB domain, affects CTCF binding and/or CTCF-mediated insulation.

4) The H3K27 ChIP experiments need to be repeated to provide better statistics to support the conclusion that changes in the levels of this mark can be accurately quantified. There are concerns this data set is not of high quality. Also, in Figure 4C-D positive and negative controls should be added.

5) The transcriptional analysis In Figure 2 could be improved. The cutoffs used to identify differentially expressed genes with DESeq2 are very loose (adjusted p-value = 0.1, no limitation is imposed on fold changes). We suggest to repeat the analysis with more than one clone per condition.

6) Please extend the discussion about the relationship between structural changes induced by STITCH and potential new loops induced by the ectopic CTCF sites in the light of the current understanding of CTCF orientation (and loop extrusion?). Reciprocal 4C viewpoints based on endogenous CTCF sites could also help clarifying this matter.

7) Upon publication we feel it is critical that the tools are made available to the community (e.g. putting plasmids on AddGene).

Reviewer #1:

The authors overstate the novelty of their results with respect to insulating elements. Insulating elements have been previously identified, and usually contain CTCF sites, similar to the STITCH sequence (Bell et al., (1999); Liao et al., (2018); Emery, (2011)). This should be more adequately referenced and introduced. In addition, it has already been shown that CTCF sites that form TAD boundaries will block enhancer-promoter interactions, yet this is presented as a novel result (subsection “Titrating blocking activity of STITCH by serial mutations of the CTCF binding sites”) (Hou et al., (2008); Guo et al., (2015); Braikia et al., (2017)).

The conclusion that the enhancer blocking activity and chromosome interaction activity are due to separate mechanisms is not sufficiently explained or supported. It is unclear why loop extrusion and enhancer blocking are introduced as separate mechanisms (Introduction), when the current understanding in the field is that enhancer blocking by CTCF sites is likely due to the creation of new TAD boundaries, which are formed by loop extrusion being blocked by CTCF sites (Recently reviewed in Schoenfelder and Fraser, (2019)).

Similarly, there appears to be confusion about the various mechanisms of long-range chromatin interactions in the Discussion section. Subsection “Mechanism of the STITCH insulation and its control by heterochromatin induction”, second papragraph is very confusing. In the current work CTCF is removed, from the locus, but cohesin is not. Therefore, there is now unblocked loop extrusion throughout the locus. Given that the interaction between the enhancer and MYC is blocked by CTCF, it is much more likely that the increased interaction between the super-enhancer and MYC when CTCF is removed is driven by loop extrusion.

Technical details and biological conclusions are not adequately explained in the text throughout this manuscript. The manuscript would benefit from improving the explanations of why specific analyses are used, and what the results signify biologically, beyond just stating the observations. In particular, the explanations of PCA analysis, component loading plots, and power law scaling between gene expression and 4C-seq contact frequency should be clarified. As it is presented now it is entirely unclear what the PCA analysis really adds. Additional technical details of the computational methods used to analyze the sequencing data is also needed, preferably an online repository for the code used to generate the plots and run the statistical tests should be included with the manuscript.

While 4C-seq is a useful technique for studying the specific interactions between MYC and surrounding loci, it would be beneficial to also compare this to an all-by-all or many-by-many chromosome conformation capture method such as Hi-C or 5C to show the endogenous organization of this region. Putting the STITCH insertion in the context of the landscape of genomic architecture (where are TADs or compartments found in this region?) would strengthen the manuscript and might help to understand how the different directionalities of the CTCF motifs in STITCH are working, as the current explanation is unclear. In addition, it would strengthen the manuscript to compare the contact frequency changes in the STITCH mutants at the MYC locus to changes in the endogenous TZ locus with similar modifications from previous publications by these authors, to determine how variable this behavior is at different genomic loci.

Reviewer #2:

The role of CTCF, cohesin, TADs and 3D genome organization in regulating gene expression and enhancer-promoter (E-P) contacts is currently being intensely studied. In particular, whether CTCF sites and TADs really regulate E-P contacts and gene expression has recently become controversial, with some studies claiming that CTCF and TADs have no role in regulating gene expression (e.g. see https://www.biorxiv.org/content/10.1101/609941v1).

Most studies (including the preprint above) have taken a "deletion" approach to this issue: take a natural E-P pair and TAD and then go in and start deleting or inverting etc. CTCF binding sites and see how gene expression is affected. What is nice about this study is that they take the opposite approach – a kind of "addition" approach. They add the STICH array of CTCF binding sites to the MYC gene at different locations and see how contact frequency (4C) and gene expression (qPCR) of MYC in hiPSCs is affected. The 2 most important points in my opinion are:

1) CTCF sites really can block E-P contact and strongly (~20-fold) affect gene expression. At least for the MYC gene.

2) Histone modifications can be used to turn ON and OFF CTCF insulation with quite high temporal resolution.

I believe the authors get about as close to causality as is realistic with the current tools of molecular biology, which is nice.

Beyond some important but addressable concerns (poor writing, at times confusing figures and presentation, occasionally poor referencing, tool availability), my major concern is this: The authors report STICH as a tool. 2 key features in a tool that are desirable to have are: (1) robustness and (2) generality. But because the authors only apply STICH to one locus (MYC) in one cell type, we cannot really tell if STICH is likely to block E-P contacts in general and robustly in many other loci and other cell types. The impact of STICH would have been greatly increased if the authors could have applied it to 2-3 case studies, ideally in at least 2 different cell types.

So overall, I believe this is a nice contribution with some really important insights, but that the general interest and impact could have been substantially improved if the authors had applied STICH to at least 2-3 different systems and if they can improve their presentation.

Specific issues:

Writing: the paper is for the most part reasonably written, but there are at least >25 cases of poor English and/or syntax/grammar issues. This is too many for a reviewer to fix and I suggest that the authors go through and clean up the issues.

Discuss results in context: First sentence of the Abstract and in the Introduction suggest that "regulation of gene-enhancer interaction is better understood,". I would argue that this is not true. In fact, recently several people (e.g. https://www.biorxiv.org/content/10.1101/609941v1) have begun arguing that CTCF plays essentially no role in the regulation of gene expression. The fact that the authors see such clear results on MYC, in my opinion only increases the impact and value of this study. Therefore, it would be nice if the authors could discuss their MYC result a little bit more clearly in the Discussion section in the context of the many recent studies arguing that CTCF plays no or only a minor role in regulating E-P contacts and gene expression.

Key resources should be available: First of all, my apologies if the authors already did this. But I tried to find this information and was unable to. The authors must put the key STICH plasmids on AddGene for the community, since the value of a tool is largely derived from it being readily accessible for the community. The DNA sequences of STICH must also be available with the paper. I could not find the DNA sequences of the full STICH sequence nor could I find the sequences of the specific CTCF binding sites. These must be available.

RNA-seq Results: The RNA-Seq studies in Figure 2 were really nice. But I could not understand why the STICH +30kb cell line would have ~2-3x more deregulated genes than the del(30-440) cell line. Although STICH is powerful, deleting the enhancer should still have a stronger effect on expression than just blocking it. Could the authors better explain this?

PCA-analysis: In Figure 3A-B, the analysis of how 4C reads in the different regions depend on the STICH construct was really nice. It was also very interesting to see the highly non-linear scaling between 4C contact frequency and gene expression (Figure 3I-J). Both of these are really important contributions in my opinion.

But the PCA-analysis was extremely confusing and convoluted. I really tried to follow the text and the figures, but it was very difficult for me to understand what the point was. In the Results section, the authors spend a lot of text and a huge number of figure panels on this, but I really could not understand it. My suggestion would be to remove all the figures and text pertaining to PCA or at least radically simplify the figures and the text to make it easier to understand. What is the major biological insight coming from this PCA analysis? What is component loading?

Subsection “Insulation and deletion of the enhancer resulted in similar transcriptome profiles”: authors out-of-the-blue reference VP-MYC1 and VP-MYC2, without any figure REF. I could not understand this.

tetR-KRAB studies: The Tet-R KRAB studies were very nice. I may have missed prior studies, but to my knowledge this is the first clear and causal demonstration that histone modifications can turn OFF CTCF insulation. However, one control I was missing was a DNA-binding control. TetR-KRAB binding could disrupt CTCF binding and insulation through 2 ways: DNA-binding competition (e.g. TetR-binding outcompetes CTCF binding) or KRAB-deposition of histone modifications. I would have liked to see a control showing that TetR binding alone – without the KRAB-domain – does not affect CTCF-mediated insulation.

But pretty neat to see that STICH insulation directly affects cell proliferation (subsection “Titrating blocking activity of STITCH by serial mutations of the CTCF binding sites”).

Figure 6. F and G have errors bars, but Figure 6 C, D, and E do not. Need to add errors bars to these.

Otherwise, the time-course results were also pretty cool.

Reviewer #3:

The manuscript describes a strategy to modulate chromosomal contacts in the vicinity of the endogenous MYC gene in human iPS cells through the ectopic insertion of an array of CTCF sites. The approach (named STITCH) seems to be able to alter MYC transcription levels, which correlates with changes in interaction frequencies between the MYC locus and a super-enhancer region downstream. The authors further monitor chromatin states at the engineered locus upon the induction of H3K9 trimethylation by targeted recruitment of a KRAB domain at the STITCH cassette, which is shown to disrupt CTCF binding and restore wild-type chromosomal contacts. The authors conclude that CTCF-mediated modulation of chromosome interactions is the driver of transcriptional changes.

The study is interesting and well designed, and has the potential to bring insight into how gene expression could be modulated by manipulating chromosome structure. However, it suffers from several major drawbacks that should be thoroughly addressed.

1) Many of the native ChIP-seq experiments in the manuscript are difficult to interpret and it is often difficult to agree with the authors on the changes they describe. The zoom level in all Figures is way too low to visually appreciate any local changes at the MYC locus, the STITCH cassette and the neighboring region. More importantly, some crucial experiments (notably the H3K27ac and H3K27me3 ChIP-seq reported in Figure 1, Figure 4 and Figure 4—figure supplement 1) appear to be strongly suboptimal. It is hard to imagine that a local increase of ~2 counts in a 0-6 range as reported in Figure 1 and Figure 4 really does correspond to a specific enrichment as opposed to technical noise.

The authors should perform new H3K27ac/me3 ChIP-seq experiments, provide statistics to support the notion that changes in these chromatin marks can really be quantified and discuss their findings in the light of the new experiments. Crosslinking ChIP-seq would be a viable option in this context – in fact I found it quite unclear why native ChIP was required in this particular study.

2) The correlation between MYC transcription levels and contact frequencies with the super-enhancer region (Figure 3) in mutant STITCH cell lines are interesting, and well supported by the large number of independent clones analyzed. Unfortunately, the structural changes induced by ectopic CTCF sites were not correlated with the position and orientation of endogenous CTCF sites. MYC itself is highly bound by CTCF, as can be more or less seen (again at regrettably low resolution) in Figure 5F. I would suggest the authors to thoroughly discuss the relationship between structural changes induced by STITCH and potential new loops induced by the ectopic CTCF sites in the light of the current understanding of CTCF orientation (and loop extrusion?). Reciprocal 4C viewpoints based on endogenous CTCF sites could also help clarifying this matter.

3) It is unclear how many copies of the STITCH cassette have been integrated at the MYC locus. The authors should provide evidence that a single insertion of 6 CTCF sites is actually responsible for the observed structural changes, as opposed to multiple tandem repeat insertions (especially since lipofection -and hence large amounts of DNA per cell- was used to generate the Cas9 assisted knock-in).

4) The transcriptional analysis In Figure 2 could be improved. The cutoffs used to identify differentially expressed genes with DESeq2 are very loose (adjusted p-value = 0.1, no limitation is imposed on fold changes). In the absence of differential gene expression analysis on more than one STITCH and del(30-440) clones, it is difficult to assess what the >1000 genes detected as differentially expressed under these loose criteria actually represent. I would suggest to repeat the analysis including more than one clone per condition and using more stringent criteria (e.g. padj<0.01, |log2(FC)|>1) in order to identify mis-regulated genes more robustly and reliably. Also, a qPCR validation of significantly up- or down-regulated genes is missing.

Finally, there is no explanation for the fact that the effect on transcription in the deletion mutant is smaller than in the STICH mutant. If the changes are indeed due to the insulation of the super-enhancer region from the MYC gene, then deletion of the super-enhancer region should lead to an even stronger effect on transcription.

5) It would be nice to prove that transcriptional changes in the STITCH and del(30-440) lines are really caused by downregulation of MYC, which could be done notably by overexpressing MYC and testing if normal expression programs are rescued.

6) The PCA analysis is Figure 3 is in principle interesting and laudable as an attempt to quantify differences in 4C profiles in a quantitative and unbiased way. However, the text is somewhat obscure and panels 3C-H are difficult to interpret. It is unclear why the results shown in Figure 3E-H, where PCA is performed on a subset of the data, are so different from panel 3D. These differences are acknowledged in the main text but I did not understand how they are interpreted by the authors. I would actually suggest that the text relative to Figure 3 is entirely re-written and clarified (e.g. please explain what "component loading" means in this context). In addition, the PCA results should be integrated with a discussion of whether they correlate or not with the position and orientation of endogenous CTCF sites (see point 2 above).

7) Transcriptional downregulation of MYC is attributed to changes in contact frequencies due to the presence of ectopic CTCF sequences at the STITCH cassette, which is supported by the strong correlation observed in Figure 3I. If this is really the case, and is due to CTCF looping from STITCH onto endogenous CTCF sites, then it should be possible to recapitulate the phenotype by deleting the endogenous partner CTCF sites. This would significantly strengthen the interpretation of the data.

8) in Figure 4C-D, negative and positive controls are missing (i.e. one or more regions where H3K4me3 should not be detected, and a region that is heavily bound by H3K27me, such as a poised gene, or a Hox gene). This is a very important control though, because one of the most interesting observations in the manuscript is that the transcriptional downregulation of MYC correlates with higher H3K27me3 levels. However, how much H3K27me3 is deposited? How does it compare with poised and/or inactive loci?

9) In Figure 5, it is impossible to understand which changes are occurring at the MYC promoter in terms of H3K9me3 and CTCF levels. This is nonetheless crucial to interpret the gene expression changes upon Dox induction and how they are related to targeted recruitment of KRAB. It seems that the CTCF signal is decreased also in the MYC promoter in the absence of Dox, and not only at the STITCH region. A zoom-in and quantification of ChIP-seq experiments (peak calling, integrated intensities of signals) should be provided, and the results should be discussed accordingly.

10) Along the same line, in Figure 6 a crucial missing information is how CTCF binding evolves in time at the STITCH cassette and at the MYC locus.

11) It is unclear what 'control' in Figure 7 refers to.

12) One very interesting observation is that MYC gains H3K27me3 upon STITCH insertion, which correlates with the observed level of insulation in the various mutants and with transcriptional activation/deactivation in time course experiments. However why is it so? If this happens as a consequence of physical insulation from the super enhancer, how do the authors interpret it? An alternative explanation is that PRC2 is recruited by sequences in the STITCH cassette, and helps repressing transcription. The delay observed between MYC deactivation/reactivation and the corresponding differences of H3K27me3 are not large enough to exclude this second hypothesis. Based on the experiment shown in Figure 7J it cannot be excluded that H3K27me3 levels are unchanged upon treatment with EPZ, in the absence of a carefully quantified ChIP experiment.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "Controlling gene activation by enhancers through a drug-inducible topological insulator" for consideration by eLife. Your article has been reviewed by Jessica Tyler as the Senior Editor, a Reviewing Editor, and two reviewers. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This study describes development of a inducible insulator cassette, STITCH, that can act as boundary. This can be a very valuable tool to dissect rules of promoter – enhancer communication, and chromosome folding through mechanisms such as loop extrusion.

We request that the authors address the following issues.

Essential revisions:

1) The authors were requested to put the structural changes induced by STITCH in the context of the overall CTCF binding patterns in the MYC region. They addressed this point on the one hand by performing new 4C experiments using the STITCH sequence as a viewpoint. These experiments, now in Figure 1, unfortunately do not seem to reveal how the various endogenous CTCF sites could be used to make new connections with the ectopic STITCH cassette and even a re-analysis of the 4C data with the MYC promoter as a viewpoint are inconclusive. The authors conclude vaguely that "It might be extrapolated from these previous results that there are not very specific endogenous regions that singly form loops with STITCH to organize the conformational changes induced by STITCH". The interpretation of the data in the rebuttal is also highly speculative and does really not address the reviewers' request that structural changes are evaluated in the light of a more global view of chromosome contacts such as the one provided by Hi-C data. In 4C it is always hard to detect loop extrusion-associated structural features such as loops and stripes (or flares), and it is not surprising that specific connections between CTCF sites might be missed without performing matched Hi-C or 5C experiments. The authors need to at least these limitations of their 4C-based analyses.

2) In the new Figure 8, new experiments are provided to support the applicability of STITCH to additional contexts. However, the data do not appear to fully support the conclusion that STITCH-mediated modifications of chromosome interactions are at the basis of the observed transcriptional effects. First, 4C experiments in panel F do not allow to conclude in any manner that STITCH alters conformation at the targeted allele. Certainly, the presence of the wild-type allele confounds the readout, but even considering this, the fluctuating small-% differences observed do not seem to be robust (also, information on replicates is not provided unless I am mistaken). The right experiment to address this point would have been to perform 4C from the STITCH cassette itself, which would allow to detect mainly contacts within the mutant allele and which is apparently technically possible given that similar experiments are reported in the new version of Figure 1. It would be important to add such 4C analysis.

Second, it appears that NEUROG2 downregulation is only shown for one population of cells following piggyBac-mediated insertion of the TetR-KRAB transgene. It is unclear whether these results would hold true if the transposition of the transgene is repeated in independent experiments. Given that piggyBac insertions typically occur in multiple genomic locations simultaneously in every cell, a possibility that cannot be excluded is that the transcriptional effect on NEUROG2 is a secondary effect of mis-regulation of one or more upstream genes that are accidentally targeted by the transposon. This should be discussed.

3) Introduction and Discussion section: the authors seem to confuse several concepts of loop extrusion, insulation and contact domains. Insulation by CTCF is the result of its ability to block loop extrusion. Insulation does not necessarily involve CTCF sites to loop with each other (in fact such loops barely insulate). Insulation and contact domains can occur even when interactions are not strictly divergent at boundaries: insulation can occur in only one direction when CTCF sites are all in the same direction. Such unidirectional insulation can demarcate contact domains. The authors are asked to consider these issues when revising the Introduction and Discussion section.

4) All reviewers found the manuscript extremely difficult to read. Please do not use track changes. The manuscript should be carefully edited.

5) All plasmids should be made available through AddGene.

https://doi.org/10.7554/eLife.47980.sa1

Author response

Summary:

In this paper, the authors present a new tool to modify chromosome structure and enhancer-promoter interactions. The major advance over previously identified insulators is that this tool (STITCH) is designed to include not only divergent tandem arrays of CTCF sites to create a boundary, but also interspersed tetO arrays to allow inducible regulation of CTCF binding. The tetO arrays are added between the CTCF sites such that with the addition of a tetR-KRAB transgene to the cell line, CTCF binding can be disrupted by inducing heterochromatin formation at STITCH in a doxycycline repressible manner. Disruption of CTCF binding to STITCH leads to increased interactions and enhancer-promoter contacts across the STITCH insertion site, and can modify gene expression. Therefore, STITCH is an inducible insulating element, which would be a broadly useful tool in the field of chromosome structure and beyond, such as in genome editing applications.

Essential revisions:

1) Given that STITCH is presented as a tool, all reviewers felt that the approach should be used for 2-3 case studies, ideally in at least 2 different cell types. This will help show how generally applicable the tool is and how robust the results are.

We agree that generality as a tool is very important for readers. As suggested, we newly tested the functionality of STITCH at another locus, namely near NEUROG2, in human iPSCs and neural progenitor cells (NPCs). Thanks to the suggestions, we could now include a discussion regarding this issue in the manuscript, as explained below.

Firstly, the previous studies have shown that the TZ at the mouse Tfap2c-Bmp7 locus is constantly bound by CTCF and organize chromatin contacts in various tissues (Tsujimura et al., 2015; 2018). In the present study, we show that the binding elements extracted from the TZ are also bound by CTCF and organize the chromatin conformation as a reconstituted cassette (STITCH) at a different locus (MYC) in a different species (human). These results well argue that STITCH, as well as the TZ, should be able to control chromatin contacts in various contexts through the binding of CTCF.

To further show the generality of the STITCH function, we inserted STITCH near NEUROG2 in human iPSCs and integrated the tetR-KRAB transgene into the cells (Figure 8). Then, we differentiated the iPSCs to NPCs where NEUROG2 is expressed. Our results clearly show that STITCH at this position also recruits CTCF and blocks the chromatin contact and that the KRAB induction again controls the CTCF binding and the chromatin in a drug-dependent manner in both cell types. Further, our data show that the expression is significantly down-regulated by STITCH in differentiating NPCs on day 4. We believe these results well support that the functionality of STITCH is quite robust and encourage researchers to apply the system for their researches.

Also, many other studies, as pointed out by the comment of the Reviewer#1 below, have already shown that inserting CTCF binding sequences insulate gene-enhancer interaction. Therefore, we firmly think that STITCH should be applicable to many genomic and cellular contexts, as presented in this study. However, as discussed below in our response to essential revision 6 and the revised manuscript, and also as remarked by reviewer#2, it is still quite elusive how CTCF manages this insulation process. In this sense, to fully describe the generality of the system, we may still need to wait for an accumulation of knowledge in the CTCF function.

2) All reviewers felt that the PCA analysis was confusing and did not make a critical contribution. Please thoroughly revise the corresponding text and figures, or remove this analysis from the manuscript.

We wish to apologize for not having explained well the methodology in our previous manuscript. Based on the reviewers' comments, we thoroughly simplified and revised both texts and figures. We also add the R codes that we used to analyze the data. So it should be now much more comprehensible than the previous version.

In brief, we think that applying PCA to analyze 4C-seq is a simple and powerful approach to extract crucial information from the data. Our PCA could, in fact, describe the preferential contact of MYC with the super-enhancer region more than with other non-enhancer regions in the vicinity in the absence of the STITCH insertion. As discussed in the following, we believe this finding should contribute to understanding how CTCF regulates the gene-enhancer interaction. Therefore, we would like to insist that this part is worth being kept and reported broadly to the community.

Improved parts in the revision

First, to improve the readability, we mainly revised the following four points:

1) We improved the appearance of PCA plots. We now use consistent colors and shapes to plot each allele in different panels, as suggested by Reviewer#1. We illustrate CTCF configurations nearby the plots to better compare the results. We also add interpretation of the PCA plots so our text should become more comprehensible while referencing the figure panels.

2) We simplified the organization of this PCA part and removed four panels in total from the main and supplementary figures.

3) We now explain how component loadings are calculated and what they mean, while providing the R code that we used to calculate the values and make the plots.

4) We thoroughly rewrote the texts to make what we think PCA could provide more precise and understandable.

Applying PCA to 4C-seq

Next, we would like to discuss what PCA provides. PCA allows us to compare 4C-seq data from multiple conditions (alleles) at once and at the same time to extract genomic regions that characteristically change contact patterns in correlation to the conditions.

Plotting the PC1 and PC2 values of each sample as a PCA plot illustrates how much the different alleles affect the contact profiles in comparison with each other (Figure 3C). If different alleles are arranged on the plot according to compositions of CTCF binding sequences, it means that the compositions of the CTCF binding sites are the major drivers to change the contact profiles within a given region. For example, in Figure 3C, the "non-blocking" alleles, namely WT(Hap) and del(L1-R3), show the lowest values of PC1, while the other alleles with CTCF binding sequences show higher values of PC1, illustrating that the largest variance among the samples is most likely due to the arrangements of the CTCF binding elements.

Then the component loadings of PC1 show which genomic regions (bins) are more loaded onto the PC1 values. The component loadings of PC1 are calculated as the product of the eigenvector and the square root of the eigenvalue of PC1. Component loadings correspond to the correlative coefficients between the component values of PC1 and the original frequency values of the binned genomic region. For example, in Figure 3D, the component loadings of the bins in the left 900-kb region are all nearly 1, showing that the contact frequencies of these bins well correlate positively with the PC1 values. It means that the samples with higher PC1 values have higher contact frequencies with the 900-kb region, which is, in fact, the case (Figure 3B). On the other hand, the bins in the right region, particularly the immediate 570-kb region, have component loadings of minus values, meaning that the contact frequencies of these bins correlate negatively with the PC1 values. In fact, the alleles with lower values of PC1 show higher contact frequencies with the right 570-kb region (Figure 3B). We think these data show a proof-of-concept that PCA can collectively represent the 4C contact patterns of multiple samples as a simple and powerful analytical method.

What PCA adds to this work

The PCA plot in Figure 3C indicates that the arrangements of CTCF differentiate the samples by two different effects: One is the blocking effect by the presence of CTCF against the non-blocking alleles; the other is the directionality effect segregating the CTCF arrays pointing to leftward from those to rightward. To disentangle the intermingled effects, we next performed PCA for subsets of the alleles. First, we removed the non-blocking alleles, namely WT(Hap) and del(L1-R3), leaving only the STITCH and the mutant alleles (Figure 3—figure supplement 4). This subset should reduce the blocking versus non-blocking effect. In fact, in Figure 3—figure supplement 4, the segregation is mainly seen between the leftward and the rightward alleles. We think this segregation well represents the differences of the directionality scores from VP-MYC1/2 in different alleles, which we now add as Figure 3—figure supplement 3B in this revision (for the directionality score, please refer to our response to essential revision 6 and the revised manuscript).

Next, we only used the non-blocking alleles and the non-directional blocking alleles, namely the original STITCH and inv(L1-R3), to reduce the effect of the directionality and enhance the blocking effect. Then the two groups are segregated along PC1 (Figure 3E). Also, the component loadings show complete switching at the insertion site of STITCH, which, of course, makes sense (Figure 3F).

Then, we asked if the left 900-kb or the right 600-kb regions contain internal regions that specifically associate with MYC depending on the absence and presence of CTCF. For this, we performed PCA using the contact frequencies only within the left or right regions (Figure 3G, H). Note that the contact frequencies are re-normalized only among the bins subject to PCA. If all the bins in either left or right side change the contact frequency more or less equally in response to CTCF, which perhaps might be a predominant expectation from the current understanding, there should not be clear segregation in PCA by the compositions of CTCF. In fact, we see no clear segregation in PCA for the left side region, suggesting that the binned regions in the left 900-kb region behave more or less equally with each other (Figure 3G).

However, for the right 600-kb region, this was not the case (Figure 3H). The plot shows segregation between blocking and non-blocking alleles. Most notably, the component-loading plot shows that the non-blocking alleles are more associated with the region corresponding to the super-enhancer. This means that STITCH does not just block the contact as a whole beyond the insertion site. STITCH instead seems to interrupt the interaction of MYC with the super-enhancer actively. As discussed later in our response to essential revision 6, we think this finding is important to understand how CTCF insulates the gene-enhancer interaction. Without PCA, at least for us, it would have been impossible to describe this feature. Thus, PCA can (and in fact did) extract valuable information from the 4C-seq data.

3) It is critical to show whether TetR binding alone, without the KRAB domain, affects CTCF binding and/or CTCF-mediated insulation.

Thanks for raising this point. We agree that this control is important. We newly performed nChIP-qPCR to confirm that CTCF binds to STITCH when the tetR-3xFLAG-HA was induced (Figure 5—figure supplement 1J). This result strongly suggests that KRAB induction, beyond simple binding of protein, is essential to disable the insulation by STITCH. Unfortunately, we failed to see significant enrichment of tetR-3xFLAG-HA at STITCH by several protocols of ChIP. We tried various conditions with help from Dr. Tomohiko Akiyama (Keio University School of Medicine), who has expertise in doing ChIP against FLAG-tagged transcription factors (Akiyama et al., 2015; Yukawa et al., 2014), but we could not establish a protocol to detect the binding of tetR at STITCH in the timeframe of the revision. We are aware of previous reports showing binding of tetR at tetO using similar tags (Moussa et al., 2019; Pourfarzad et al., 2013; Ragunathan et al., 2015). We think that, when compared to these studies, the difficulty we encountered might be attributable to the fact that our STITCH includes a total of only four elements of tetO that are sparsely arranged with each other within the 1.3-kb length of STITCH, while those successful studies used a clustered array of 7x or 10x tetO sequences.

Instead, we further tested if the STITCH at the same location but with the puromycin resistant cassette (PUROr), which is the one before Cre recombination was carried out, would be bound by CTCF. Here, the PUROr is transcribed from the cassette, so there should be binding of certain proteins at STITCH. Our nChIP-qPCR shows that CTCF still binds to STITCH at similar levels, and MYC is repressed (Figure 5—figure supplement 1K, L). These results strongly support that the KRAB induction is essential to expel the CTCF binding. We believe these data clarify the concerns raised by the reviewers.

4) The H3K27 ChIP experiments need to be repeated to provide better statistics to support the conclusion that changes in the levels of this mark can be accurately quantified. There are concerns this data set is not of high quality. Also, in Figure 4C-D positive and negative controls should be added.

To clarify the concerns in the revised manuscript, we first re-analyzed the ChIP-seq data (now Figure 4A-E). We also added positive and negative controls in Figure 4C-D (now Figure 4F-G). Further, we repeated the experiments in Figure 6F, G, Figure 7A-F, and obtained the same conclusion.

Reanalysis of nChIP-seq for H3K4me3, H3K27me3, and H3K27ac (Figure 4A-E)

Firstly, to demonstrate that the H3K4me3 and H3K27me3 profiles only changed at MYC, but not other loci, we presented magnified views of the profiles at MYC, T (repressive), ACTB (active), HOXD13 (repressive), and DPPA4 (repressive) loci (Figure 4A, Figure 4—figure supplement 4C). These panels well illustrate that while the epigenetic states at MYC were considerably changed, those at the other loci are unchanged at all. We also think that the magnified views well show that the enrichment we detect is well more than the backgrounds. Also, these figures should support that our assays are good enough to distinguish changed and unchanged epigenetic states in different conditions. Of note, the MYC locus lacks one allele, so we adjusted the count ranges to show in the tracks accordingly (0-6 for MYC, but 0-12 for others, as for H3K27me3).

Next, to further provide more statistic support to our observation, we calculated fold changes of read counts over peaks in genome-wide in STITCH+30kb against Hap. As shown in Figure 4C-E, the H3K4me3 and H3K27me3 peaks are ranked as one of the top peaks showing the most extensive changes. By contrast, the same analysis shows that the H3K27ac peaks detected in the super-enhancer/PVT1 region did not change much by the STITCH insertion. We think these results adequately support our conclusion that the epigenetic states were only altered at MYC.

Positive and negative controls in Figure 4F-G

We newly quantified the enrichment at DPPA4 and T as positive and negative controls for H3K4me3, respectively (Figure 4F). Similarly, the enrichment at HOXD13 and ACTB was added as positive and negative controls, respectively, for H3K27me3 (Figure 4G).

Repeated analysis of the time-course change of H3K4me3 and H3K27me3 at MYC (Figure 6F, G, Figure 7A-F)

We repeated the time-course experiments in Figure 6F, G, and Figure 7A-F. For Figure 6F, G, in this revision, we extended the timepoint up to 48 hours after the addition or removal of DOX, to compare the 24hour point with the later point. As we have shown in the previous manuscript, we could see again that the change of the H3K4me3 level is rapid, while the change of the H3K27me3 level is slow.

In Figure 7, we again sampled cells at 24 and 72 hours after DOX addition/removal together with DOX plus/minus controls, and examined the histone mark levels. Then again, we could reproduce our previous results that 24 hours is not enough for H3K27me3 to be entirely switched. Thus, our finding that H3K27me3 level only follows the gene expression change should be very robust.

5) The transcriptional analysis In Figure 2 could be improved. The cutoffs used to identify differentially expressed genes with DESeq2 are very loose (adjusted p-value = 0.1, no limitation is imposed on fold changes). We suggest to repeat the analysis with more than one clone per condition.

Thanks a lot for the suggestions. In the previous version of our manuscript, we used the threshold (p-adjusted < 0.1), as it was used for demonstration and benchmarking in the paper reporting the development of DESeq2 (Love et al., 2014). In this revision, we re-analyzed the data with a tighter threshold (p-adjusted < 0.05, log2-fold-change > 0.5). This setting called less number of differentially expressed genes (Figure 2B-D, Figure 2—figure supplement 2B, C). Nonetheless, the GO enrichment analysis among the commonly down-regulated genes still shows enrichment of categories such as cholesterol synthesis and rRNA processing, as we described in the previous version (Figure 2—figure supplement 2B, C).

As everyone might agree, deciding on the threshold is a little bit arbitrary. In this sense, we found Gene Set Enrichment Analysis (GSEA) is attractive, because this algorithm does not impose threshold setting (Subramanian et al., 2005). Also, as pointed out by reviewer#1's comment, the primary aim of this analysis should be on whether the target of MYC is affected by the mutations or not. We found that the "HALLMARK 50" from MSigDB includes categories of MYC targets (Liberzon et al., 2015). Therefore, in this revised manuscript, we newly performed GSEA against HALLMARK 50. Figure 2E and F show that the analysis detected quite strong enrichment of the MYC target categories in down-regulated genes in both STITCH+30kb and del(30-440). The other enriched categories are shared between the two mutations, suggesting that the tendency of the transcriptional change is quite similar between them (Figure 2E, F). Moreover, many of the enriched categories have already been suggested to be subject to MYC function in various cell types. Thus, the reanalysis could show that MYC was effectively down-regulated by the mutations to affect the expression of its target genes that seem to be shared in many cell types.

Regarding the suggestions to use different clones per condition, the three replicates that we used were actually from three different (sub-)clones, as described in the Materials and methods section in the previous manuscript. We now also indicate it in the corresponding part in the revised Results section. As indicated here, the used clones were: as for Hap the parental clone and two isolated sub-clones derived from the parental one; as for STITCH+30kb and del(30-440), three different clones isolated upon the final Cre recombination step, respectively. To analyze completely independent clones for each should have been ideal for fully controlling the clonal variations. However, it would be pragmatically very complicated to do so here, because introducing these mutations needs quite a few steps of cloning.

We still think our results are valid enough for the following reasons. Firstly, we did obtain different clones at one step for each condition, so much of the cloning effects should have been well controlled. Secondly, we confirmed that the deletion of STITCH or the KRAB induction well recovered the MYC expression level back to the normal level (Figure 3A and Figure 5C), so at least the repression of MYC observed in this study should not be attributed to the clonal difference. Therefore, we do not think that our data are suffering much from variations due to the cloning processes.

6) Please extend the discussion about the relationship between structural changes induced by STITCH and potential new loops induced by the ectopic CTCF sites in the light of the current understanding of CTCF orientation (and loop extrusion?). Reciprocal 4C viewpoints based on endogenous CTCF sites could also help clarifying this matter.

Thanks a lot for this suggestion. In the previous version, the 4C-seq was only from viewpoints at MYC. So, we could not discuss much how STITCH itself was involved in the chromatin structure. In this revised manuscript, we newly performed 4C-seq from the flanking sites of STITCH as viewpoints, as we thought it should be more direct to discuss this matter (now Figure 3—figure supplement 3). We also put the CTCF binding sites and their orientations, as well as the endogenous domain structures along some of the 4C-seq tracks (Figure 1B, C, Figure 3—figure supplement 3A). With these data, we extended the discussion in the context of CTCF loops. In summary, we cannot conclude that any specific loops or loop/contact domains are essential for STITCH insulation. We instead claim that considering a model linking the functionality of CTCF (and loop extrusion) directly to the disruption of gene-enhancer interaction should be required, as explained below.

Formation of loops?

We visually inspected the new 4C-seq plots and found that there seem to be several peak-like bumps in some of these plots (Figure 3—figure supplement 3). These bumps may represent new loops that STITCH creates. However, we cannot tell if there is anything special that would be regulating the insulation process, simply because the bumps are not very striking. To clarify if there is a newly formed loop that would be important, it is necessary to delete endogenous CTCF binding sites (or whatever) that are involved in this loop formation, as suggested by reviewer#3. However, we could not perform this experiment for this revision, firstly because the new 4C-seq plots did not indicate encouraging peaks to test and secondly because it would be certainly impossible to finish the analysis in a reasonable timeframe here. Also, please note that the deletion of one region might only result in reestablishment of another loop with some remaining regions, possibly the CTCF binding sites next to the deleted one, which would make the experimental design very complicated.

However, we would like to point out that we had already performed analogous experiments in our previous study (Tsujimura et al., 2018). In Figure 5 of this paper, we analyzed contact profiles of Tfap2c, which carries CTCF binding sites nearby, upon several mutations of the TZ. First, the inversion experiment of Tfap2c indicated that the directional folding around Tfap2c greatly depends on the locally associated CTCF binding sites. On the other hand, the mutations around the TZ did not affect the folding directionality of Tfap2c almost at all. This result suggests that the conformational change imposed by a CTCF binding site (near Tfap2c) does not rely on looping with another CTCF binding site located distantly (around the TZ). Also, the recent paper, which was kindly introduced by reviewer#2, deleted CTCF binding sites that bridge loops to establish the Shh loop domain. Interestingly, it did not result in an appreciable change of the Shh expression, strongly suggesting that a specific loop may not be relevant to genome regulation (Williamson et al., 2019).

Based on these results, we now have a view that discussing the formation of new stable loops with specific endogenous sites might not be directly fruitful to gain insights into how CTCF insulates the gene-enhancer interaction. In the revised manuscript, we describe these thoughts along with the presentation of Figure 3—figure supplement 3.

Of important note, we believe that loops anchoring the insulating CTCF sites should be rather crucial for the disruption of the gene-enhancer interaction, as explained in the following. However, according to what we think, it might not matter much whether the loops are formed stably with specific CTCF binding sites or dynamically/promiscuously with any other sites, as long as they are anchored at the insulating CTCF site (now Figure 9E). In this sense, we do NOT think that the current understanding of the insulation-by-looping model should be excluded.

Loop/contact domains?

We also have extended discussion regarding the current dogma of enhancer regulation by contact domains. As discussed right above, our methodology could not identify specific loops created by STITCH. Therefore, this study is not able to tell the overall re-organization of domains induced by STITCH. As suggested by reviewer#1, Hi-C or 5C should be required to do so. However, domains are basically defined by "boundaries" that exhibit diverging directionality of chromatin folding. So, it was possible for us to analyze the transition of folding directionality around the insertion site of STITCH (Figure 3—figure supplement 3B). We found that the divergence of folding directionality is not evident across STITCH, particularly in the delL allele, suggesting that the formation of domains or establishment of domain boundaries is not prerequisite for the STITCH insulation (Figure 3—figure supplement 3B).

We are quite aware that many previous studies (including (Tsujimura et al., 2015)) supports the currently recognized model that contact domains restrict enhancer targets (Schoenfelder and Fraser, 2019), as indicated by reviewer #1. However, we now wonder if this is really the case. We even find that questioning or neglecting this dogma seems helpful to interpret the data presented in this study and others. So please allow us to explain what we think regarding this idea here.

As far as we understand, what have been shown so far are primarily the followings:

1) Boundaries/CTCF binding sites exhibit the orientation-dependent directionality of chromatin folding most likely through the loop extrusion.

2) Contact domains emerge as a consequence of the directional folding (or the loop formation) by the function of the boundaries/CTCF binding sites (and cohesin loops).

3) Enhancer allocation is mostly restricted within contact domains.

4) Genetic manipulation of boundaries/CTCF binding sites (but not directly of contact domains) alters the gene-enhancer interaction.

These data underlie the current understanding that CTCF creates (boundaries of) domains, which then restrict the enhancer allocation. However, it should be questioned if taking the contact domains into account is genuinely essential to explain the enhancer regulation because there do not seem studies directly showing the causative role of the domains per se, but not the CTCF/loop extrusion. The above data should also corroborate another idea that CTCF/loop extrusion limits the enhancer targets and at the same time, establish contact domains (or chromatin organization that can be called contact domains). What has been shown so far is the only correlation between the presence of contact domains and enhancer regulation through analysis of CTCF binding sites.

For this reason, we had put loop extrusion and enhancer blocking as separate mechanisms in the Introduction. As our text was pointed out as unclear by reviewer #1, we have now added more of these explanations in the revised manuscript for it.

From this perspective, it should not be surprising that we did not find evidence of domain boundaries at STITCH from the analysis of folding directionality. Of course, absence of evidence is not evidence of absence. However, at least our data does not encourage the domain-centric view.

Besides, we would like to point out some data of this work and other studies that the regulation-by-domain model does not explain well. First, (Bintu et al., 2018) has shown that contact domains represent averaged projection of various domain-like structures. This means that CTCF/cohesin allows domain-like association across domain boundaries. Then, why does the gene-enhancer interaction not take place across boundaries?

Similarly, we (and surely many other studies), in fact, detect inter-domain contacts between genes and enhancers. For example, our 4C-seq results show that STITCH reduces the contacts with the enhancer region only by half. However, the change in the expression level is more than 20 times. We think this discrepancy is quite puzzling. Is there really any difference between inter-domain and intra-domain contacts? Is there any evidence for it? Then what was the domain-like structure across the boundary observed by DNA FISH (Bintu et al., 2018)?

CTCF-centric view

We think the most straightforward interpretation of these observations should be that CTCF/loop extrusion somehow interrupts the gene-enhancer interaction when inserted in between, regardless of the formation of loop/contact domains. In this sense, we think our PCA has provided valuable insights. The analysis shows that the presence of CTCF more pronouncedly decreases the contacts of MYC with the enhancer region more than the other non-enhancer regions around. Based on this finding, we have proposed two models for enhancer regulation by CTCF, as illustrated in Figure 9D-E, which is added for this revision. One is that the gene-enhancer interaction is boosted upon an increase of overall contact frequency, possibly by an enhanced phase-separation process (Figure 9D). The other is that (either dynamic or stable) loop extrusion anchored at the insulating CTCF efficiently disrupts the gene-enhancer interaction (Figure 9E). Of course, these models are elusive. However, we believe they are reasonable enough to be proposed in this manuscript. Along this line, it should be pointed out that the regulation of gene-enhancer interaction by CTCF is not fully understood. Therefore, as stated in our response to the essential revision 1, the generality of STITCH cannot be described entirely for the moment.

7) Upon publication we feel it is critical that the tools are made available to the community (e.g. putting plasmids on AddGene).

Thanks a lot for this suggestion. It will be our pleasure if many people use our system for their researches. Accordingly, we have deposited the three plasmids carrying STITCH, tetR-KRAB-2A-Puro, and tetR-3xFLAG-HA-2A-Puro, respectively, to AddGene. They should become soon available to the broad community from them. Please also see the Key Resources Table in the revised manuscript.

Reviewer #1:

The authors overstate the novelty of their results with respect to insulating elements. Insulating elements have been previously identified, and usually contain CTCF sites, similar to the STITCH sequence (Bell et al., (1999); Liao et al., (2018); Emery, (2011)). This should be more adequately referenced and introduced.

Thanks for this comment. It was not our intention to claim novelty for it. We have now introduced some of these studies in the Introduction to explain previous works regarding the CTCF insulation.

In addition, it has already been shown that CTCF sites that form TAD boundaries will block enhancer-promoter interactions, yet this is presented as a novel result (subsection “Titrating blocking activity of STITCH by serial mutations of the CTCF binding sites”) (Hou et al., (2008); Guo et al., (2015); Braikia et al., (2017)).

Thanks a lot for bringing our attention to these previous studies in this respect. Notably, we realized that Hou et al., (2008) in Figure 5A and Figure S4 similarly claims that the CTCF insertion specifically reduces the gene-enhancer interaction, but not the contacts with the other regions in the vicinity. We newly mentioned about their finding in this part.

The conclusion that the enhancer blocking activity and chromosome interaction activity are due to separate mechanisms is not sufficiently explained or supported. It is unclear why loop extrusion and enhancer blocking are introduced as separate mechanisms (Introduction), when the current understanding in the field is that enhancer blocking by CTCF sites is likely due to the creation of new TAD boundaries, which are formed by loop extrusion being blocked by CTCF sites (Recently reviewed in Schoenfelder and Fraser, (2019)).

Please see our response to the essential revision 6. We have tried to explain this better in the revised Introduction.

Similarly, there appears to be confusion about the various mechanisms of long-range chromatin interactions in the Discussion section. Subsection “Mechanism of the STITCH insulation and its control by heterochromatin induction”, second papragraph is very confusing. In the current work CTCF is removed, from the locus, but cohesin is not. Therefore, there is now unblocked loop extrusion throughout the locus. Given that the interaction between the enhancer and MYC is blocked by CTCF, it is much more likely that the increased interaction between the super-enhancer and MYC when CTCF is removed is driven by loop extrusion.

Depletion of cohesin enhances compartmentalization (Rao et al., 2017; Schwarzer et al., 2017). Therefore, we think it is less likely that the unblocked loop extrusion is required to establish the gene-enhancer interaction here. Rao et al., (2017) has shown that the links between super-enhancers are established even between different chromosomes, suggesting this can be achieved without any loop extrusion. We have mentioned this in the revised discussion. We think it is more likely that the loop extrusion interferes with the gene-enhancer interaction when CTCF is present (Figure 9), as explained in our response to the essential revision 6.

Technical details and biological conclusions are not adequately explained in the text throughout this manuscript. The manuscript would benefit from improving the explanations of why specific analyses are used, and what the results signify biologically, beyond just stating the observations. In particular, the explanations of PCA analysis, component loading plots, and power law scaling between gene expression and 4C-seq contact frequency should be clarified. As it is presented now it is entirely unclear what the PCA analysis really adds. Additional technical details of the computational methods used to analyze the sequencing data is also needed, preferably an online repository for the code used to generate the plots and run the statistical tests should be included with the manuscript.

Thanks a lot for the suggestion. We have tried to make the aims, the methodology, and the biological conclusions clearer throughout the revised manuscript. Particularly, for PCA, we now have added an extensive explanation of the methodology and clarified the story. We also put the R codes as supplementary files.

While 4C-seq is a useful technique for studying the specific interactions between MYC and surrounding loci, it would be beneficial to also compare this to an all-by-all or many-by-many chromosome conformation capture method such as Hi-C or 5C to show the endogenous organization of this region. Putting the STITCH insertion in the context of the landscape of genomic architecture (where are TADs or compartments found in this region?) would strengthen the manuscript and might help to understand how the different directionalities of the CTCF motifs in STITCH are working, as the current explanation is unclear.

Thanks for the comment. Now we have added Hi-C data and TADs organization in human ESCs (Dixon et al., 2015) to Figure 1B to illustrate how the domains are organized around the locus. We see that MYC is located inside a vast domain. Accordingly, we have introduced this data at the beginning of the Results section. Moreover, we newly performed 4C-seq from flanking regions of the insertion as viewpoints to see how STITCH affects the organization, as explained in our response to the essential revision 6. Also as explained in the response, we believe that we could provide a reasonable model of how STITCH insulates gene expression.

In addition, it would strengthen the manuscript to compare the contact frequency changes in the STITCH mutants at the MYC locus to changes in the endogenous TZ locus with similar modifications from previous publications by these authors, to determine how variable this behavior is at different genomic loci.

Thanks a lot for the suggestion. It is very true. As mentioned above, we have performed new 4C-seq experiments and analyzed the directionality of chromatin folding around the STITCH insertion, as was performed in our previous work (Tsujimura et al., 2018). We did similarly observe skewed transition of the folding directionality across inserted STITCH as in the endogenous TZ. This strongly suggests that the CTCF binding sites behave similarly in both endogenous and synthetic contexts. Interestingly, however, the divergence of folding directionality around STITCH was not evident in contrast to the TZ. As explained in our response to the essential revision 6, this may suggest that the creation of diverging directionality of chromatin folding is not an essential prerequisite for the insulation of gene-enhancer interaction.

Reviewer #2:

The role of CTCF, cohesin, TADs and 3D genome organization in regulating gene expression and enhancer-promoter (E-P) contacts is currently being intensely studied. In particular, whether CTCF sites and TADs really regulate E-P contacts and gene expression has recently become controversial, with some studies claiming that CTCF and TADs have no role in regulating gene expression (e.g. see https://www.biorxiv.org/content/10.1101/609941v1).

Most studies (including the preprint above) have taken a "deletion" approach to this issue: take a natural E-P pair and TAD and then go in and start deleting or inverting etc. CTCF binding sites and see how gene expression is affected. What is nice about this study is that they take the opposite approach – a kind of "addition" approach. They add the STICH array of CTCF binding sites to the MYC gene at different locations and see how contact frequency (4C) and gene expression (qPCR) of MYC in hiPSCs is affected. The 2 most important points in my opinion are:

1) CTCF sites really can block E-P contact and strongly (~20-fold) affect gene expression. At least for the MYC gene.

2) Histone modifications can be used to turn ON and OFF CTCF insulation with quite high temporal resolution.

I believe the authors get about as close to causality as is realistic with the current tools of molecular biology, which is nice.

Beyond some important but addressable concerns (poor writing, at times confusing figures and presentation, occasionally poor referencing, tool availability), my major concern is this: The authors report STICH as a tool. 2 key features in a tool that are desirable to have are: (1) robustness and (2) generality. But because the authors only apply STICH to one locus (MYC) in one cell type, we cannot really tell if STICH is likely to block E-P contacts in general and robustly in many other loci and other cell types. The impact of STICH would have been greatly increased if the authors could have applied it to 2-3 case studies, ideally in at least 2 different cell types.

So overall, I believe this is a nice contribution with some really important insights, but that the general interest and impact could have been substantially improved if the authors had applied STICH to at least 2-3 different systems and if they can improve their presentation.

Thanks a lot for this comment. Regarding the generality and robustness of STITCH, please see our response to essential revision 1.

Specific issues:

Writing: the paper is for the most part reasonably written, but there are at least >25 cases of poor English and/or syntax/grammar issues. This is too many for a reviewer to fix and I suggest that the authors go through and clean up these issues.

We apologize for our poor language. We have tried to correct mistakes and to improve readability.

Discuss results in context: First sentence of the Abstract and in the Introduction suggest that "regulation of gene-enhancer interaction is better understood,". I would argue that this is not true. In fact, recently several people (e.g. https://www.biorxiv.org/content/10.1101/609941v1) have begun arguing that CTCF plays essentially no role in the regulation of gene expression.

We agree with this comment. While recent studies have provided valuable insights into the genome regulation, there remain still many puzzling phenomena. We changed the sentence as "While regulation of gene-enhancer interaction is intensively studied,".

The fact that the authors see such clear results on MYC, in my opinion only increases the impact and value of this study. Therefore, it would be nice if the authors could discuss their MYC result a little bit more clearly in the Discussion section in the context of the many recent studies arguing that CTCF plays no or only a minor role in regulating E-P contacts and gene expression.

Thanks for this suggestion. We have now extended our Discussion section. For the details, please see our response to essential revision 6.

Key resources should be available: First of all, my apologies if the authors already did this. But I tried to find this information and was unable to. The authors must put the key STICH plasmids on AddGene for the community, since the value of a tool is largely derived from it being readily accessible for the community. The DNA sequences of STICH must also be available with the paper. I could not find the DNA sequences of the full STICH sequence nor could I find the sequences of the specific CTCF binding sites. These must be available.

Thanks a lot for this suggestion. We have now deposited the plasmids to AddGene. Please see our response to essential revision 7. The DNA sequences of STITCH were provided in Supplementary file 2. The sequences should also become available from AddGene soon.

RNA-seq Results: The RNA-Seq studies in Figure 2 were really nice. But I could not understand why the STICH +30kb cell line would have ~2-3x more deregulated genes than the del(30-440) cell line. Although STICH is powerful, deleting the enhancer should still have a stronger effect on expression than just blocking it. Could the authors better explain this?

Thanks for the comment. We speculate that the deletion allowed contact of MYC with regions with some enhancer activity located further than the +440kb position (Figure 1E), which led to a slight upregulation of MYC. We now discuss this in the corresponding part of the Results section.

PCA-analysis: In Figure 3A-B, the analysis of how 4C reads in the different regions depend on the STICH construct was really nice. It was also very interesting to see the highly non-linear scaling between 4C contact frequency and gene expression (Figure 3I-J). Both of these are really important contributions in my opinion.

Thanks a lot for the comment. We thought that the feasibility of serial mutagenesis of CTCF binding sites could be an advantage of our system, so we tried to gain as much insight as possible from this.

But the PCA-analysis was extremely confusing and convoluted. I really tried to follow the text and the figures, but it was very difficult for me to understand what the point was. Aorund lines 200-250, the authors spend a lot of text and a huge number of figure panels on this, but I really could not understand it. My suggestion would be to remove all the figures and text pertaining to PCA or at least radically simplify the figures and the text to make it easier to understand. What is the major biological insight coming from this PCA analysis? What is component loading?

We again apologize for our inadequate explanation. We have revised this part thoroughly. Please see our response to essential revision 2.

Subsection “Insulation and deletion of the enhancer resulted in similar transcriptome profiles”: authors out-of-the-blue reference VP-MYC1 and VP-MYC2, without any figure REF. I could not understand this.

Thanks for the comment. We now add the figure references.

tetR-KRAB studies: The Tet-R KRAB studies were very nice. I may have missed prior studies, but to my knowledge this is the first clear and causal demonstration that histone modifications can turn OFF CTCF insulation. However, one control I was missing was a DNA-binding control. TetR-KRAB binding could disrupt CTCF binding and insulation through 2 ways: DNA-binding competition (e.g. TetR-binding outcompetes CTCF binding) or KRAB-deposition of histone modifications. I would have liked to see a control showing that TetR binding alone – without the KRAB-domain – does not affect CTCF-mediated insulation.

But pretty neat to see that STICH insulation directly affects cell proliferation (subsection “Titrating blocking activity of STITCH by serial mutations of the CTCF binding sites”).

Thanks for the comments and suggestions. Please see our response to essential revision 3.

Figure 6. F and G have errors bars, but Figure 6 C, D, and E do not. Need to add errors bars to these.

Otherwise, the time-course results were also pretty cool.

Thanks for the comments. We only performed n=1 experiments for Figure 6C, D, E as the objective here was to capture the time-course change of the system. Besides, the experiments, particularly for Figure 6D and E, give relatively robust results. Based on these observations in Figure 6, we newly performed n=3 experiments for the statistical test presented in Figure 7A-F. We made the numbers of replicates explicit in the revised manuscript.

Reviewer #3:

The manuscript describes a strategy to modulate chromosomal contacts in the vicinity of the endogenous MYC gene in human iPS cells through the ectopic insertion of an array of CTCF sites. The approach (named STITCH) seems to be able to alter MYC transcription levels, which correlates with changes in interaction frequencies between the MYC locus and a super-enhancer region downstream. The authors further monitor chromatin states at the engineered locus upon the induction of H3K9 trimethylation by targeted recruitment of a KRAB domain at the STITCH cassette, which is shown to disrupt CTCF binding and restore wild-type chromosomal contacts. The authors conclude that CTCF-mediated modulation of chromosome interactions is the driver of transcriptional changes.

The study is interesting and well designed, and has the potential to bring insight into how gene expression could be modulated by manipulating chromosome structure. However, it suffers from several major drawbacks that should be thoroughly addressed.

Thanks a lot for appreciating our work.

1) Many of the native ChIP-seq experiments in the manuscript are difficult to interpret and it is often difficult to agree with the authors on the changes they describe. The zoom level in all Figures is way too low to visually appreciate any local changes at the MYC locus, the STITCH cassette and the neighboring region. More importantly, some crucial experiments (notably the H3K27ac and H3K27me3 ChIP-seq reported in Figure 1, Figure 4 and Figure 4—figure supplement 1) appear to be strongly suboptimal. It is hard to imagine that a local increase of ~2 counts in a 0-6 range as reported in Figure 1 and Figure 4 really does correspond to a specific enrichment as opposed to technical noise.

The authors should perform new H3K27ac/me3 ChIP-seq experiments, provide statistics to support the notion that changes in these chromatin marks can really be quantified and discuss their findings in the light of the new experiments. Crosslinking ChIP-seq would be a viable option in this context – in fact I found it quite unclear why native ChIP was required in this particular study.

Thanks for the comment. As the MYC locus was made haploid, the locus demanded more coverage than the rest of the genome, which might have made our ChIP data look as if compromised. However, we re-analyzed our data more quantitatively and now could provide statistical support for our conclusion (Figure 4C-E). Please see our response to essential revision 4 for more details. Also, we think our nChIP for H3K27me3 assays are good enough to discuss the following assays in Figure 5, Figure 6, Figure 7. In fact, we repeated H3K27me3 and H3K4me3 experiments in Figure 6F, G, and Figure 7A-F, and obtained the same conclusions. As indicated in Figure 4A, the background does not look very high when inspected in magnified views. It seems that the zoomed-out view of our tracks exaggerates the background levels.

ChIP with crosslinking in principle can enrich regions that are only indirectly associated with the chromatin mark through 3D association with regions that possess the mark genuinely, as exemplified in (Skene and Henikoff, 2017). Our work tries to challenge the chromatin organization with STITCH. Therefore, we think it should be more appropriate to utilize nChIP to discuss local chromatin states, as it avoids detecting pseudo-epigenetic changes due to the re-organization of chromatin conformation.

2) The correlation between MYC transcription levels and contact frequencies with the super-enhancer region (Figure 3) in mutant STITCH cell lines are interesting, and well supported by the large number of independent clones analyzed. Unfortunately, the structural changes induced by ectopic CTCF sites were not correlated with the position and orientation of endogenous CTCF sites. MYC itself is highly bound by CTCF, as can be more or less seen (again at regrettably low resolution) in Figure 5F. I would suggest the authors to thoroughly discuss the relationship between structural changes induced by STITCH and potential new loops induced by the ectopic CTCF sites in the light of the current understanding of CTCF orientation (and loop extrusion?). Reciprocal 4C viewpoints based on endogenous CTCF sites could also help clarifying this matter.

Thanks a lot for the suggestions. Please see our response to essential revision 6.

3) It is unclear how many copies of the STITCH cassette have been integrated at the MYC locus. The authors should provide evidence that a single insertion of 6 CTCF sites is actually responsible for the observed structural changes, as opposed to multiple tandem repeat insertions (especially since lipofection -and hence large amounts of DNA per cell- was used to generate the Cas9 assisted knock-in).

We agree that tandem repeat insertions could happen. We genotyped the insertion with PCR using several combinations of primers (Supplementary file 4). One of the primer pairs was designed to anneal to the flanking sites of the insertion, which tells the total length of the inserted DNA. With this, we could safely conclude that only one copy is integrated. Besides, we carry out Cre recombination after the targeting. We, of course, confirmed the correct integration at both sides of the insertion site. In this case, it was very improbable that we could obtain insertions with multiple copies afterward, because all the extra internal copies regardless of their orientations, which we confirmed are absent, should have been deleted out by the recombination.

4) The transcriptional analysis In Figure 2 could be improved. The cutoffs used to identify differentially expressed genes with DESeq2 are very loose (adjusted p-value = 0.1, no limitation is imposed on fold changes). In the absence of differential gene expression analysis on more than one STITCH and del(30-440) clones, it is difficult to assess what the >1000 genes detected as differentially expressed under these loose criteria actually represent. I would suggest to repeat the analysis including more than one clone per condition and using more stringent criteria (e.g. padj<0.01, |log2(FC)|>1) in order to identify mis-regulated genes more robustly and reliably. Also, a qPCR validation of significantly up- or down-regulated genes is missing.

Finally, there is no explanation for the fact that the effect on transcription in the deletion mutant is smaller than in the STICH mutant. If the changes are indeed due to the insulation of the super-enhancer region from the MYC gene, then deletion of the super-enhancer region should lead to an even stronger effect on transcription.

Thanks a lot for the suggestions. Please see our response to essential revision 3 for the reanalysis that we performed. The RNA-seq results are well consistent with our qPCR assays for the MYC expression. Therefore, we think our RNA-seq data are valid. Regarding the difference between the STITCH insertion and the enhancer deletion, we add our speculation to the revised manuscript. Please see our response to reviewer #2.

5) It would be nice to prove that transcriptional changes in the STITCH and del(30-440) lines are really caused by downregulation of MYC, which could be done notably by overexpressing MYC and testing if normal expression programs are rescued.

Thanks for the suggestion. We agree that this rescue experiment should clarify the cause of the transcriptomic changes more. However, our (re-)analysis shows that the most significantly down-regulated gene groups are those of known MYC target genes. Therefore, the transcriptomic alteration observed here should be reasonably attributable to the MYC down-regulation.

6) The PCA analysis is Figure 3 is in principle interesting and laudable as an attempt to quantify differences in 4C profiles in a quantitative and unbiased way. However, the text is somewhat obscure and panels 3C-H are difficult to interpret. It is unclear why the results shown in Figure 3E-H, where PCA is performed on a subset of the data, are so different from panel 3D. These differences are acknowledged in the main text but I did not understand how they are interpreted by the authors. I would actually suggest that the text relative to Figure 3 is entirely re-written and clarified (e.g. please explain what "component loading" means in this context). In addition, the PCA results should be integrated with a discussion of whether they correlate or not with the position and orientation of endogenous CTCF sites (see point 2 above).

Thanks a lot for the appreciation of our attempt and also for giving kind suggestions. We thoroughly revised our text and figures, as described in our response to essential revision 2. Regarding the relevance to the endogenous CTCF sites, please see our response to essential revision 6.

7) Transcriptional downregulation of MYC is attributed to changes in contact frequencies due to the presence of ectopic CTCF sequences at the STITCH cassette, which is supported by the strong correlation observed in Figure 3I. If this is really the case, and is due to CTCF looping from STITCH onto endogenous CTCF sites, then it should be possible to recapitulate the phenotype by deleting the endogenous partner CTCF sites. This would significantly strengthen the interpretation of the data.

Please see our response to essential revision 6.

8) in Figure 4C-D, negative and positive controls are missing (i.e. one or more regions where H3K4me3 should not be detected, and a region that is heavily bound by H3K27me, such as a poised gene, or a Hox gene). This is a very important control though, because one of the most interesting observations in the manuscript is that the transcriptional downregulation of MYC correlates with higher H3K27me3 levels. However, how much H3K27me3 is deposited? How does it compare with poised and/or inactive loci?

Thanks a lot for the comment and suggestion. We now have the suggested controls in the revised manuscript. Please see our response to essential revision 4.

9) In Figure 5, it is impossible to understand which changes are occurring at the MYC promoter in terms of H3K9me3 and CTCF levels. This is nonetheless crucial to interpret the gene expression changes upon Dox induction and how they are related to targeted recruitment of KRAB. It seems that the CTCF signal is decreased also in the MYC promoter in the absence of Dox, and not only at the STITCH region. A zoom-in and quantification of ChIP-seq experiments (peak calling, integrated intensities of signals) should be provided, and the results should be discussed accordingly.

Thanks for the comment. We now show a zoom-in view of these tracks, including MYC promoter, which clearly shows that H3K9me3 is not deposited around the promoter (Figure 5—figure supplement 1G). We also provide quantitative representations of CTCF ChIP-seq for this revision. The results show that striking changes of CTCF binding only took place at STITCH, but not in the other genomic regions, including those near the MYC promoter (Figure 5—figure supplement 1F).

10) Along the same line, in Figure 6 a crucial missing information is how CTCF binding evolves in time at the STITCH cassette and at the MYC locus.

We agree that this would be critical to understanding our STITCH system fully. Technically speaking, CTCF nChIP demands immediate processing and IP of chromatin after sampling, as freezing somehow abolishes most of the signals. Therefore, carrying out time-course experiments is quite difficult for our current situation. What we intend here is basically to describe the transition by the KRAB induction, which we think is well represented by our 4C-seq. We are, of course, very interested in what kind of events are serially involved in this process. However, we would like to leave this open for future works in the present study.

11) It is unclear what 'control' in Figure 7 refers to.

Thanks for the comment. As suggested by reviewer#1, we now indicate in the figure panels that they are either + or – DOX controls. They were kept in either + or – DOX for a much longer time than 24 or 72 hours (more than one passage) without switching.

12) One very interesting observation is that MYC gains H3K27me3 upon STITCH insertion, which correlates with the observed level of insulation in the various mutants and with transcriptional activation/deactivation in time course experiments. However why is it so? If this happens as a consequence of physical insulation from the super enhancer, how do the authors interpret it? An alternative explanation is that PRC2 is recruited by sequences in the STITCH cassette, and helps repressing transcription. The delay observed between MYC deactivation/reactivation and the corresponding differences of H3K27me3 are not large enough to exclude this second hypothesis. Based on the experiment shown in Figure 7J it cannot be excluded that H3K27me3 levels are unchanged upon treatment with EPZ, in the absence of a carefully quantified ChIP experiment.

Thanks for the comment. We have repeated our analysis presented in Figures 6F-G, 7A-F, and confirmed that our conclusion was reproduced, as described in our response to essential revision 4. Therefore, we think that our conclusions are valid. It has been already shown that H3K27me3 or PRC2 are not required to establish transcriptome in certain cellular contexts (Riising et al., 2014). This indicates that PRC2 is not necessarily required to repress gene expression. The delayed change of H3K27me3 was also reported before by a few studies (Hosogane et al., 2013). Our conclusions are consistent with these studies.

On the other hand, H3K27me3 levels seem to correlate with gene expression levels greatly not only at the MYC locus but also globally in the genome. We also wonder what this correlation would mean. Surely, as we discuss in the manuscript, PRC2 is critical for organismal development and homeostasis. Therefore, the protein complex should have essential roles. Perhaps the effects on transcriptome observed in PRC2 mutants so far might have been just an indirect effect of something more direct. However, we really cannot add speculations more. This needs to be studied in the future.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Summary:

This study describes development of a inducible insulator cassette, STITCH, that can act as boundary. This can be a very valuable tool to dissect rules of promoter – enhancer communication, and chromosome folding through mechanisms such as loop extrusion.

We request that the authors address the following issues.

Essential revisions:

1) The authors were requested to put the structural changes induced by STITCH in the context of the overall CTCF binding patterns in the MYC region. They addressed this point on the one hand by performing new 4C experiments using the STITCH sequence as a viewpoint. These experiments, now in Figure 1, unfortunately do not seem to reveal how the various endogenous CTCF sites could be used to make new connections with the ectopic STITCH cassette and even a re-analysis of the 4C data with the MYC promoter as a viewpoint are inconclusive. The authors conclude vaguely that "It might be extrapolated from these previous results that there are not very specific endogenous regions that singly form loops with STITCH to organize the conformational changes induced by STITCH". The interpretation of the data in the rebuttal is also highly speculative and does really not address the reviewers' request that structural changes are evaluated in the light of a more global view of chromosome contacts such as the one provided by Hi-C data. In 4C it is always hard to detect loop extrusion-associated structural features such as loops and stripes (or flares), and it is not surprising that specific connections between CTCF sites might be missed without performing matched Hi-C or 5C experiments. The authors need to at least these limitations of their 4C-based analyses.

Thanks for the comment. We agree that discussing based on the “extrapolation” from other studies had been more speculative than based on actual experiments in the manuscript of the previous version. Hence, we deleted the corresponding paragraph in which we had discussed the issue in relation with the studies of the Tfap2c locus (Tsujimura et al., 2018) and the Shh locus (Williamson et al., 2019). Instead, we added the following sentence to the present manuscript (subsection “Titrating blocking activity of STITCH by serial mutations of the CTCF binding sites”).

“Also, more comprehensive analysis methods such as 5C or Hi-C are required to fully describe the locus-wide conformational change induced by STITCH.”

We also added the following sentence in Discussion section to emphasize the limitation of our study.

“Moreover, applying 5C or Hi-C might be more appropriate to describe formation of contact domains than the present 4C-based analyses.”

2) In the new Figure 8, new experiments are provided to support the applicability of STITCH to additional contexts. However, the data do not appear to fully support the conclusion that STITCH-mediated modifications of chromosome interactions are at the basis of the observed transcriptional effects. First, 4C experiments in panel F do not allow to conclude in any manner that STITCH alters conformation at the targeted allele. Certainly, the presence of the wild-type allele confounds the readout, but even considering this, the fluctuating small-% differences observed do not seem to be robust (also, information on replicates is not provided unless I am mistaken). The right experiment to address this point would have been to perform 4C from the STITCH cassette itself, which would allow to detect mainly contacts within the mutant allele and which is apparently technically possible given that similar experiments are reported in the new version of Figure 1. It would be important to add such 4C analysis.

Thanks for the suggestion. Accordingly, we newly performed 4C-seq from a viewpoint at STITCH inserted in the NEUROG2 locus and compared the contact pattern between in the presence and absence of DOX (Figure 8—figure supplement 2D-G). The data well show that while the contact considerably extended to distant regions in the presence of DOX (i.e. with CTCF binding; see Figure 8E), the contact attenuated relatively in a short distance in the absence of DOX (Figure 8—figure supplement 2E, F). We think this change should reflect the extrusion mediated contact of the CTCF binding sites at STITCH. Thus, the new experiments could well support that the chromatin conformation is altered by the functionality of STITCH.

The viewpoint was designed at the right edge of the inserted cassette (Figure 8—figure supplement 2D). So, we also compared the directionality of chromatin folding. However, we could not detect striking changes by DOX (Figure 8—figure supplement 2E, G). We think this is because the viewpoint captures the contact pattern of not only the rightward CTCF binding sites but also the leftward ones, which are only 600-bp apart from the viewpoint. In the above analysis, indeed, the contact extension was observed on both right- and left-sides, indicating that the viewpoint captures contacts of both rightward and leftward CTCF arrays (Figure 8—figure supplement 2E, F).

We agree that the difference of the contact frequencies depicted in Figure 8F appears to be small. However, we would like to emphasize that the PCA plot in Figure 8—figure supplement 2C unbiasedly shows that the DOX alters the contact pattern exactly at the STITCH insertion site. Therefore, we think that it should be safe to attribute the conformational change to the functionality of STITCH.

Second, it appears that NEUROG2 downregulation is only shown for one population of cells following piggyBac-mediated insertion of the TetR-KRAB transgene. It is unclear whether these results would hold true if the transposition of the transgene is repeated in independent experiments. Given that piggyBac insertions typically occur in multiple genomic locations simultaneously in every cell, a possibility that cannot be excluded is that the transcriptional effect on NEUROG2 is a secondary effect of mis-regulation of one or more upstream genes that are accidentally targeted by the transposon. This should be discussed.

Thanks for the comment. Actually, we have split the cells in the same one dish, which had been transfected with the TetR-KRAB transgene in the piggyBac vector, equivalently to all the samples and replicates. Therefore, the composition/heterogeneity of the transgene insertions should be controlled well and equivalent between with and without DOX conditions. So, the possible secondary effects should also be equivalent. To clarify the procedures, we described the experiment as following (subsection “Blocking NEUROG2 activation in differentiating neural progenitor cells with STITCH”).

“We split the NEUROG2/KRAB cells derived from a single dish equivalently to different dishes, and then either did or did not add DOX upon the start of the differentiation into NPCs.”

3) Introduction and Discussion section: the authors seem to confuse several concepts of loop extrusion, insulation and contact domains. Insulation by CTCF is the result of its ability to block loop extrusion. Insulation does not necessarily involve CTCF sites to loop with each other (in fact such loops barely insulate).

Thanks for the comment. This is actually how we think. We simplified the Introduction to prevent potential confusion.

Insulation and contact domains can occur even when interactions are not strictly divergent at boundaries: insulation can occur in only one direction when CTCF sites are all in the same direction.

We agree. Our data are also very consistent to this statement. Moreover, we believe that the direct comparison of different CTCF configurations in our serial mutagenesis experiments could contribute to this understanding.

Such unidirectional insulation can demarcate contact domains.

We agree. This is for example described as exclusion domains by Sanborn et al., 2015. As far as we understand, however, emergence of such contact domains is not conclusively shown to be a prerequisite for the specification of enhancer allocation.

The authors are asked to consider these issues when revising the Introduction and Discussion section.

Thanks for the suggestions. Based on the above comments, we have tried to improve the Introduction and Discussion section.

4) All reviewers found the manuscript extremely difficult to read. Please do not use track changes. The manuscript should be carefully edited.

We have re-edited the manuscript carefully. This time, we upload the manuscript without the track changes.

5) All plasmids should be made available through AddGene.

Now all the three plasmids used in this study are already available through AddGene.

https://doi.org/10.7554/eLife.47980.sa2

Article and author information

Author details

  1. Taro Tsujimura

    1. Department of iPS Cell Research & Epigenetic Medicine, Keio University School of Medicine, Tokyo, Japan
    2. Department of Physiology, Keio University School of Medicine, Tokyo, Japan
    Present address
    Institute for the Advanced Study of Human Biology (WPI-ASHBi), Kyoto University, Kyoto, Japan
    Contribution
    Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Visualization, Methodology
    For correspondence
    taro.tsujimura@keio.jp
    Competing interests
    An inventor on the Japanese patent application (2018-154577, filed on 21 August 2018) and the PCT international patent application (PCT/JP2019/032106, filed on 16 August 2019) in respect of the STITCH/KRAB system by Keio University.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3281-0150

  2. Osamu Takase

    1. Department of iPS Cell Research & Epigenetic Medicine, Keio University School of Medicine, Tokyo, Japan
    2. Department of Physiology, Keio University School of Medicine, Tokyo, Japan
    Contribution
    Resources, Funding acquisition, Methodology
    Competing interests
    No competing interests declared

  3. Masahiro Yoshikawa

    1. Department of iPS Cell Research & Epigenetic Medicine, Keio University School of Medicine, Tokyo, Japan
    2. Department of Physiology, Keio University School of Medicine, Tokyo, Japan
    Contribution
    Resources, Funding acquisition, Methodology
    Competing interests
    No competing interests declared

  4. Etsuko Sano

    1. Department of iPS Cell Research & Epigenetic Medicine, Keio University School of Medicine, Tokyo, Japan
    2. Department of Physiology, Keio University School of Medicine, Tokyo, Japan
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared

  5. Matsuhiko Hayashi

    Apheresis and Dialysis Center, Keio University School of Medicine, Tokyo, Japan
    Contribution
    Project administration, MH made critical contribution to the decision of starting the project, and supported the set-up. MH have read the manuscript and approved it to be published
    Competing interests
    No competing interests declared

  6. Kazuto Hoshi

    1. Division of Tissue Engineering, University of Tokyo Hospital, Tokyo, Japan
    2. Department of Oral and Maxillofacial Surgery, University of Tokyo Hospital, Tokyo, Japan
    Contribution
    Resources, Project administration
    Competing interests
    No competing interests declared

  7. Tsuyoshi Takato

    1. Division of Tissue Engineering, University of Tokyo Hospital, Tokyo, Japan
    2. Department of Oral and Maxillofacial Surgery, University of Tokyo Hospital, Tokyo, Japan
    Contribution
    Resources, Project administration
    Competing interests
    No competing interests declared

  8. Atsushi Toyoda

    Department of Genomics and Evolutionary Biology, National Institute of Genetics, Mishima, Japan
    Contribution
    Data curation, Investigation, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0728-7548

  9. Hideyuki Okano

    Department of Physiology, Keio University School of Medicine, Tokyo, Japan
    Contribution
    Resources, Funding acquisition, Validation, Investigation, Methodology, Project administration
    Competing interests
    An inventor on the Japanese patent application (2018-154577, filed on 21 August 2018) and the PCT international patent application (PCT/JP2019/032106, filed on 16 August 2019) in respect of the STITCH/KRAB system by Keio University.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-7482-5935

  10. Keiichi Hishikawa

    1. Department of iPS Cell Research & Epigenetic Medicine, Keio University School of Medicine, Tokyo, Japan
    2. Department of Physiology, Keio University School of Medicine, Tokyo, Japan
    Contribution
    Conceptualization, Resources, Funding acquisition, Validation, Investigation, Methodology, Project administration
    For correspondence
    hishikawa-tky@umin.ac.jp
    Competing interests
    An inventor on the Japanese patent application (2018-154577, filed on 21 August 2018) and the PCT international patent application (PCT/JP2019/032106, filed on 16 August 2019) in respect of the STITCH/KRAB system by Keio University.

Funding

Japan Society for the Promotion of Science (Grants-in-Aid for Young Scientists (B) (17K16072))

  • Taro Tsujimura

Japan Society for the Promotion of Science (Grants-in-Aid for Scientific Research (B) (15H03001))

  • Keiichi Hishikawa

Japan Society for the Promotion of Science (Grants-in-Aid for Scientific Research (C) (16K09602))

  • Osamu Takase

Japan Society for the Promotion of Science (Grants-in-Aid for Scientific Research (C) (15K09244))

  • Masahiro Yoshikawa

Mutou Group

  • Hideyuki Okano

APA Group

  • Hideyuki Okano

IMS Group

  • Hideyuki Okano

Alba Lab

  • Hideyuki Okano

Kobe One Medicine, One Health

  • Hideyuki Okano

Japan Agency for Medical Research and Development (Acceleration Program for Intractable Disease Research Utilizing Disease-specific iPS Cells)

  • Hideyuki Okano

Keio University (Program for the Advancement of Research in Core Projects on Longevity of KGRI)

  • Hideyuki Okano

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Prof. Shinya Yamanaka (Kyoto University) for providing us the hiPSC line. We would also like to thank Drs. Sumihiro Maeda, Kent Imaizumi, Tsukasa Sanosaka, and Tomohiko Akiyama for their scientific advice and generous supports in revising the manuscript.

Senior Editor

  1. Jessica K Tyler, Weill Cornell Medicine, United States

Reviewing Editor

  1. Job Dekker, University of Massachusetts Medical School, United States

Reviewer

  1. Job Dekker, University of Massachusetts Medical School, United States

Publication history

  1. Received: April 26, 2019
  2. Accepted: April 6, 2020
  3. Version of Record published: May 5, 2020 (version 1)

Copyright

© 2020, Tsujimura et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 3,304
    Page views
  • 312
    Downloads
  • 3
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Taro Tsujimura
  2. Osamu Takase
  3. Masahiro Yoshikawa
  4. Etsuko Sano
  5. Matsuhiko Hayashi
  6. Kazuto Hoshi
  7. Tsuyoshi Takato
  8. Atsushi Toyoda
  9. Hideyuki Okano
  10. Keiichi Hishikawa
(2020)
Controlling gene activation by enhancers through a drug-inducible topological insulator
eLife 9:e47980.
https://doi.org/10.7554/eLife.47980

Categories and tags

Research organisms

Further reading

    1. Chromosomes and Gene Expression
    2. Genetics and Genomics

    Mouse B2 SINE elements function as IFN-inducible enhancers

    Isabella Horton, Conor J Kelly ... Edward B Chuong
    Research Article Updated

    Regulatory networks underlying innate immunity continually face selective pressures to adapt to new and evolving pathogens. Transposable elements (TEs) can affect immune gene expression as a source of inducible regulatory elements, but the significance of these elements in facilitating evolutionary diversification of innate immunity remains largely unexplored. Here, we investigated the mouse epigenomic response to type II interferon (IFN) signaling and discovered that elements from a subfamily of B2 SINE (B2_Mm2) contain STAT1 binding sites and function as IFN-inducible enhancers. CRISPR deletion experiments in mouse cells demonstrated that a B2_Mm2 element has been co-opted as an enhancer driving IFN-inducible expression of Dicer1. The rodent-specific B2 SINE family is highly abundant in the mouse genome and elements have been previously characterized to exhibit promoter, insulator, and non-coding RNA activity. Our work establishes a new role for B2 elements as inducible enhancer elements that influence mouse immunity, and exemplifies how lineage-specific TEs can facilitate evolutionary turnover and divergence of innate immune regulatory networks.

    1. Biochemistry and Chemical Biology
    2. Chromosomes and Gene Expression

    An acetylation-mediated chromatin switch governs H3K4 methylation read-write capability

    Kanishk Jain, Matthew R Marunde ... Brian D Strahl
    Short Report Updated

    In nucleosomes, histone N-terminal tails exist in dynamic equilibrium between free/accessible and collapsed/DNA-bound states. The latter state is expected to impact histone N-termini availability to the epigenetic machinery. Notably, H3 tail acetylation (e.g. K9ac, K14ac, K18ac) is linked to increased H3K4me3 engagement by the BPTF PHD finger, but it is unknown if this mechanism has a broader extension. Here, we show that H3 tail acetylation promotes nucleosomal accessibility to other H3K4 methyl readers, and importantly, extends to H3K4 writers, notably methyltransferase MLL1. This regulation is not observed on peptide substrates yet occurs on the cis H3 tail, as determined with fully-defined heterotypic nucleosomes. In vivo, H3 tail acetylation is directly and dynamically coupled with cis H3K4 methylation levels. Together, these observations reveal an acetylation ‘chromatin switch’ on the H3 tail that modulates read-write accessibility in nucleosomes and resolves the long-standing question of why H3K4me3 levels are coupled with H3 acetylation.

    1. Chromosomes and Gene Expression

    Hypermetabolism in mice carrying a near-complete human chromosome 21

    Dylan C Sarver, Cheng Xu ... G William Wong
    Research Article

    The consequences of aneuploidy have traditionally been studied in cell and animal models in which the extrachromosomal DNA is from the same species. Here, we explore a fundamental question concerning the impact of aneuploidy on systemic metabolism using a non-mosaic transchromosomic mouse model (TcMAC21) carrying a near-complete human chromosome 21. Independent of diets and housing temperatures, TcMAC21 mice consume more calories, are hyperactive and hypermetabolic, remain consistently lean and profoundly insulin sensitive, and have a higher body temperature. The hypermetabolism and elevated thermogenesis are likely due to a combination of increased activity level and sarcolipin overexpression in the skeletal muscle, resulting in futile sarco(endo)plasmic reticulum Ca2+ ATPase (SERCA) activity and energy dissipation. Mitochondrial respiration is also markedly increased in skeletal muscle to meet the high ATP demand created by the futile cycle and hyperactivity. This serendipitous discovery provides proof-of-concept that sarcolipin-mediated thermogenesis via uncoupling of the SERCA pump can be harnessed to promote energy expenditure and metabolic health.