Transcriptional Dysregulation of MYC Reveals Common Enhancer-Docking Mechanism

1Elevated expression of the c-MYC transcription factor occurs in a broad spectrum of human cancers and is associated with tumor aggression and poor clinical outcome. Activation of the MYC gene, which encodes c-MYC, is a hallmark of cancer initiation and maintenance. Dysregulation of MYC is often achieved through the formation of large tumor-specific super-enhancers in the region surrounding the MYC gene. These large enhancer clusters differ in size, composition, and distance from the MYC promoter, yet all accomplish the same task of stimulating MYC overexpression across a broad spectrum of tumors.

2Selective gene activation is essential to the gene expression programs that define both normal and cancer cells. During gene activation, transcription factors (TFs) bind enhancer elements and regulate transcription from the promoters of nearby or distant genes through physical contacts that involve looping of DNA between enhancers and promoters. The mechanisms that ensure that specific enhancers interact with specific promoters are not fully understood. Some enhancer-promoter interactions are likely determined by the nature of TFs bound at the two sites.

3Recent studies have revealed that specific chromosome structures play important roles in gene control. Enhancer-promoter interactions generally occur within larger chromosomal loop structures formed by the interaction of CTCF proteins bound to each of the loop anchors. These loop structures, variously called topologically associated domains (TADs), sub-TADs, loop domains, CTCF contact domains, and insulated neighborhoods, tend to insulate enhancers and genes within the CTCF-CTCF loops from elements outside those loops. Constraining DNA interactions within CTCF-CTCF loop structures in this manner may facilitate proper enhancer-promoter contacts.

4CTCF does not generally occupy enhancer and promoter elements. Another TF, YY1, generally binds to enhancers and promoters and facilitates their interaction through YY1 dimerization. However, when CTCF does bind these regulatory elements, it can also contribute to enhancer-promoter interactions.

5Here, we investigate DNA looping structures in the MYC locus in multiple cancers and identify a CTCF-occupied site at the MYC promoter that facilitates docking with essentially any enhancers that are formed within the 2.8 Mb MYC locus. The CTCF-occupied site at the MYC promoter, which we call the MYC enhancer-docking site, can be abrogated by genetic and epigenetic editing. Similar enhancer-docking sites occur at other oncogenes. This suggests a mechanism by which tumor cells can generally hijack enhancers and, with editing, a potential therapeutic vulnerability.

6Previous studies have established that tumor cells acquire tumor-specific super-enhancers at various sites throughout the MYC locus (Figures 1A and S1A), but the mechanisms by which these diverse enhancer structures control MYC are not clear. In one case, for example, a super-enhancer located ~2 Mb downstream of the MYC gene has been shown to physically interact with MYC, but the mechanisms responsible for this specific interaction are unclear. To gain insights into the potential role of DNA loop structures in gene control at the MYC locus, we generated cohesin HiChIP data for HCT-116 cells and collected published DNA interaction data for three other cancer cell types for comparison (Figure 1B; Tables S1 and S5). Among the DNA loop structures observed in these datasets, a large 2.8 Mb DNA loop was evident in all four cell types. This loop connects CTCF sites encompassing the MYC gene and qualifies as an insulated neighborhood. The DNA anchor sites of this 2.8 Mb DNA loop occur at the boundaries of a TAD found in all cells (Figure S1B). The MYC TAD encompasses a region previously described as a “gene desert,” because this large span of DNA contains no other annotated protein-coding genes.

7While all cells examined appear to share the TAD-spanning 2.8 Mb loop encompassing MYC, the loop structures within the neighborhood were found to be markedly different among the tumor types. The internal loops were dominated by interactions between a MYC promoter-proximal CTCF site and the diverse super-enhancers (Figures 1B and 1C). The major differences between these internal structures in the different tumor cells involved the different positions of the tumor-specific super-enhancer elements. Examination of Hi-C data for a broader spectrum of tumor cell types suggests that tumor cells generally have DNA contacts between the MYC promoter-proximal site and other sites within the 2.8 Mb MYC TAD (Figure S1B). This looping was not limited to cancer cells, because examination of enhancer and promoter-capture Hi-C data in a variety of normal cell types that express MYC revealed that cell-type-specific enhancers do indeed loop to the MYC proximal CTCF site (Figures S1C and S1D). This indicated that this CTCF site is also used during normal development by cell-type-specific enhancers to facilitate MYC expression and cellular proliferation.

8Further examination of the MYC promoter-proximal region revealed three constitutive CTCF binding sites (Figure 1C). All three sites were found to be occupied by CTCF in a wide variety of normal cells and tumor cells, and this binding pattern is shared across species (Figure S1C). Previous studies have examined the role of CTCF binding at all three sites. The two sites located within the MYC gene have been shown to play roles in MYC transcript start site selection and in promoter-proximal pausing of RNA polymerase II. The CTCF binding site located 2 kb upstream of the major transcription start site has been reported to protect the promoter from methylation and to be an insulator element. The DNA interaction data described here, however, suggests that this upstream site dominates connections with distal enhancer elements, as the majority of reads in the DNA interaction data are associated with this site in all tumor cells examined (Figures 1C and S1E). The −2 kb CTCF binding site contains a number of putative CTCF binding motifs; one of these most closely matches the canonical CTCF motif in the JASPAR database and occurs within a highly conserved sequence (Figure 1D). These features, the presence of CTCF sites in tumor super-enhancers, and the ability of two CTCF-bound sites to be brought together through CTCF homodimerization led us to further study the possibility that the −2 kb site has an enhancer-docking function critical to MYC expression.

9To determine whether the putative enhancer-docking site plays a functional role in MYC expression through DNA loop formation, small perturbations of the CTCF binding site were generated in both alleles of the tumor cell lines K562, HCT-116, Jurkat, and MCF7 using clustered regularly interspaced short palindromic repeats (CRISPR)/Cas9 (Figures 2A and 2B). Attempts at genetic perturbation by transfection with constructs carrying CRISPR/Cas9 with a guide RNA specifically targeting the CTCF motif upstream of the MYC gene did not yield viable clones. To allow cells to continue to proliferate if the CTCF motif deletion was lethal, cells were virally transduced with an exogenous MYC gene driven by a phosphoglycerate kinase (PGK) promoter (Figure S2A). This construct contained sequence differences in the 3′ UTR that allowed discrimination between the endogenous and exogenous MYC mRNAs. Cells expressing this exogenous MYC gene were then subjected to CRISPR/Cas9 perturbation. Clones were selected with small deletions or insertions disrupting the canonical CTCF motif (Figures 2B and S2B) and these cells were further characterized. CTCF chromatin immunoprecipitation quantative polymerase chain reaction (ChIP-qPCR) showed complete loss of CTCF binding to this site in K562 and HCT-116 cells and a 60%–70% reduction in CTCF binding at this site in Jurkat and MCF7 cells (Figure 2C). RNA analysis revealed a 70%–80% reduction of endogenous MYC mRNA in the absence of the enhancer-docking site in all of these cell types (Figure 2D). Furthermore, an inducible CRISPR/Cas9 perturbation model showed reduced proliferation for these four cell types upon induction of CTCF-site deletions (Figures S2C–S2G). These results indicate that the CTCF motif in the MYC enhancer-docking site is necessary for CTCF binding, for high levels of MYC expression and for cellular proliferation.

10If the putative MYC enhancer-docking site contributes to looping interactions with distal enhancers, then the loss of this site should cause a decrease in DNA interactions between the MYC promoter and the distal super-enhancers. We used chromosome conformation capture combined with high-throughput sequencing (4C-seq) to compare super-enhancer interactions in K562 and HCT-116 cells with normal or perturbed CTCF binding motifs. The 4C-seq data in K562 cells indicated that the MYC enhancer-docking site interacts predominantly with distal super-enhancers located ~0.3 Mb and ~2 Mb downstream of the MYC gene and that the majority of these interactions were significantly reduced when the putative CTCF motif was perturbed (Figures 3A and S3A). In order to control for any direct effects of a genetic alteration near the viewpoint, 4C-seq was performed with a viewpoint placed in the downstream super-enhancer. This showed clear interactions with the MYC enhancer-docking site as well as with the nearby super-enhancer, and these interactions were significantly reduced upon perturbation of the CTCF motif (Figures 3B and S3B). Similar results were obtained in HCT-116 cells, where the viewpoint was centered on the super-enhancer located ~0.4 Mb upstream of the MYC gene (Figure S3C). These results showed that the CTCF site in the promoter-proximal region of MYC is important for optimal interaction with distal enhancers and supports the idea that this CTCF site functions as an enhancer-docking site.

11CTCF binding is abrogated when its sequence motif is methylated, and the MYC enhancer-docking site occurs within a CpG island that is consistently hypomethylated in different tumor types as well as in different normal tissues (Figures S4A and S4B). The recent development of tools that permit site-specific DNA methylation suggested a means to disrupt MYC expression by methylation of the enhancer-docking site. To achieve targeted methylation, we created a construct to express a dCas9 fusion protein consisting of the catalytic domain of DNMT3A and the interacting domain of DMNT3L. This dCas9-DNMT3A-3L protein was targeted to the MYC enhancer-docking site in K562 and HCT-116 cells using multiple guide RNAs that span the region (Figures 4A and 4B). The targeting of dCas9-DNMT3A-3L resulted in robust local DNA methylation (Figure 4C) and a 40%–50% reduction in mRNA levels in both cell types (Figure 4D). The methylated region likely contains binding sites for additional TFs that may be sensitive to DNA methylation, so it is possible that the reduced mRNA levels are due to multiple factors. In order to test whether disruption of TFs other than CTCF contribute to the reduction in MYC mRNA levels, the dCas9-DNMT3A-3L was targeted to the MYC enhancer-docking site in CTCF-site deleted K562 cells. No further reduction of MYC mRNA levels was observed under these conditions (Figures S4C and S4D), indicating that loss of CTCF was a major contributor to the observed reduction of MYC expression upon targeted methylation of the MYC enhancer-docking site. These results demonstrate that epigenetic editing of the enhancer-docking site can reduce MYC expression.

12Previous genomic studies have noted that CTCF might engender enhancer-promoter interactions at a minority of genes. We therefore identified the set of genes whose promoter-proximal regions contain CTCF-bound sites and that show evidence of enhancer interactions in K562, Jurkat, and HCT-116 cells. We identified all active transcription start sites (TSSs) that have at least one CTCF-bound site within 2.5 kb of the TSS that interacts with at least one enhancer. This yielded between 555 and 1,108 TSSs with a nearby CTCF site that loops to an active enhancer (Figure 5A; Table S2). We define these TSSs as having a putative CTCF enhancer-docking site. The majority of TSSs identified in this analysis were identified in only one cell type, with only 52 TSSs identified in all three cell types (Figure 5B). Nonetheless, these putative enhancer-docking sites tended to be constitutively bound by CTCF in all three cell types, and the CTCF motifs in these sites showed high sequence conservation (Figures 5C and 5D). This suggests that these putative enhancer-docking sites are occupied by CTCF regardless of interaction with active enhancers and that differences in cell-type-specific enhancers are largely responsible for differential use of enhancer-docking site genes in these cells.

13Gene ontology analysis of the genes with putative enhancer-docking sites found different processes to be significantly enriched in each cell type, and these processes were dominated by the cellular identity of the cell lines (Figure S5A; Table S3). Common processes among the three cell types include cell cycle and other cancer-related processes such as gene expression and response to signaling (Figure S5A). A number of cancer-associated genes were found, including TGIF1, VEGFA, RUNX1, and PIM1 (Figure 5E), as well as others (Figure S5B). We conclude that genes other than MYC are likely regulated by CTCF-bound enhancer-docking sites and that these include multiple cancer-associated genes.

14Aberrant transcriptional activation of the MYC oncogene occurs frequently in tumor cells and is associated with tumor aggression. MYC resides within a 2.8 Mb TAD and its aberrant activation is generally accomplished by acquisition of a super-enhancer somewhere within that TAD. How these diverse cancer-specific super-enhancers loop long distances to specifically interact with MYC has not been clear. We find that the diverse super-enhancers commonly interact with, and depend on, a conserved CTCF binding site located 2 kb upstream of the MYC promoter. Because tumor super-enhancers can encompass genomic regions as large as 200 kb, and CTCF occupies sites that occur on average every 10 kb, there is considerable opportunity for super-enhancers to adventitiously contain a CTCF-bound site, which in turn could serve to interact with the MYC CTCF site (Table S6). Thus, different tumor super-enhancers have the opportunity to form through diverse mechanisms throughout this large TAD and can exploit the MYC CTCF site to interact with and activate MYC expression.

15The concept that enhancer-promoter interactions generally occur within larger chromosomal loop structures such as TADs, which are themselves often formed by the interaction of CTCF proteins bound to each of the TAD loop anchors, is supported by the observations described here. These larger loop structures tend to insulate enhancers and genes within the CTCF-CTCF loops from elements outside those loops. Constraining DNA interactions within CTCF-CTCF loop structures in this manner may facilitate proper enhancer-promoter contacts.

16The evidence described here argues that diverse human tumor cell super-enhancers depend on the MYC CTCF site for optimal levels of enhancer-promoter looping and mRNA expression. A recent independent study in K562 cells used a tiling CRISPR screen to systematically perturb the MYC locus and also found that full MYC expression and cell proliferation is dependent on this region. However, deletion of the −2 kb CTCF site has limited effects on MYC expression in mice, and some translocated enhancers can drive MYC expression in the absence of this CTCF site. There are several potential explanations for these diverse results. It is possible that the −2 kb CTCF site is important for optimal MYC expression levels in human cells, but not in mice. It is conceivable that the deletion of a region containing the CTCF site can be compensated by features of the new enhancer landscape in the deletion mutations. Furthermore, additional mechanisms normally involved in enhancer-promoter interactions, such as YY1-YY1 interactions, may mask the loss of the CTCF site in vivo; YY1 is present in the MYC promoter region and is thus likely to contribute to DNA looping and expression.

17Our studies suggest that an additional set of human genes, beyond MYC, may utilize promoter-proximal enhancer-docking sites to mediate cell-type-specific enhancer-promoter interactions. Such CTCF-mediated enhancer-promoter interactions are generally nested within larger CTCF-mediated loops that would function as insulated neighborhoods. At these genes with CTCF-mediated enhancer docking, the promoter-proximal enhancer-docking sites tend to be constitutively bound by CTCF and these binding sites tend to be highly conserved. Indeed, two studies have reported that these genes tend to lose expression upon perturbation of CTCF, consistent with a role for CTCF in enhancer-promoter looping. Among these genes are cancer-associated genes that likely employ this mechanism to engender interactions with tumor-specific enhancers. For example, at CSNK1A1, a drug target in acute myeloid leukemia (AML) tumor cells, VEGFA, which is upregulated in many cancers, and RUNX1, a well-defined oncogene in AML, the evidence suggests that super-enhancers in these cancer cells use a CTCF enhancer-docking mechanism to interact with the oncogene. Thus, a CTCF-dependent enhancer-docking mechanism, which presumably facilitates interaction with different cell-specific enhancers during development, is exploited by cancer cells to dysregulate expression of prominent oncogenes.

18MYC dysregulation is a hallmark of cancer. The c-MYC TF is an attractive target for cancer therapy because of the role that excessive c-MYC levels play in a broad spectrum of aggressive cancers, but direct pharmacologic inhibition of c-MYC remains an elusive challenge in drug discovery. The MYC enhancer-docking site, and presumably those of other oncogenes, can be repressed by dCas9-DNMT-mediated DNA methylation. Oncogene enhancer-docking sites may thus represent a vulnerability in multiple human cancers.

19Further details and an outline for resources used in this work can be found in Supplemental Experimental Procedures.

20Genome editing was performed using CRISPR/Cas9 essentially as described previously. The genomic sequences complementary to all guide RNAs are listed in Table S4.

21ChIP was performed as described previously. Approximately 30 million cells were crosslinked for 10 min at room temperature by the addition of one-tenth of the volume of 11% formaldehyde solution to the growth media followed by 5 min quenching with 125 mM glycine. Cells were washed twice with PBS, and then the supernatant was aspirated and the cell pellet was flash frozen at −80°C. 100 µL Protein G Dynabeads (Thermo 10003D) were blocked with 0.5%BSA (w/v) in PBS. Magnetic beads were bound with 40 µL anti-CTCF antibody (Millipore 07–729). Nuclei were isolated as previously described and sonicated in lysis buffer on a Misonix 3000 sonicator for 5 cycles at 30 s each on ice (18–21 W) with 60 s on ice between cycles. Sonicated lysates were cleared once by centrifugation and incubated overnight at 4°C with magnetic beads bound with antibody to enrich for DNA fragments bound by the indicated factor. Beads were washed with wash buffers A, B, C, and D sequentially. DNA was eluted, cross-links were reversed, and DNA was purified with phenol chloroform extraction and ethanol precipitation. Libraries for Illumina sequencing were prepared following the Illumina TruSeq DNA Sample Preparation v2 kit and sequenced on the Illumina HiSeq 2500 for 40 bases in single-read mode.

22A modified version of 4C-seq was developed (Supplemental Experimental Procedures). The major change was the ligation is performed in intact nuclei (in situ). This change was incorporated because previous work has noted that in situ ligation dramatically decreases the rate of chimeric ligations and background interactions.

23HiChIP was performed essentially as described. 10 million HCT116 cells were crosslinked for 10 min at room temperature by the addition of one-tenth of the volume of 11% formaldehyde solution to the growth media followed by 5-min quenching with glycine. Cells were washed twice with PBS, and then the cell pellet was flash frozen in liquid nitrogen. Frozen samples were processed according to protocol (Supplemental Experimental Procedures).

24To perform targeted methylation, cells were transfected with a dCas9-DNMT3A-3L construct and five guides. To generate the dCas9-DNMT3A-3L construct, dCas9 was isolated from pSQL1658 (Addgene 51023) by PCR. Cas9 was removed from pX330-Cas9 (Addgene 42230) and replaced by DNMT3A-3L. Guide RNAs used for targeting can be found in Table S4.

25ChIP-seq datasets were generated for this study as well as collated from previous studies (Table S5) and were aligned using Bowtie (version 0.12.2) to the human genome (build hg19, GRCh37) with parameter -k 1 -m 1 -n 2. We used MACS version 1.4.2 with the parameter “-no-model-keep-dup=auto.” A p value threshold of enrichment of 1e-09 was used.

264C-seq reads were trimmed and mapped using bowtie with options -k 1 -m 1 against the hg19 genome assembly. We only used the reads from non-blind fragments for further analysis. The normalized profile of each sample was smoothened using a 6-kb running mean at 500-bp steps across the genome. Quantification of the 4C signal counted the reads per fragment per million sequenced reads in the super-enhancers or the CTCF MACS peak calls.

27We developed a new software pipeline and analytical method called origami to process HiChIP and chromatin interaction analysis by paired-end tag sequencing (ChIA-PET). The software and releases can be found at https://github.com/younglab/origami using version alpha20160828. The ChIA-PET datasets analyzed along with their corresponding linker sequence and called interactions in and around the MYC TAD can be found in Table S4. Each ChIA-PET datasets was processed as follows: the reads were first trimmed and aligned using origami-alignment. Each end of a paired end tag (PET) with a linker sequence were separately mapped to the hg19 genome assembly using bowtie with the following options: -v 1 -k 1 -m 1. After alignment, the separated PETs were re-paired in the final BAM output. After repairing, all duplicated PETs within the data were removed. Peaks were called on the re-paired ChIA-PET reads using MACS1 v1.4.2 with the following parameters: -nolambda -nomodel -p 1e–9.