Writing of H3K4Me3 overcomes epigenetic silencing in a sustained but context-dependent manner

Introduction

Epigenetics controls gene expression patterns in a cell-specific and mitotically stable manner. Genome-wide analysis and gene expression profiling studies identified specific combinations of modifications of DNA and histones, as well as transcriptional regulators, to correlate with chromatin accessibility and expression1,2,3. For example, methylation of lysine residue 4 or 79 on histone H3 (H3K4me3 and H3K79me2–3) or monoubiquitination of histone H2B (H2Bub1) situated at transcription start sites (TSSs) are associated with transcriptionally active euchromatin4,5,6,7,8,9,10. On the other hand, DNA methylation in core promoter regions is mainly involved in gene silencing. Elucidating the distinction between the mere associative presence versus the actual causality of transcription by chromatin marks is an important area of investigation11,12. Current approaches to studying chromatin function often make use of small-molecule inhibitors and RNA interference to unravel the role of epigenetic enzymes in transcription regulation. Although these studies have yielded basic insights into epigenetic regulation, they are hampered by genome-wide effects13,14. Identifying the conditions that drive transcriptional changes is critical to understanding how cell identity is established and how genes become permanently dysregulated in human diseases.

An innovative approach to study transcriptional changes is by synthetic modulation of gene expression. Gene expression modulation can be achieved using artificial transcription factors by coupling transcriptional activators or repressors to DNA-targeting platforms such as zinc-finger (ZF) domains, transcription activator-like effector domains and the clustered regularly interspaced palindromic repeats (CRISPR–dCas)15,16,17,18,19,20,21,22,23,24,25,26,27,28,29. Even though changes in gene expression have been successful, the sustainability of such induced transcriptional reprogramming is still under debate. Indeed, these artificial systems merely act as scaffolds to recruit multiple transcriptional components and have no enzymatic activity on the chromatin state directly. Therefore, methods for directly linking transcriptional function with the presence or absence of epigenetic marks are needed to establish general principles for (sustained) cell reprogramming. One elegant method to establish those general rules is epigenome editing30,31,32,33. Since the dynamic remodelling of the chromatin landscape is tightly regulated by a conglomerate of enzymes and macromolecules, there is an extensive array of epigenetic effector domains suitable for gene expression modulation34,35,36.

Several studies have already shown the potency of epigenome editing in inducing37,38,39,40,41,42 or repressing41,42,43,44,45,46,47,48,49 gene expression. Despite the fact that gene expression could be modulated, little is known about the stability of the acquired epigenetic states. In this respect, gene repression by DNA methylation has been shown to be stable and heritable using engineered ZFs fused to DNA methyltransferases to target the SOX2 gene46, while the repressive effect was not achieved in another context for the VEGF-A gene49. In contrast, sustained gene reactivation remains largely unexplored. Studies so far have been focusing on activating gene expression by VP64-based artificial transcription factors and, in a few cases, epigenetic enzymes (TET and p300)37,38,39. Although gene induction has been achieved, the sustainability of this overexpression has not been documented yet. To fully exploit the potentials of epigenome editing, it is necessary to understand how the chromatin microenvironment affects mitotic stability of reprogrammed gene expression patterns.

Trimethylation of H3K4 is a hallmark of gene expression and the presence of this mark at promoters of protein coding genes might serve as a transcriptional on-off switch50. H3K4me3 is found in ∼75% of all human active gene promoters in several cell types, suggesting that it plays a key role in mammalian gene expression2. Here we have employed epigenome editing to investigate the role and stability of H3K4me3 in transcriptional activation. In light of the fact that H3K4me3 at TSSs is frequently associated with active transcription, we aimed to achieve targeted gene re-expression of epigenetically silenced genes by local induction of this mark. Using the histone methyltransferase PRDM9 (refs 51, 52, 53) fused to either dCas9 or ZF proteins, we examined the role of H3K4me3 in upregulating the expression of several model genes in different chromatin contexts. We also identified potential reinforcing marks to achieve stable gene activation. As such, our study identified H3K4me3 and H3K79me as well as the absence of DNA methylation to be critical in allowing sustained re-expression of epigenetically silenced genes.

Results

Effect on gene expression by PRDM9 induced H3K4me3

To investigate the potency of H3K4me3 in inducing gene expression, we fused the SET domain of the human PRDM9 to dCas9. We transiently co-transfected HEK293T and A549 cells to express the proteins (dCas9-empty, the transcriptional activator dCas9-VP64 and dCas9-PRDM9) with a combination of guide RNAs (gRNAs) to activate the endogenous promoters of intercellular adhesion molecule 1 (ICAM1), Ras association domain-containing protein 1 (RASSF1a) or epithelial cell adhesion molecule (EpCAM; Fig. 1a,b). We used a combination of gRNAs to target each promoter based on previous reports indicating that multiple gRNAs at a single promoter are more effective for gene activation15,16,18,19,21,22. dCas9-VP64 was able to induce EpCAM gene expression in both cell lines, whereas gRNA-directed dCas-PRDM9 and dCas-VP64 were ineffective in activating ICAM1 and RASSF1a (Fig. 1c). There were no clear beneficial effects when changing the binding orientation of the gRNAs. Analysis of ENCODE data depicted that the target regions of ICAM1 and RASSF1a in both HEK293T and A549 were hypermethylated, not associated with H3K4me3 marks and lacked DNAse hypersensitive sites, whereas the promoters of EpCAM were unmethylated and contained H3K4me3 peaks (Supplementary Fig. 1a). We confirmed these differential DNA methylation levels of our three model genes around the promoter area in both cell lines (Supplementary Fig. 1b). This suggested that the dCas9 is not able to access the promoters of hypermethylated genes, explaining the lack of effect of VP64 and the PRDM9 catalytic domain. To further confirm this, we choose to target procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2 (PLOD2) in C33a cells, which is transcriptionally repressed, although its promoter has low DNA methylation levels (Supplementary Fig. 1b). Indeed, dCas9-VP64 was able to induce high levels of gene transcription from the endogenous PLOD2 promoter (≈2,000-fold) (Fig. 1d). In addition, dCas9-PRDM9 was able to moderately but significantly upregulate PLOD2 expression up to 1.7-fold compared with its catalytically inactive mutant (MutPRDM9; P<0.05, two-tailed unpairedt-test), which did not change the target gene expression. To test whether the actual binding of dCas9 was indeed impaired by DNA hypermethylation, we performed anti-FLAG chromatin immunoprecipitation (ChIP) using cells transfected with dCas9-3XFLAG and a combination of gRNAs for the different cell lines and genes (Supplementary Fig. 1c). dCas9 is not able to efficiently bind regions located in CpG islands (CGIs), where DNA hypermethylation is present (ICAM1 and RASSF1a) as compared with regions outside of CGIs or without DNA hypermethylation (EpCAM and PLOD2). Taken together, our data suggests that DNA hypermethylation of CGIs severely hampers the binding or effect of dCas9 fusions. Importantly, for a susceptible silenced locus, H3K4me3 could directly induce gene expression.

Figure 1: Local induction of H3K4me3 activates transcription of endogenous genes from promoter regions using CRISPR–dCas9.

(a) Schematic representation of the targeted genes and the (overlapping) locations where the ZFs and gRNAs bind (the letter or number of each region refers to the name of the ZF or gRNA, for the regions marked with * a ZF as well as a gRNA were designed). The yellow bars represent the location of the CpG islands. (b) Schematic of dCas9-VP64 targeting sense and antisense strands of DNA, and dCas9 and ZF fused to the epigenetic editor PRDM9 to locally induce H3K4me3. (c) Relative messenger RNA (mRNA) expression of ICAM1, RASSF1a andEpCAM in HEK293T and A549 cells, and (d) PLOD2 in C33a cells, by the indicated dCas9 fusion protein co-transfected with a combination gRNAs targeted to each promoter region. (n=3 independent experiments; error bars±s.d.).