Ezh2-dCas9 and KRAB-dCas9 enable engineering of epigenetic memory in a context-dependent manner

Plasmids

Plasmids expressing dCas9 fusions with KRAB, Ezh2, DNMT3A effector domains, as well as the dCas9 cloning vector without any effector domain, have previously been described [26] and are available through Addgene (KRAB-dCas9 #112195 Ezh2-dCas9 #100086, dCas9 #100091). The DNMT3L expression plasmid pCDNA-DNMT3L was a kind gift from Dr. Fred Chedin [28]. DNMT3A-dCas9 (Addgene #100090) and pCDNA-DNMT3L are abbreviated in this study to D3A-dCas9 and D3L, respectively. DNMT3L was amplified from pCDNA-DNMT3L with overhangs for Gibson cloning into a KpnI- or NheI-digested dCas9 cloning vector, resulting in D3L-dCas9 (Addgene) and dCas9-D3L (Addgene), respectively. Protein sequences are shown in Additional file 1: Figure S1. A plasmid expressing a hybrid of mouse Dnmt3a and human DNMT3L fused to dCas9 (dCas9-D3a3L) was a kind gift from Dr. Albert Jeltsch [29]. The gRNA cloning vector, plasmid # 41824, was obtained from Addgene [30], and the 19-bp gRNA target sequences were selected within 500 bp of each relevant gene promoter using the online tool CHOPCHOP v2 [31]. Each gRNA sequence was cloned as G-N19 into the AflII-linearized plasmid, as previously described [26]. The gRNA sequences used to create target-specific vectors are listed in Additional file 2: Table S1. Plasmids expressing 1xMCP-effector fusions were generated by cloning effector domains into NcoI- and ClaI-digested UBC-NLS-HA-2XMCP-tagRFPt [32] Addgene plasmid #64541). 2xMCP-effector fusion plasmids were created using XbaI- and ClaI-digested UBC-NLS-HA-2XMCP-tagRFPt. KRAB and Ezh2 effector domains were amplified with overhangs from KRAB-dCas9 and Ezh2-dCas9 and cloned using the Gibson method into the appropriately cut vector. Protein sequences for 2xMCP-KRAB (Addgene) and 2xMCP-Ezh2 (Addgene) are shown in Additional file 1: Figure S1. MS2-gRNA expressing plasmids were created using the sgRNA(MS2) backbone [33], Addgene plasmid #61427) following the authors’ instructions (http://sam.genome-engineering.org/protocols/). MS2-gRNA sequences are listed in Additional file 2: Table S1.

Cell culture and transfections

HCT116 cells (ATCC #CCL-247) were maintained in McCoy’s 5A medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. LNCaP cells (ATCC #CRL-1740) were grown in RPMI-1640 medium supplemented with 10% FBS and 1% penicillin/streptomycin. 3T3 cells were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% FBS and 1% penicillin/streptomycin. Neuro2A (ATCC #CCL-131) cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal calf serum (FCS). All cells were grown at 37 °C under 5% CO₂. Cells of 60–70% confluency were transfected using Lipofectamine 3000 (Life Technologies) following the manufacturer’s instructions. Transfections for RNA extraction and FACS sorting were performed in 12-well plates using 625 ng individual or combinations of dCas9 expression vector, 500 ng of equimolar pooled gRNA expression vectors. Transfections for MS2-based recruitment were carried out using the same protocol with the following plasmid amounts: 337 ng MS2-gRNA pool, 450 ng dCas9 expression vector and 338 ng MCP-effector plasmid. All experiments were co-transfected with 125 ng of puromycin-resistant plasmid pBABE-puro to select for transfected cells. Transfection medium was replaced 24 h post-transfection with growth medium containing 3-µg/ml puromycin to enrich for transfected cells. After 72 h of puromycin selection, cells for RNA isolation were stored at 4 °C in RNAlater (Ambion) or collected for sorting by flow cytometry. To assay for persistent repression, media was switched to standard growth media 4 days after transfection. For ChIP assays, cells were plated in 10-cm culture dishes and transfection followed the same protocol except that reagents were scaled up accordingly.

Flow cytometry

Transfected HCT116 cells were assayed by flow cytometry after 72-h puromycin selection (96 h after transfection). For flow cytometry collection of the HER2-negative cell population, cells were harvested from two independent biological replicates, washed once in PBS, and resuspended in 1% bovine serum albumin (BSA) in PBS. After adding 3 µl APC-conjugated antihuman CD340 (erbB2/HER-2) antibody (Biolegend #324408), cells were incubated for 30 min at 4 °C. Labeled cells were washed once and resuspended in 1% BSA in PBS. Cell sorting was performed using an Astrios EQ cell sorter (Beckman Coulter) at the UC Davis Flow Cytometry Shared Resource Core. Untreated HCT116 control cells were used to determine the APC signal for HER2-expressing cells, while unlabeled cells were used to determine the sorting gate for HER2-negative cells. Four days after combinatorial epi-dCas9 transfections, HER2-negative cells were collected and replated using standard media. Cells were harvested at indicated time points, dependent on cells reaching ~ 80% confluency. Cells were expanded in 24-well dishes for RNA isolation (5 days, 14 days, 23 days, 39 days, 44 days and 50 days) and in 6-well dishes for DNA methylation analysis (10 days, 17 days and 24 days) and Western blot analysis (40 days). For ChIP assays (24 days), cells were plated in 10-cm dishes.

RNA extraction and reverse-transcription quantitative PCR (RT-qPCR)

Cells for RNA isolation were stored in RNAlater (Ambion) at 4 °C for up to 1 week. RNAlater was removed, and total RNA was isolated using the RNeasy Plus RNA isolation kit (Qiagen). RNA was reverse-transcribed using the SuperScript VILO MasterMix (Invitrogen). RT-qPCR was performed in triplicate using 2 × iQ SYBR mix (Bio-Rad) with the CFX384 Real-Time System C1000 Touch system (Bio-Rad). Gene expression analysis was performed with GAPDH as a reference gene using at least three biological replicates and the following primer sequences: human HER2 primers (HER2-F 5′-GGGAAACCTGGAACTCACCT-3′; HER2-R 5′-GACCTGCCTCACTTGGTTGT-3′), SNURF (human) primers (SNURF-F 5′-CTGTCTGAGGAGCGGTCAGT-3′; SNURF-R 5′-CAGGTACTTGCTGCTGCTGA-3′), human GAPDH primers (GAPDH-F 5′-AATCCCATCACCATCTTCCA-3′; GAPDH-R 5′-CTCCATGGTGGTGAAGACG-3′), mouse Snurf primers (mSnurf-F 5′-TTGGTTCTGAGGAGTGATTTGC-3′; mSnurf-R 5′-CCTTGAATTCCACCACCTTG-3′), and mouse Gapdh primers (mGapdh-F 5′-TGACCACAGTCCATGCCATC-3′; mGapdh-R 5′-GACGGACACATTGGGGGTAG-3′). Primer sequences used for RT-qPCR of potential off-targets are listed in Additional file 2: Table S1. Relative target gene expression was calculated as the difference between the target gene and the GAPDH reference gene (dCq = Cq[target] − Cq[GAPDH]). Gene expression results are indicated as fold change to a reference sample (usually dCas9 without any effector domain), using the ddCq method. A one-way ANOVA (ANalysis Of VAriance) with post hoc Tukey HSD (honestly significant difference) or Dunnett’s test was used to determine statistical significance for different epi-dCas9 treatments.

Western blot analysis

Transfected HER2-negative cells were collected by flow cytometry 4 days after transfection and harvested for protein extraction 40 days after transfection. Cells were lysed in 1 × RIPA buffer (Millipore) supplemented with protease inhibitor cocktail (Roche), and protein concentrations were determined by Bradford assay (Bio-Rad). For Western blots, 30 μg of protein was separated on a 4–15% TGX gel (Bio-Rad) with Tris/glycine/sodium dodecyl sulfate buffer. Protein was subsequently transferred onto a nitrocellulose membrane, and protein loading was evaluated by Ponceau S stain. The membrane was rinsed with deionized water and incubated in blocking solution (5% nonfat dry milk in TBST; 50 mM Tris, 150 mM NaCl, 0.1% Tween-20) for 30 min at room temperature. Membranes were incubated with primary antibody in blocking solution at 4 °C overnight. We used Anti-c-ErbB2/c-Neu (Ab-3) mouse mAb (3B5) antibody at 1:1000 dilution (SIGMA #OP15). After three 10-min washes with TBST, the membrane was incubated with horseradish peroxidase-conjugated antimouse secondary antibody (1:2000 dilution in TBST) for 45 min at room temperature. After three more washes in TBST, proteins were visualized with Amersham ECL Prime Western Blotting Detection Reagent (GE Healthcare) using the ChemiDoc XRS Imaging System (Bio-Rad).

Chromatin immunoprecipitation (ChIP) and ChIP-qPCR

HCT116 cells were transfected with D3A-dCas9/D3L and either KRAB-dCas9 or Ezh2-dCas9 and three gRNAs targeting the HER2 promoter. Control cells were transfected with dCas9 and the same three gRNAs. For transient repression timepoints, cells were cross-linked 4 days after transfection and growth in puromycin-containing media. For persistent repression timepoints, cells were sorted as described above and dCas9 control cells were cross-linked 24 days after transfection. Cross-linking was carried out in 1% formaldehyde for 10 min at room temperature and was stopped with 0.125 M glycine. Cross-linked cells were lysed with ChIP lysis buffer (5 mM PIPES pH8, 85 mM KCl, 1% Igepal) with a protease inhibitor (PI) cocktail (Roche). Nuclei were collected by centrifugation at 2000 rpm. for 5 min at 4 °C and lysed in nuclei lysis buffer (50 mM Tris pH8, 10 mM EDTA, 1% SDS) supplemented with PI cocktail. Chromatin was fragmented using the Bioruptor 2000 (Diagenode) and diluted with 5 vol RIPA buffer (50 mM Tris pH 7.6, 150 mM NaCl, 1 mM EDTA pH8, 1% Igepal, 0.25% deoxycholic acid). ChIP enrichment was performed by incubation with 3 µg H3K9me3 antibody (Diagenode C15410056), 2 µg H3K27me3 antibody (MP07–449), 2 µg H3K27ac antibody (Active Motif #39133) or 2 µg normal rabbit IgG (Abcam ab46540) for 16 h at 4 °C. Immune complexes were bound to 20 µl magnetic protein A/G beads (ThermoFisher) for 2 h at 4 °C. Beads were washed 2 × with RIPA and 3 × with ChIP wash buffer (100 mM Tris pH8, 500 mM LiCl, 1% deoxycholic acid). The final wash was performed in ChIP wash buffer with 150 mM NaCl. Cross-links were then reversed by heating beads in 100 µl ChIP elution buffer (50 mM NaHCO₃, 1% SDS) overnight at 65 °C, and DNA was purified using the QIAquick PCR Purification Kit (Qiagen). qPCR was performed with 2 × SYBR FAST mastermix (KAPA Biosystems) using the CFX384 Real-Time System C1000 Touch Thermo Cycler (Bio-Rad). HER2 ChIP amplification primers are as follows: HER2-ChIP-F (5′-TTGGAATGCAGTTGGAGGGG-3′) and HER2-ChIP-R (5′-GGTTTCTCCGGTCCCAATGG-3′). ChIP enrichment was calculated relative to input samples using the dCq method (dCq = Cq[HER2-ChIP] − Cq[input]). Statistical significance was determined by Student’s t test.

ChIP-sequencing and data analysis

Each entire ChIP sample was used to prepare Illumina sequencing libraries using the KAPA Hyper Prep Kit (Roche) and NEXTflex DNA barcodes (BIOO Scientific). Illumina sequencing libraries were pooled and sequenced using the HiSeq 4000 platform (Illumina) at the UC Davis DNA Technologies Sequencing Core. Short sequence reads (SR50) were aligned to the hg19 genome assembly using using BWA [34], and data were preprocessed using the ENCODE3 ChIP-seq pipeline (https://www.encodeproject.org/chip-seq/). ChIP-seq peaks were called using MACS2 [35] from each replicate first, and then, reproducible peaks were selected using the naïve overlap tool, as suggested in the ENCODE3 ChIP-seq standards document (https://www.encodeproject.org/pages/pipelines/) and previously described [36]. To find reduced H3K27ac ChIP-seq promoter peaks in cells treated with Ezh2 plus D3A and D3L compared to dCas9 control, we selected top 15 k robust reproducible peaks from each dataset. H3K27ac peaks only found in the dCas9 control were identified by overlap analysis with treated cells using bedtools (https://github.com/arq5x/bedtools2). Lastly, peaks located in promoter regions (1-kb windows of transcription start sites) of protein coding genes (Gencode version 28, https://www.gencodegenes.org) were identified (n = 469 peaks). Using the DiffBind R package [37], ChIP-seq signals were normalized across samples and fold change of reduced peaks was calculated using average signal values (Additional file 6: Table S2). ChIP-seq data have been submitted to the Gene Expression Omnibus (GEO) and are available under accession number GSE123882.

Analysis of methylation by bisulfite conversion

Genomic DNA from biological replicates was isolated using the Quick-DNA Miniprep Kit (ZYMO), and bisulfite conversion was performed starting with 500 ng genomic DNA with the EZ DNA Methylation-Gold Kit (ZYMO) following the manufacturer’s instructions. Bisulfite-Sequencing PCR primers were designed using MethPrimer [38] (BSS-HER2-F 5′-GGAGGGGGTAGAGTTATTAGTTTTT-3′ and BSS-HER2-R 5′- CACCTCCTCCTTCTCCTATAATTAAA-3′); 100 ng bisulfite converted DNA was used for PCR amplification with ZymoTaq polymerase (ZYMO), and the 229-bp PCR product was purified with the QIAquick PCR Purification Kit (Qiagen). Amplicons were subcloned into the pCR2.1-TOPO TA vector using the TOPO TA cloning kit (ThermoFisher) and transformed into NEB5α competent cells (NEB). Individual clones were subjected to Sanger sequencing (Genewiz), and methylation status of 11 CpGs immediately upstream of the HER2 TSS (hg19, chr17:37,856,035–37,856,263) was determined. The same amplicons from bisulfite converted DNA were sent for high-throughput sequencing at the CCIB DNA Core Facility at Massachusetts General Hospital (Cambridge, MA). Sequencing data were processed using FLASH2 to merge overlapping paired-end reads and single long reads were then demultiplexed using FASTX barcode splitter by matching barcodes at the beginning or end of the sequences, with one mismatch allowed. Methylation levels were of bisulfite-treated sequencing reads and were analyzed with Bismark [39] using a bisulfite reference and default settings.

Global DNA methylation analysis

HCT116 cells were transfected with D3A-dCas9/D3L and either KRAB-dCas9 or Ezh2-dCas9 and three gRNAs targeting the HER2 promoter. Each replicate (n = 2) was performed as an entirely separate transfection experiment. Transfected cells were sorted by flow cytometry as described above and genomic DNA was extracted. Control cells were transfected with three HER2 gRNAs and dCas9 (no effector domain). Genomic DNA from sorted and control cells was isolated 17 and 24 days after transfection using the Quick-gDNA MiniPrep kit (ZYMO). The Infinium Human MethylationEPIC BeadChip (Illumina) was used to analyze global DNA methylation. Genomic DNA was bisulfite converted and EPIC DNA methylation array data were processed in the USC Molecular Genomics Core; the methylation status of over 850,000 CpG probes was reported. For each probe, methylated (M) and unmethylated (U) signal intensities were recorded and the beta value (M/(M + U)) was determined for each probe. Samples were normalized (background corrected) using the ‘noob’ function in the minfi software program in R computing language. All downstream analysis was conducted using the hg19/GRCh37 human genome assembly. To identify hypermethylated probes located in promoter regions, we first selected probes located in promoter regions (Illumina MethylationEPIC Manifest RefGene annotation 5′UTR, TSS200 and TSS1500) (511,700 CpG probes), and then, probes showing increased DNA methylation levels in treated cells (KRAB + D3A + D3L 17 days, Ezh2 + D3A + D3L 17 days, and Ezh2 + D3A + D3L 24 days) compared to dCas9 control samples (mean beta value difference cut off 0.2) were selected (223, 2966, 2018 probes for KRAB + D3A + D3L 17 days, Ezh2 + D3A + D3L 17 days, and Ezh2 + D3A + D3L 24 days, respectively; Additional file 9: Table S3). On the other hand, hypomethylated probes located in promoter regions were identified by selecting probes showing decreased DNA methylation levels in treated samples compared to control samples (mean beta value difference cut off 0.2). Genes that include more than 3 hypermethylated promoter probes were selected to generate Venn diagrams. EPIC array data have been submitted to the Gene Expression Omnibus (GEO) and are available under accession number GSE123830.

Ezh2-dCas9, but not KRAB-dCas9, enables long-term silencing by creating a persistent heterochromatin environment at the HER2 locus

We have previously created a toolbox of dCas9 fusions with epigenetic effector domains (epi-dCas9), which can deposit their respective epigenetic marks at the endogenous HER2 target site in HCT116 cells when transfected individually [26]. DNMT3A-dCas9 (D3A-dCas9) establishes 5C DNA methylation, Ezh2[FL]-dCas9 (Ezh2-dCas9) establishes H3K27 trimethylation, and KRAB-dCas9 induces H3K9 trimethylation at the target site (Fig. 1a). Individual epi-dCas9 fusions were capable of transient HER2 repression, but they failed to induce long-term repression. Although reports have emerged that a combination of the KRAB repressor domain and DNA methyltransferases is required to engineer epigenetic memory [25, 27], we were unable to use this combination to engineer long-term repression at the HER2 locus in HCT116 cells [26]. In this study, we aim to gain a better understanding of the requirements to stably engineer epigenetic memory at different loci. First, we compared combinatorial treatment of DNA methylation with either KRAB or with the histone methyltransferase Ezh2 and the effect on long-term repression and establishment and maintenance of a repressive epigenetic state. For this purpose, HCT116 cells were co-transfected with either Ezh2-dCas9 or KRAB-dCas9 cocktail, plus D3A-dCas9 and D3L, and three gRNAs targeting the HER2 gene promoter (Fig. 1a). To more easily monitor gene expression over time, we took advantage of the transmembrane properties of the HER2 protein and employed flow cytometry to collect cells with no/low APC-HER2 fluorescence (HER2−) 4 days after transfection (Fig. 1b). On average, 14% of the cells from three biological replicates were HER2− after exposure to the Ezh2-dCas9 cocktail and 28% were HER2− after treatment with the KRAB cocktail (Fig. 1b). Thus, the stronger transient repressive activity of KRAB-dCas9 was reflected in the higher number of HER2− cells, indicating that this method can be used to monitor HER2 gene expression (repression of HER2 expression also leads to persistent reduction in HER2 protein; see Additional file 3: Figure S2A). HER2− cells were then maintained without selection for 50 days after transfection. Unsorted cells transfected with dCas9 (no effector domain) were grown in parallel and used as a control. Time points for individual experiments were determined by expansion of sorted cells to 80% confluency (see "Materials and methods"). As expected, transfected epi-dCas9 fusions were transiently expressed at high levels 5 days after transfection, but expression of dCas9 fusions was no longer detectable 12 days after transfection (Fig. 1b). Ezh2-dCas9 in combination with D3A-dCas9 and D3L produced persistent twofold repression for the entire 50-day period, supporting the hypothesis that a repressive state was established and maintained through approximately 57 mitotic cell divisions (doubling time ~ 21 h) (Fig. 1c). Although KRAB-dCas9 together with D3A-dCas9 and D3L initially repressed HER2 expression by 3.3-fold, repression was lost and normal HER2 expression was re-established by day 14 after transfection. Although we sorted and expanded HER2− cells, we observed persistent repression of only twofold. This could either represent partial repression in individual cells or result from a bimodal population where a subset of cells maintained the HER2− state and another subset regained HER2 expression. We performed FACS analysis to evaluate the HER2 protein level in cells 54 days post-transfection (50 days after the initial sort). FACS analysis demonstrates all-or-none maintenance of silencing events (Additional file 3: Figure S2B). We observed 31% HER2− cells in the population treated with Ezh2-dCas9, D3A-dCas9 and D3L, while 2% of cells remained HER2− after exposure to the KRAB-dCas9 cocktail. No HER2− cells were present after treatment with dCas9 (no effector).

We then investigated whether the persistent change in HER2 expression was accompanied by a change in the chromatin environment at the HER2 promoter. We interrogated a 229-bp region in the HER2 promoter (Fig. 1a) by bisulfite sequencing to determine the methylation status in HER2− cells 10 days after treatment with Ezh2-dCas9 or KRAB-dCas9 cocktail (plus D3A-dCas9 and D3L) and three gRNAs targeting the HER2 gene promoter. Treatment with KRAB-dCas9 cocktail resulted in 39% methylation, while Ezh2-dCas9 facilitated treatment showed increased methylation of 54% (Fig. 1d). No methylation was observed in cells treated with dCas9 with no effector domain (Fig. 1d). We further wanted to investigate whether hypermethylation was associated with spreading of DNA methylation. We used the Illumina Infinium HumanMethylationEPIC (EPIC) array to investigate DNA methylation beyond the target site 17 days after treatment. Cells treated with Ezh2-dCas9/D3A-dCas9 and D3L showed hypermethylation of all 12 CpGs in the HER2 promoter region and the promoter-proximal 5′UTR region with an average methylation of 48% as compared to 7% in dCas9 control samples (Fig. 1e). In addition to the 416-bp nucleation site (spanned by the 3 HER2 gRNAs), hypermethylation spread through the downstream CpG island, creating a 1.25-kb hypermethylated region (Fig. 1e). On the other hand, only a single HER2 promoter CpG was hypermethylated 17 days after treatment with KRAB-dCas9/D3A-dCas9 and D3L. The single hypermethylated CpG was located in the CpG island downstream of the gRNA target region and was not associated with HER2 gene repression (Fig. 1c). This suggests that higher initial methylation level (Fig. 1d) and/or spreading of DNA methylation (Fig. 1e) facilitated by Ezh2-dCas9 plus DNA methyltransferases enables persistence of a repressive chromatin state and thereby enables long-term epigenetic memory at the target locus.

We next determined whether co-targeting with Ezh2-dCas9 or KRAB-dCas9 resulted in deposition and maintenance of the corresponding histone marks (H3K27me3 and H3K9me3, respectively). Because cell number was limited after sorting for HER2− cells, we performed ChIP-qPCR on unsorted cells for the transient time point (4 days), but used the collected HER2− population (Fig. 1b) for the persistent time point (24 days). ChIP-qPCR confirmed the presence of the expected histone marks 4 days after transfection (Fig. 1f), when epi-dCas9 was still present (see Fig. 1b). In addition, the 6.6-fold increase in H3K27 trimethylation at the HER2 promoter mediated by Ezh2-dCas9 was maintained at a similar level (4.7-fold) at 24 days after transfection (Fig. 1f). On the other hand, KRAB-dCas9 initially triggered a significant increase in H3K9 trimethylation (102-fold), but this epigenetic mark was completely lost 24 days after transfection; this loss corresponded to the loss of HER2 repression. To determine whether the gain in the repressive H3K27 methylation mark is accompanied by a loss of the active H3K27 acetylation mark, we performed H3K27ac ChIP-qPCR. Indeed, we observed a 1.4-fold decrease in H3K27ac 4 days after transfection, which was more pronounced (5.2-fold) 24 days after exposure to Ezh2-dCas9 plus D3A-dCas9 and D3L (Fig. 1f). These data clearly demonstrate that targeting Ezh2-dCas9, D3A-dCas9 and D3L establishes and maintains a repressive chromatin environment comprised of DNA and histone methylation.

Long-term silencing by combinatorial treatment with Ezh2-dCas9 or KRAB-dCas9 is locus specific

Co-targeting DNA methylation with Ezh2-dCas9, but not with KRAB-dCas9, stably silenced HER2 expression. We then wanted to investigate whether another locus that is readily silenced by a combination of KRAB-dCas9 with D3A-dCas9 and D3L also responds to combinatorial Ezh2-dCas9 treatment, or whether the two modes of action are mutually exclusive. For this purpose, we evaluated different combinations of epi-dCas9 fusions using three gRNAs targeting the Snurf promoter in Neuro2A cells. Transient repression was measured 4 days post-transfection under puromycin selection to enrich for transfected cells, while persistent repression was measured after cells were grown for an additional 10 days in puromycin-free media (Fig. 2a). Informed by our results (Fig. 1c) and previous reports [25, 26], we evaluated persistent silencing 14 days after transfection. In addition to Ezh2-dCas9, we included dCas9-FOG [N + C], which deposits similar levels of H3K27 trimethylation when targeted to the HER2 promoter [26]. All epi-dCas9 fusions were able to repress Snurf gene expression temporarily, ranging from 2.6- to 14.2-fold downregulation (Fig. 2b). Importantly, dCas9 without any effector domain was also able to repress Snurf gene expression 3.6-fold. This effect, also known as CRISPR inhibition (or CRISPRi), is due to binding of dCas9 near the transcription start site and subsequent blocking access of endogenous factors required for transcription [40]. As expected, individual epi-dCas9 fusions (KRAB-dCas9, Ezh2-dCas9, D3A-dCas9, dCas9-FOG [N + C]) could cause transient repression, but in each case Snurf gene expression was re-established by 14 days after transfection. The observed CRISPRi effect of dCas9 without an effector domain was also of transient nature. Long-term repression of the Snurf gene by KRAB-dCas9 required co-expression of both D3A-dCas9 and D3L (twofold; Fig. 2b). On the other hand, long-term repression of the Snurf gene by Ezh2-dCas9 was observed in combination with D3A-dCas9 (1.6-fold) or after combinatorial treatment with D3A-dCas9 and D3L (1.6-fold). dCas9-FOG [N + C] by itself, or in combination with D3A-dCas9 and D3L, was not able to confer long-term repression of the Snurf gene. Similar results were obtained when targeting the TRPM4 promoter in C42B cells (Additional file 4: Figure S3). Interestingly, these data demonstrate that targeted H3K27me3 deposition by dCas9-FOG [N + C] is not sufficient for long-term repression, even in the presence of targeted DNA methyltransferase, while H3K27me3 deposition by Ezh2-dCas9, in combination with targeted DNA methylation by D3A-dCas9 and overexpression of D3L, does lead to long-term repression of Snurf expression. This suggests that either interaction of the epigenetic effectors themselves (Ezh2 and D3A) or recruitment of PRC2 complex partners specific to Ezh2 is necessary for long-term repression.

We next investigated whether simultaneous recruitment of the histone methyltransferase Ezh2 and DNA methyltransferases is also required in other cell types. It has previously been shown that long-term repression by a targetable fusion with the KRAB repressor domain required simultaneous recruitment of the histone methyltransferase DNMT3A and DNMT3L in some, but not all cell types tested [25]. We observed that KRAB-dCas9 by itself was able to induce modest long-term repression of Snurf in 3T3 cells as well as SNURF and HER2 in LNCaP cells. However, long-term silencing efficiency was increased when a combination of KRAB-dCas9, D3A-dCas9 and D3L was used (Fig. 2c). Efficient long-term repression by Ezh2-dCas9 also required addition of D3A-dCas9 and D3L. We note that the combination of D3A-dCas9 and D3L with Ezh2-dCas9 or KRAB-dCas9 induced comparable levels of long-term repression: Ezh2-dCas9 facilitated 1.9- and 2.2-fold repression, while KRAB-dCas9 established 1.8- and 1.7-fold repression of Snurf/SNURF expression in 3T3 and LNCaP cells, respectively. In LNCaP cells, addition of D3A-dCas9 and D3L to Ezh2-dCas9 or KRAB-dCas9 established 3.4- and 1.9-fold persistent HER2 repression, respectively. Among the two loci tested in different mouse and human cell types, only the HER2 locus in HCT116 cells was resistant to long-term silencing by KRAB-dCas9 in combination with D3A-dCas9 and D3L. This demonstrates that different combinations of targeted histone and DNA methyltransferases are required to achieve maximal repression at the same locus in different cell types or at different loci in the same cell type. Altogether, these data indicate that there are locus- and/or cell-type-specific requirements for the specific tools needed to silence an endogenous gene by targeted recruitment of histone and DNA methylases.

Genome-wide H3K27ac analysis demonstrates that combinatorial Ezh2-dCas9 treatment leads to loss of H3K27ac specific to the HER2 target promoter

We demonstrated that the targeted increase in the repressive histone mark H3K27me3 is associated with long-term loss of H3K27ac (Fig. 1f), which is associated with distal regulatory regions and with gene promoters [9]. Therefore, loss of H3K27ac in promoter regions can be used as an indicator for reduced expression. We performed ChIP-seq 24 days after transfection with either dCas9 (control) or Ezh2-dCas9 cocktail (including D3A-dCas9 and D3L) to analyze H3K27ac signals genome-wide; comparisons of H3K27ac peaks showed strong correlation between biological replicates (Spearman’s ρ > 0.98; Additional file 5: Figure S4). Biological replicates were then combined, and the top 15,000 robust ChIP-seq peaks were called in control and combinatorial Ezh2-dCas9 treatment, independently (Additional file 6: Table S2). When we compared the robust peaks, 13,528 (91%) of H3K27ac peaks found in the control treatment were also present in cells that received Ezh2-dCas9 in combination with D3A-dCas9 and D3L (Additional file 7: Figure S5A and S5B). As expected, a H3K27ac binding peak at the HER2 target site was only identified in control cells, but no robust binding signal was observed after combinatorial Ezh2-dCas9 treatment (Fig. 3a). We were interested whether other promoters showed similarly reduced levels of H3K27ac, which may indicate loss of transcription of the associated gene. 1392 H3K27ac peaks were only present in control cells, but not in cells treated with Ezh2-dCas9 cocktail. Among these control-specific peaks, only 469 H3K27ac binding peaks were localized within ± 1 kb of the transcription start site (TSS). H3K27ac binding enrichment at these 469 TSS proximal regions was calculated after normalization using DiffBindR program [37], and the change in H3K27ac binding signals between control and combinatorial Ezh2-dCas9 treatment was calculated. Only 20 of the 469 promoter-proximal regions showed a more than twofold reduction in H3K27ac binding (Fig. 3b, Additional file 6: Table S2). The strongest reduction, which was 3.7-fold, was observed at the HER2 target site with the H3K27ac peak centered in the proximal promoter (− 288 bp relative to TSS). ARHGEF19 and RNF128 showed a reduction of 3.6- and threefold, respectively. The H3K27ac ChIP-seq signal was smaller at ARHGEF19 and RNF128 (approximately half and a fourth, respectively) than at the HER2 target promoter. It is noteworthy that among the 20 promoter binding peaks showing more than twofold reduction, H3K27ac ChIP-seq signal was highest at the HER2 promoter (Additional file 7: Figure S5C). If H3K27ac changes were due to off-target binding of epi-dCas9 fusions, one would expect an overlap with predicted Cas9 off-targets. We used Cas-OFFinder [41] to identify predicted Cas9 off-targets in the human genome with a ‘NGG’ PAM sequence and allowing up to three mismatches to each of the three HER2 gRNA target sequences. There were 3, 58 and 28 predicted off-target sites for HER2 gRNAs 1,2 and 3, respectively (Additional file 2: Table S1). Using bedtools, we determined that there was no overlap between predicted off-target sites and the 469 TSS proximal regions that showed reduced H3K27ac binding. To determine whether changes in H3K27ac binding affect gene expression, we performed RT-qPCR of the top 10 promoters with reduced H3K27ac enrichment. We found that 4 out of the 10 genes (GBX1, ADCK5, BRINP3, CD8A) were either not expressed or not detectable by RT-qPCR in HCT116 cells. The remaining six gene promoters did not show a change in gene expression 23 days after transfection (Fig. 3c). Thus, we observed very few off-target H3K27ac changes in the genome of the transfected cells and did not observe any changes in gene expression at the off-target loci tested.

Global analysis of persistent DNA methylation after combinatorial targeting with Ezh2-dCas9 or KRAB-dCas9 and DNA methyltransferase (D3A-dCas9 and D3L)

We have demonstrated that treatment with Ezh2-dCas9, plus D3A-dCas9 and DNMT3L, cannot only establish, but also stably maintain DNA methylation (Fig. 1d, e). Some groups have reported very precise targeted methylation [25, 42, 43], and others have found significant amounts of off-target methylation [43–45]. We used the Illumina Infinium HumanMethylationEPIC (EPIC) array to obtain a global view of DNA methylation at human promoters. Global DNA methylation was analyzed 17 days after cells were treated with HER2 gRNAs and Ezh2-dCas9 or KRAB-dCas9 cocktails (containing D3A-dCas9 + D3L) and compared to cells transfected with dCas9 (no effector domain); biological replicates showed strong bivariate correlation (R² > 0.99; Additional file 8: Figure S6). Methylation levels are recorded as beta values (β), which are determined by the ratio of intensities between methylated (M) and unmethylated (U) alleles (β = M/M + U). EPIC arrays reliably detect a change in beta values of 0.2 (methylation change of 20%) with FDR < 0.01. Since hypermethylation of gene promoters is frequently associated with transcriptional repression, average methylation was calculated at promoter regions encompassing from 1.5 kb upstream of the transcription start site (TSS) through the 5′ UTR. On average, each promoter region is represented by 14 CpG probes on the EPIC array. Promoter CpGs showing a 20% increase in DNA methylation are highlighted in Fig. 3d regardless of the original promoter methylation state. Differentially methylated regions (DMRs) usually involve adjacent CpGs or a group of CpGs close together that have different methylation patterns between samples, with a minimum requirement of > 3 differentially methylated CpGs [46, 47]. We determined the frequency of methylated CpGs occurring at each promoter region and observed that the majority of promoter regions only contained a single hypermethylated CpG after transfection with the epi-dCas9 fusions (Fig. 3e). For example, KRAB-dCas9 in combination with D3A-dCas9 and D3L resulted in hypermethylation of 223 CpGs corresponding to 199 genes (GenCode Basic v12). 91% of these genes contained only a single hypermethylated CpG in the entire promoter region (Additional file 9: Table S3). Only one possible off-target gene with > 3 hypermethylated promoter CpGs was identified. The SHOX2 gene promoter contains 4 hypermethylated CpGs with an average methylation of 33% as compared to 6% in dCas9 control cells. We note that the KRAB-dCas9 plus D3A-dCas9 and D3L treated cells were only transiently exposed to DNA methyltransferase. Our results demonstrate that hit-and-run DNA methyltransferase targeting does not display promiscuous off-target activity under these experimental conditions.

Ezh2-dCas9 treatment in combination with D3A-dCas9 and D3L resulted in hypermethylation of more CpG probes (2966) than the KRAB-dCas9 experiment. However, even in this case, only 51 promoters contained > 3 hypermethylated CpGs (Additional file 9: Table S3). Importantly, the targeted HER2 promoter contained the most hypermethylated CpGs (12 hypermethylated CpGs), showing an average methylation of 48% as compared to 7% in dCas9 control samples.

It has previously been reported that gain of off-target methylation by transient overexpression of DNMT3A-dCas9 fusions is of a transient nature [45, 48]. We wanted to determine whether promoter hypermethylation was stabilized 17 days after combinatorial Ezh2-dCas9 treatment with D3A-dCas9/D3L. DNA methylation 24 days after transfection showed high concordance (93%) with 1876 overlapping hypermethylated CpGs at 17 and 24 days after transfection with 2966 and 2018 hypermethylated CpGs, respectively. We then compared the 89 off-target sites predicted by Cas-OFFinder (as described earlier; Additional file 2: Table S1) to the 1876 overlapping hypermethylated CpGs. It is possible that a hypermethylated CpG is not directly within, but adjacent to a predicted off-target site. We therefore expanded the hypermethylated region by the length of a nucleosome up- and downstream of the CpG (± 146 base pairs). There were no overlaps between predicted off-target sites and the expanded hypermethylated regions. Using the criteria of > 3 hypermethylated CpGs per gene promoter, half of the gene promoters we identified to be hypermethylated at 17 days post-transfection were still hypermethylated at 24 day post-transfection (Fig. 3f; Additional file 9: Table S3). Among the 25 persistently hypermethylated gene promoters (represented by 132 CpG probes), HER2 still ranked highest having 12 hypermethylated CpGs and an average methylation level of 46%. SHOX2, which was the only potential off-target identified by D3A-dCas9/D3L treatment with KRAB-dCas9, was also among the 25 persistently hypermethylated gene promoters. To evaluate the impact of DNA methylation on expression of potential off-target sites, we performed RT-qPCR in cells 23 days after co-transfection with the Ezh2-dCas9 cocktail (plus D3A-dCas9/D3L) and compared expression to control cells treated with dCas9 without effector domain (Fig. 3g). To evaluate possible downstream effects of HER2 repression, HCT116 cells with siRNA mediated knockdown of HER2 were used as a control. SHOX2 and MPP1 encompassed 9 and 4 hypermethylated CpG probes, respectively. Neither SHOX2 nor MPP1 showed changes in gene expression when compared to control samples, while HER2 was robustly downregulated in cells treated with HER2 siRNA and in cells 23 days after treatment with Ezh2-dCas9 cocktail (Fig. 3g).

We further hypothesized that true off-targets from combinatorial Ezh2-dCas9/DNMT3A-dCas9 and D3L treatment would contain both an increase in DNA methylation and a decrease in H3K27 acetylation. When we overlapped the 25 hypermethylated CpG probes identified at 24 days after transfection with H3K27ac ChIP peaks that showed reduced binding in Ezh2-dCas9/D3A-dCas9/D3L treated cells, only the target gene HER2 was identified. Taken together, our data demonstrate that the Ezh2-dCas9 cocktail can engineer epigenetic memory at the HER2 locus conferring long-term repression with very few potential off-targets. Persistent repression is associated with a stably maintained epigenetic switch to a heterochromatin environment stably marked by both, histone H3K27 trimethylation and DNA methylation.

Alternative strategies for combinatorial epi-dCas9 recruitment while maintaining a reduced number of gRNAs

Global ChIP-seq and DNA methylation analysis revealed very few potential off-target sites. However, the use of multiple gRNAs constitutes an increased risk for off-target binding of epi-dCas9 fusions. Since a combination of different epi-dCas9 fusions is required to engineer persistent epigenetic silencing, we explored alternative recruitment strategies that will allow us to recruit multiple epi-dCas9 fusions while maintaining a reduced number of gRNAs. First, we explored whether a single gRNA could facilitate targeted epigenetic silencing with multiple epi-dCas9 fusions. Multiple epi-dCas9 fusions are exploiting the same gRNA, possibly leading to DNA binding competition. To test this, we used two epi-dCas9 fusions (KRAB-dCas9 and D3A-dCas9) in combination with D3L, with a pool of three gRNAs or a single gRNA targeting the Snurf promoter in Neuro2A cells. Surprisingly, we found that a single gRNA was just as effective as the gRNA pool in inducing long-term repression. Snurf gRNA 1 induced 2.3-fold repression, while the gRNA pool (gRNAs 1,2 and 3) resulted in 1.6-fold repression of Snurf expression (Additional file 10: Figure S7A).

Alternatively, recruitment of effector domains through the gRNA component may offer a solution to increasing the number of epi-dCas9 fusions while reducing the number of gRNAs and therefore the number of possible off-target sites (Additional file 10: Figure S7B). In previous studies, gRNA scaffolds were used that contain two MS2 aptamers (MS2-gRNA), one in the gRNA tetraloop and one in stem loop 2. These aptamers can be recognized by the MS2 bacteriophage coat protein (MCP), to which various effector domains (VP64, P65, P65-HSF1, SS18 or HP1) can be attached [33, 49–52]. For our study, we first created fusions of KRAB and Ezh2 to either one or a tandem array of two MCPs, resulting in either 1x or 2x MCP-effector fusions. MS2-gRNAs were designed to target the same endogenous sites as their respective traditional gRNAs. When dCas9 was targeted to the HER2 promoter by a pool of three MS2-gRNAs, 2xMCP-KRAB showed twofold repression at 4 days post-transfection in HCT116 cells, while 1xMCP-KRAB was not able to repress HER2 expression. We then determined whether effector pairing of epi-dCas9 and MCP fusions could increase the repressive capacity of gRNA-based recruitment. Simultaneous recruitment of KRAB-dCas9 and 2xMCP-KRAB decreased repression to 1.3-fold, perhaps by destabilizing the dCas9 complex. However, the direct fusion of KRAB to the N-terminus of dCas9 (KRAB-dCas9) was most effective resulting in tenfold repression of HER2 expression. 1xMCP-Ezh2 and 2xMCP-Ezh2 fusions were unable to repress HER2 gene expression. We additionally performed this test at the Snurf promoter in Neuro2A cells. Again, only the direct fusion of KRAB-dCas9 was able to persistently repress Snurf expression, while one or two MS2-gRNAs recruiting 2xMCP-KRAB to the dCas9 complex did not cause significant long-term repression (Additional file 10: Figure S7B). While the MS2-gRNA recruitment strategy holds great potential, direct epi-dCas9 fusions in combination with a single-gRNA induced persistent epigenetic silencing more efficiently under these conditions.

Co-targeting of full-length DNMT3L is necessary and sufficient to enable persistent gene silencing by Ezh2-dCas9

The catalytic activity of the de novo DNA methyl transferase DNMT3A is enhanced by dimerization with DNMT3L [28]. In fact, a direct fusion of the C-terminal portion of DNMT3L to the catalytic domain of Dnmt3a (D3a3L) to a DNA targeting module (zinc finger protein or dCas9) enabled efficient DNA methylation at endogenous target sites [29, 42, 53]. We have shown that the D3A-based recruitment approach can initiate long-term repression in the presence of overexpressed DNMT3L (Fig. 4a). Alternatively, we sought to explore the impact of a targetable version of DNMT3L on long-term silencing by co-targeting D3A-dCas9 and targetable DNMT3L (D3L) fusions (Fig. 4b). In order to avoid overexpression of the catalytically active DNMT3A, we tested the ability of targetable D3L fusions to recruit endogenous D3A activity (D3L-based recruitment; Fig. 4c).

Although the C terminus of DNMT3L is able to dimerize with DNMT3A, we hypothesized that the ADD domain located in the N-terminal portion also plays an important role. Therefore full-length DNMT3L (D3L) was fused to either the N or the C terminus of dCas9 resulting in D3L-dCas9 and dCas9-D3L, respectively (Fig. 4d). Overexpression of DNMT3L in addition to KRAB-dCas9 and D3A-dCas9 yielded the strongest long-term repression (2.7-fold) at the Snurf promoter in Neuro2A cells (Fig. 4a). The combination of KRAB-dCas9 with dCas9-D3a3L resulted in long-term repression, but at a reduced level (1.4-fold) (Fig. 4a). Comparable repression (1.5-fold) was obtained when KRAB-dCas9 was combined with D3A-dCas9 and targetable DNMT3L fusions (D3L-dCas9 and dCas9-D3L; Fig. 4b). Ezh2-dCas9 also showed the strongest Snurf repression (threefold) when added with D3A-dCas9 and overexpressed D3L. On the other hand, Ezh2-dCas9 did not show any long-term repression when accompanied by dCas9-D3a3L or the combination of D3A-dCas9 and D3L-dCas9 (N-terminal D3L fusion). Thus, dCas9-D3L (C-terminal D3L fusion) was the only targetable version of DNMT3L capable of inducing long-term Snurf repression (1.8-fold) in combination with Ezh2-dCas9 and D3A-dCas9 (Fig. 4b). The same trend was observed at the HER2 locus in HCT116 cells. Ezh2-dCas9 and D3A-dCas9 were only able to persistently repress HER2 expression when combined with D3L overexpression or with the targetable C-terminal version dCas9-D3L (1.6 or 1.4-fold, respectively). Both the addition of dCas9-D3a3L and the combination of D3A-dCas9 with N-terminal D3L-dCas9 were not able to induce long-term repression, suggesting that full-length DNMT3L is necessary for long-term repression by Ezh2-dCas9 and D3A-dCas9, either due to recruitment of protein partners or to effects on activity due to conformational differences between the different constructs.

Overexpression of targetable DNA methyltransferase DNMT3A has recently raised concerns about off-target DNA methylation. The enzymatic activity of DNMT3A is enhanced by its dimerization partner DNMT3L, which does not display catalytic activity by itself. So far, we have introduced D3L by overexpression or by co-targeting with dCas9-D3L (Fig. 4a, b). We hypothesized that targeting of dCas9-D3L to an endogenous promoter is sufficient for recruitment of endogenous D3A to stably repress the target locus. To test this hypothesis, we co-transfected LNCaP cells with a combination of dCas9-D3L and either KRAB-dCas9 or Ezh2-dCas9 together with a gRNA pool targeted to the SNURF and HER2 promoters. Long-term repression was determined 14 days after transfection and compared to dCas9 with no effector. Significant long-term repression was observed with dCas9-D3L and either KRAB-dCas9 or Ezh2-dCas9 with 1.8- or 1.3-fold repression of SNURF expression, respectively and 2.5- or twofold of HER2 expression, respectively (Fig. 4c). We performed bisulfite sequencing to determine the methylation status achieved with the D3L-based recruitment approach at the HER2 promoter (Fig. 4e). To directly compare this approach to the previously applied combinatorial treatment of KRAB-dCas9 or Ezh2-dCas9 with D3A-dCas9 and overexpressed D3L (Fig. 1d), we collected HER2− cells by FACS 4 days after co-transfection with three gRNAs targeting the HER2 promoter in HCT116 cells. HER2− cells were expanded until sufficient amounts were available for bisulfite sequencing (10 days). Co-targeting of KRAB-dCas9, D3A-dCas9 and dCas9-D3L resulted in 22% methylation, while combinatorial treatment of KRAB-dCas9/D3A-dCas9 with overexpressed D3L increased methylation to 46%. When D3A-dCas9 was omitted, combinatorial treatment of KRAB-dCas9 or Ezh2-dCas9 with dCas9-D3L resulted in an average methylation of 20% and 29%, respectively (Fig. 4e). Surprisingly, comparable methylation levels (20%) were reached when dCas9-D3L itself was targeted to the HER2 promoter. The HER2 promoter region was unmethylated in the untreated control and in cells treated with dCas9 (no effector domain). Co-targeting of KRAB-dCas9, D3A-dCas9 and dCas9-D3L to the HER2 promoter resulted in low levels of methylation (9%) suggesting that this approach is less efficient, which is also reflected in lower repressive ability (Fig. 4b).

Long-term repression facilitated by dCas9-D3L is advantageous for two reasons; (1) co-targeting requires only one epi-dCas9 (D3L fusion) instead of two components (D3A-dCas9 plus dCas9-D3L) for efficient long-term repression and (2) co-targeting with dCas9-D3L avoids overexpression of enzymatically active D3A and may therefore offer another strategy to limit off-target effects.