Dual Inhibition of DNA and Histone Methyltransferases Increases Viral Mimicry in Ovarian Cancer Cells

Recent advances in cancer genomics and epigenomics have revealed that the vast majority of human cancers harbor both genetic and epigenetic alterations (1–3). Epigenetic mechanisms, including DNA and histone modifications, as well as chromatin accessibility, determine how cells express genes and play a crucial role in regulating normal cellular functions (4). Aberrant DNA methylation occurs in almost every cancer type and is among the earliest and most common event during tumorigenesis (5). Unlike genetic abnormalities, which are difficult to reverse, epigenetic alterations are readily reversible, making them attractive therapeutic targets for cancer treatment (6).

A group of inhibitors that target epigenetic modifiers has emerged as an exciting group of compounds for use in the clinic (6). For example, the FDA has approved DNA methyltransferase inhibitors (DNMTi), such as 5-aza-2′-deoxycytidine (5-aza-CdR), for the treatment of myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML). Currently, there are ongoing clinical trials for solid tumors including colon, ovarian, liver, lung, and breast cancer (6, 7). Proposed mechanisms underlying the clinical efficacies of DNMTis include demethylation of the promoters of tumor suppressor genes and bodies of oncogenes, thereby restoring more normal expression levels of these genes (8). Moreover, recent work has suggested a novel mechanism of action for DNMTis, termed “viral mimicry” (9–11). Viral mimicry is characterized by the upregulation of endogenous retrovirus (ERV) transcripts, formation of cytoplasmic double-stranded RNA, and the induction of viral defense pathways (9–11). Importantly, the sensing of these ERVs by viral defense proteins leads to the death of colon cancer stem cells (11). The response to DNMTis in a subset of patients was impressive, with long-term durable antitumor effects (12–15).

However, primary and secondary resistance to the epigenetic therapies is common, likely because epigenetic processes are established and reinforced by both positive and negative feedback loops (16, 17). For example, while DNMTis induce immediate and genome-wide demethylation of DNA, their effectiveness can be limited by a fairly rapid remethylation after the drug is removed (8), and/or by the involvement of alternative silencing mechanisms such as histone modifications (18). The future of epigenetic therapies, especially for solid tumors, relies on the rational design of combination treatments taking alternative silencing mechanisms into consideration.

DNA methylation plays an important role in silencing ERVs in somatic cells and in the male differentiating germline (19–22), yet it is dispensable for the repression of a subset of ERVs in the early germline, early embryo, and embryonic stem (ES) cells (23–25). Silencing of ERVs in early embryo and germline development depends primarily on histone methylation, notably at lysine 9 of histone H3 (H3K9). Many studies have shown that H3K9 lysine methyltransferases, including G9A, GLP, and SETDB1, are essential for repressing ERV expression in these stages (24–27). Importantly, G9A, which is responsible for catalyzing mono- or dimethylation of H3K9, is upregulated in many types of cancers (28–30). G9A and H3K9me2 are present at DNA hypermethylated promoters of tumor suppressor genes in cancers and are lost from the promoters when these genes are reactivated by DNMTis or in DKO cells (31, 32). The level of G9A overexpression has been associated with poor prognosis of ovarian, colon, breast, and lung cancers (30, 33–36). We have previously shown that knocking down G9A sensitizes a colon cancer cell line to 5-aza-CdR treatment (37). In addition, G9Ai in combination with DNMTis has been proposed as a potential therapy for sickle cell disease, because this combination further activates fetal hemoglobin genes (38). This combination has also been shown to effectively increase the expression of immune genes in a colon cell line (39). These findings urged us to evaluate this novel combination as a potential epigenetic therapy approach for ovarian cancer. Here, we show that pharmacologic inhibition of G9A synergistically enhances the efficacy of 5-aza-CdR by restoring strong promoter activity of ERVs and enhancing the viral mimicry effects in ovarian cancer cells with high levels of G9A and GLP expression.

A2780, CAOV3, PEO14, and OAW42 ovarian cancer cell lines were authenticated using the ATCC human cell line authentication service. All cell lines were cultured according to standard mammalian tissue culture protocols using sterile technique. A2780 were maintained in McCoy's 5A medium, PEO14 cells in RPMI 1640 medium and CAOV3 and OAW42 in Dulbecco's Modified Eagle Medium. All media (from GIBCO) were supplemented with 10% fetal bovine serum (Sigma-Aldrich) and 1% penicillin/streptomycin (GIBCO). Venor GeM Mycoplasma Detection Kit (Sigma-Aldrich) was used every 3 months to confirm all cell lines were Mycoplasma-free. 5-Aza-CdR and G9Ai (UNC0638) were purchased from Sigma-Aldrich and APExBIO, respectively.

To test for dose-dependent response, 400 A2780 cells or 1,000 CAOV3, PEO14, and OAW42 cells were plated in each well of the 96-well plates 24 hours prior to the treatment. A2780 and OAW42 cells were exposed to a dose of 5-aza-CdR (from 25 to 4,800 nmol/L) for 48 hours, while CAOV3 and PEO14 cells were treated with 2 consecutive daily doses of 5-aza-CdR (from 25 to 4,800 nmol/L) for 48 hours. Subsequently G9Ai UNC0638 (from 100 to 20,000 nmol/L) was added to the culture media until 7 days after the treatment. Cells were then incubated with CellTiter-Glo assay reagent (Promega) for 10 minutes and luminescence was measured using a Synergy HT multimode microplate reader (BioTek).

Cell viability data were normalized to their corresponding untreated controls for each treatment condition and were expressed as percentage fractional affect (Fa). CompuSyn software (ComboSyn, Inc.) was used to calculate combination index (CI) values of Fa under different conditions using the Chou–Talalay equation (40) CI = (D)_Vc/((D_m)_Vc (Fa/(1-Fa))^1/m1) + (D)_Aza/((D_m)_Aza (Fa/(1-Fa))^1/m2), where D is the concentration of G9Ai and 5-aza-CdR either alone or in combination to achieve a given Fa. The median-effect dose (D_m), m1 (G9Ai), and m2 (5-aza-CdR) values were determined using the median-effect equation (41) (Fa)/(1-Fa) = ((D)/(D_m))^m for G9Ai and 5-aza-CdR treatment alone. The CI values define synergistic effect (42) when CI < 1, additive effect when CI = 1, and antagonism when CI > 1.

Cells were harvested and then stained with the amine reactive viability dye Ghost Dye Violet 450 (Tonbo Biosciences) at 4°C for 30 minutes according to the manufacturer's protocol. Cells were then fixed with 66% ethanol and stored at 4°C until ready to stain with propidium iodide (PI). Fixed cells were pelleted and resuspended in 500 μL of PI staining solution (PBS + 100 μg/mL RNase A + 50 μg/mL PI) and incubated overnight at 4°C. For flow cytometry analysis, cells were filtered through cell strainer snap caps (Fisher Scientific) and then analyzed on a CytoFLEX S (Beckman Coulter). Cell cycle was analyzed using ModFit LT software (Verify Software House, www.vsh.com), and the percentage of cells alive and dead was measured using FlowJo v10.0.7 (FlowJo, LLC).

Cell lysis and washing steps were performed in cold buffer containing 10 mmol/L PIPES, pH 7.0, 300 mmol/L sucrose, 100 mmol/L NaCl, 3 mmol/L MgCl2, 1 × EDTA-free protease inhibitor (Roche), 1 × phosphatase inhibitor cocktail (Sigma), and 0.1% Triton X-100. Whole-cell and chromatin fractions were treated with benzonase (Sigma-Aldrich) prior to Western blot analysis. Whole-cell, chromatin-associated or soluble fractions were mixed with SDS/β-mercaptoethanol loading buffer and resolved on a Bio-Rad 4% to 15% gradient SDS/PAGE gel. Antibodies against G9A (Perseus Proteomics Inc., PP-A8620A-00), Tubulin (Cell Signaling, 86298S), TATA-binding protein (TBP; Santa Cruz Biotech sc-74596), H3K9me1 (Epicypher 13-0014), H3K9me2 (Abcam #1220), H3K9me3 (Active Motif #39161), and total H3 (Abcam #12079) were used. Proteins were visualized using the Clarity Western ECL substrate (Bio-Rad) and ChemiDoc XRS+ imaging system (Bio-Rad).

Total RNA was extracted with TRIzol reagent (Invitrogen), followed by cleanup and Turbo DNase I (Invitrogen) treatment with a Zymo Direct-Zol RNA mini prep kit (Zymo Research) according to the manufacturer's protocol. RNA quality was assessed using Agilent 2100 bioanalyzer with RNA Nano chips (Agilent Technologies, Inc.). For directional RNA-seq with ribosomal RNA (rRNA) reduction, libraries were prepared using the KAPA Stranded RNA-seq Kit with RiboErase (HMR; KapaBiosystems) and sequenced as single-end 75 bases on a NextSeq 500 instrument (Illumina) at the Van Andel Research Institute Genomics Core. The RNA-seq reads were mapped to the human transcriptome using TopHat version 2.1.0 with NCBI RefSeq as the reference annotation of transcripts. The transcripts were assembled and quantified using Cufflinks version 2.2.1. Differential expression is measured using edgeR package from the Bioconductor project.

To quantify the transcription of repetitive element at a specific locus, we used Repeatmasker (http://www.repeatmasker.org) annotation as our input and considered only uniquely mapped reads (with mapping quality threshold 10) as previously described (10). For each transcript, we separated reads mapped to the two strands. We considered a transcript as bidirectionally transcribed if the smaller read count divided by the greater read count is over 0.5.

On day 5 after treatment, A2780, CAOV3, PEO14, and OAW42 cells were harvested, and genomic DNA was purified by phenol–chloroform extraction and ethanol precipitation. Bisulfite conversion was performed using the EZ DNA Methylation Kit (Zymo). MethyLight assays were then performed as previously described (43) using the primers listed in Supplementary Table S1. For A2780 cells, bisulfite PCR was performed using bisulfite-converted DNA with primers listed in Supplementary Table S1, and the products were cloned using the pGEM-T Vector System I (Promega) as previously described (10).

Single nucleosome preparation was performed according to the Dilworth lab native chromatin immunoprecipitation (ChIP) protocol (44). Briefly, A2780 cells (10⁷ cells) before and after 5-aza-CdR, G9Ai, or combination treatment were harvested and washed twice and resuspended in ice-cold buffer N (15 mmol/L Tris pH 7.5, 15 mmol/L NaCl, 60 mmol/L KCl, 8.5%(w/v) sucrose, 5 mmol/L MgCl₂, 1 mmol/L CaCl₂, 1 mmol/L DTT, 200 μmol/L PMSF, 1× cOmplete Mini EDTA-free Protease Inhibitor Cocktail; Roche). To prepare nuclei, cells were lysed in 1 mL Lysis Buffer (Buffer N supplemented with 0.3% NP-40 substitute; Sigma) for 10 minutes at 4°C, and nuclei were collected by centrifugation (500 × g for 5 minutes at 4°C), resuspended in 1 mL of Buffer N, then sedimented through 7.5 mL sucrose cushion (10 mmol/L HEPES pH7.9, 30%(w/v) sucrose, 1.5 mmol/L MgCl₂ and centrifuged 13,000 × g using Sorvall swinging bucket for 12 minutes at 4°C). To isolate single nucleosomes, the nuclei were digested with MNase (1U Worthington MNase per 70 μg of chromatin at 37°C for 10 minutes), the nucleosomes were then purified by hydroxyapatite chromatography, and adjusted to a concentration of 20 μg/mL with ChIP Buffer 1(25 mmol/L Tris pH 7.5, 5 mmol/L MgCl2, 100 mmol/L KCl, 10% (v/v) glycerol, 0.1% (v/v) NP-40 substitute) and analyzed using 2% agarose gel.

H3K9me2, H3K4me3, and H3K27ac ChIP were performed as previously described (45), using 5 μg of nucleosome pulled down with 10 μg of anti-H3K9me2 (Abcam ab1220), anti-H3K4me3 (Abcam ab1012), and anti-H3K27ac (Active Motif AM39133) antibody on Dynabeads Protein G (Invitrogen) for 2 hours at 4°C. Initial chromatin (10%) for each immunoprecipitation was set aside to serve as ChIP input. The beads were washed 3 times with ChIP Buffer 2 (10 mmol/L Tris pH 7.5, 5 mmol/L MgCl2, 300 mmol/L KCl, 10% (v/v) glycerol, 0.1% (v/v) NP-40 substitute), twice with ChIP Buffer 3 (10 mmol/L Tris pH 7.5, 250 mmol/L LiCl, 1 mmol/L EDTA, 0.5% Na • deoxycholate, 0.5%(v/v) NP 40 substitute), twice with 1 × TE buffer followed with two elution steps in elution buffer (50 mmol/L Tris pH 7.5, 1 mmol/L EDTA, 1% w/v SDS). After proteinase K (Roche) digestion (65°C 1 hour), sample DNA was purified using Agencourt AMPure XP beads (Beckman Coulter) prior to qPCR analysis. Primers used in this study are listed in Supplementary Table S1.

All data have been deposited at the Gene-Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) with accession code GSE108223.

RNA-seq data from The Cancer Genome Atlas (TCGA) project reveal that G9A expression is upregulated in most types of cancer, and that the highest levels are found in primary ovarian tumors among 33 cancer types (Supplementary Fig. S1A). In addition, the related histone methyltransferase GLP is also significantly overexpressed in ovarian tumors compared with normal tissues (Supplementary Fig. S1B). We therefore used four ovarian cancer cell lines A2780, CAOV3, PEO14, and OAW42 as our in vitro models for treatment. We first tested various concentrations of a G9A inhibitor (UNC0638) in A2780 cells and found that the global levels of H3K9me1 and H3K9me2 were efficiently reduced by 400 nmol/L G9Ai at 48 hours after the treatment (Fig. 1A). In addition, UNC0638 at 400 nmol/L exhibited low cellular toxicity for all four ovarian cancer cell lines (Supplementary Fig. S2). This is consistent with a previous report that UNC0638 has high G9A/GLP inhibition potency and low cellular toxicity with concentrations in the nanomolar range (46). We therefore used 400 nmol/L UNC0638 to determine the optimum dosing schedules for combination treatments with low-dose 5-aza-CdR (100 nmol/L) in A2780 cells. We explored different time points to add G9Ai relative to 5-aza-CdR and varied the duration of exposure to these compounds (Fig. 1B). The expression levels of HERV-Fc1, a marker for the effects of DNMTis (9, 10), were used as readout and were measured by quantitative RT-PCR. The dosing schedule in which A2780 cells were exposed to 5-aza-CdR for 48 hours and subsequently G9Ai until harvest achieved the greatest effect on upregulating HERV-Fc1 (Fig. 1B). Therefore, this schedule was applied to A2780 and OAW42 cells in subsequent experiments (Fig. 2A). Because the doubling times for CAOV3 and PEO14 cells were greater than 24 hours, we treated these cells with two consecutive daily doses of 5-aza-CdR for 48 hours to compensate for the fact that incorporation of the drug requires cell doubling (Fig. 2B).

Following these dosing schedules, we tested whether the combinations of G9Ai and 5-aza-CdR could induce synergistic effects in ovarian cancer cells. We measured dose-dependent inhibition of cell proliferation [as fraction affected (f_a)] with various concentrations of 5-aza-CdR and G9Ai alone or in combination using the CellTiter-Glo luminescent cell viability assay (Supplementary Fig. S3). Combination treatments resulted in increased inhibition of cell proliferation (increased f_a) compared with single-compound treatment in A2780 and CAOV3 cells (Supplementary Fig. S3). The Chou–Talalay analyses (42) determined that the two compounds indeed acted synergistically (CI values <1) under the majority of conditions in A2780 and CAOV3 cells (Fig. 2C). In contrast, the synergistic effects were limited in PEO14 and OAW42 cells (Fig. 2C; Supplementary Fig. S3).

As further characterization of the effects by combination treatments, we next monitored changes of total cell counts upon treatments using a Coulter counter. In this experiment, A2780, CAOV3, PEO14, and OAW42 cells were cultured in 100-mm dishes and treated with 400 nmol/L G9Ai alone or in combination with 100 nmol/L 5-aza-CdR following the dosing schedules outlined above. The results showed that A2780 and CAOV3 cells exhibited significantly reduced total cell counts by combination treatment compared with untreated or single-compound–treated cells (Supplementary Fig. S4). Because this effect could be due to reduced cell proliferation rate or increased cell death, we next examined the cellular phenotypes by flow cytometry using an amine reactive viability dye and performed cell-cycle analysis to test these possibilities. The reduced total cell counts in A2780 and CAOV3 cells were associated with increased cell death by combination treatment (Fig. 2D), while cell-cycle parameters remained unchanged compared with 5-aza-CdR treatment (Fig. 2E). These data indicated that low dose of G9Ai and 5-aza-CdR could act synergistically by promoting cell death in A2780 and CAOV3 cells. To the contrary, no significant differences in total cell counts (Supplementary Fig. S4), percentages of dead cells (Fig. 2D),and cell-cycle parameters (Fig. 2E) were found between combination treatment and 5-aza-CdR–treated PEO14 and OAW42 cells. These data further confirmed that the two compounds have no synergistic effects on cell growth or inducing cell death in PEO14 and OAW42 cells.

To investigate gene-expression changes underlying the synergistic antitumor effects of combination treatments, we sequenced total RNA in A2780, CAOV3, PEO14, and OAW42 cells after treatments. A P value of 0.05 and 2-fold expression change were used as cutoffs to identify differential expressed genes between treated and untreated cells from two independent experiments. In general, G9Ai treatment alone did not induce global gene-expression changes in the four ovarian cancer cell lines (Supplementary Fig. S5), with only a few genes significantly upregulated in A2780 and CAOV3 cells (Fig. 3A). Although genes upregulated by 5-aza-CdR and combination treatment largely overlapped, more genes were upregulated by combination treatment compared with 5-aza-CdR treatment alone in all four cell lines (Fig. 3A; Supplementary Figs. S5 and S6A–S6C). We therefore performed gene ontology analysis to identify the functions of genes that were uniquely upregulated by combination treatment. Interestingly, the immune response pathways, especially the interferon pathways, as cellular defense response to virus were overrepresented in the set of genes uniquely upregulated by combination treatments in A2780 and CAOV3 cells (Fig. 3B). We and others have previously shown that these viral defense genes can be upregulated by 5-aza-CdR treatment and are responsible for inducing apoptosis or inhibiting proliferation of cancer cells (9–11). Therefore, the expression status of a panel of 24 viral defense genes was examined after G9Ai, 5-aza-CdR, and combination treatment compared with untreated cells (Fig. 3C). Consistent with previous reports (9–11), 5-aza-CdR treatment upregulated these viral defense genes in a cell type–specific pattern (Fig. 3C; Supplementary Fig. S7A and S7B), whereas G9Ai treatment alone slightly increased the expression of these genes less than 2-fold relative to untreated cells (Fig. 3C; Supplementary Fig. S7A and S7B). These viral defense genes were further upregulated by the combination treatment (Fig. 3C; Supplementary Fig. S7A and S7B), suggesting that the synergistic antitumor effects by the combination treatment were contributed in part by the increased viral defense response in A2780 and CAOV3 cells. In contrast, the same analysis showed that the viral defense pathway were not further upregulated by combination treatments in PEO14 and OAW42 cells (Supplementary Figs. S6B–S6C and S7C–S7D), in which no synergy was found (Fig. 2C–E). This observation strengthens our conclusion that the synergistic effects between G9Ai and 5-aza-CdR are associated with further induction of the viral defense pathway.

Because induction of ERV expression, especially as dsRNA, is key for triggering the viral defense response by 5-aza-CdR (9–11), we next tested whether ERVs were further upregulated by combination treatment. We mapped transcripts unique to ERV loci in two replicates of the RNA-seq data for all four cell types after the various treatment regimens. ERVs located within coding genes (including introns) were removed from analysis to focus on expression induced by their LTRs rather than as part of host genes. G9Ai treatment alone increased the total reads of intergenic ERVs in CAOV3 but not in A2780 cells (Fig. 4A). Although 5-aza-CdR increased the total intergenic ERV counts compared with untreated A2780 and CAOV3 cells, a further increase after combination treatment was found (Fig. 4A).

Our previous data also showed that some ERVs could be bidirectionally transcribed and that these overlapping transcripts can pair to form dsRNA (10). In addition, dsRNA are preferred substrates for the RIG-I–like receptors, such as RIG-I or MDA5, and are important inducers of antiviral immunity (9, 11, 47). We therefore analyzed the strand-specific RNA-seq data and detected increases in bidirectional transcribed ERVs mainly by 5-aza-CdR and combination treatment, similar to total intergenic ERVs (Fig. 4B). G9Ai treatment alone, however, significantly increased the total counts of the bidirectionally transcribed ERVs in both A2780 and CAOV3 cells (Fig. 4B). These data support a correlation between total bidirectionally transcribed ERVs and the upregulation of viral defense genes (Figs. 3C and 4B). Therefore, combination treatment of G9Ai and 5-aza-CdR enhanced the previously reported “viral mimicry” response by 5-aza-CdR alone, with increased total ERV counts, especially the dsERV counts, that further upregulated an antiviral immune response in these cells.

Interestingly, we observed that G9A and DNA methylation appears to repress a distinct set of ERVs, because the majority of ERVs upregulated by G9Ai treatment did not overlap with those upregulated by 5-aza-CdR (Fig. 4C). This is consistent with our recent finding that there is a switch in silencing mechanisms depending on the evolutionary age of ERVs (48). Evolutionary “young” ERVs tend to be CpG-rich and are repressed mainly by DNA methylation, while “intermediate aged” ERVs with lower CpG density are repressed predominantly by histone modifications, particularly H3K9 methylation (48). The combination treatment of G9Ai with 5-aza-CdR resulted in the activation of more ERVs than single-compound–treated A2780 and CAOV3 cells (Fig. 4C; Supplementary Fig. S8), the majority of which were uniquely upregulated by the combination treatment (Fig. 4C). These data suggest that these ERVs might be silenced by both G9A and DNA methylation. Alternatively, an “epigenetic switch” might happen at these ERVs, which become silenced by histone methyltransferases such as G9A upon loss of DNA methylation. Similar mechanisms have been described previously (18, 48–50). We also observed that different ERVs were upregulated in A2780 and CAOV3 cells, showing a cell line–specific pattern of ERV upregulation (Fig. 4D). Using quantitative RT-PCR, we confirmed that HERV-Fc1 and MLT1N2 were robustly upregulated by 5-aza-CdR and further by combination treatments in both cells (Fig. 4D; Supplementary Fig. S9A and S9B). However, LTR12C, a family of ERVs previously found to be induced by 5-aza-CdR treatment (10, 51), was not further upregulated by combination treatment (Supplementary Fig. S9A and S9B).

In PEO14 and OAW42 cells, however, no further increases in intergenic ERVs or bidirectionally transcribed ERVs were found with combination treatments compared with 5-aza-CdR–treated cells (Supplementary Fig. S10A and S10B), which might explain why no further upregulation of the viral defense genes is elicited by combination treatments. In summary, the cell lines exhibited variable responses to the combination treatments with respect to ERV upregulation.

To address the cell line–dependent effects on ERV upregulation by the combination treatments, we next examined the effectiveness of 5-aza-CdR in inhibiting DNA methylation. We previously observed that 5-aza-CdR treatment in HCT116 cells with stable knockdown of G9A causes a decrease in DNA methylation levels at the promoters of the G9A-targeted genes (37). We therefore used a MethyLight assay (43) to determine DNA methylation level at LINE-1 elements, an approximate measure of global DNA methylation levels. DNA methylation was not inhibited by G9Ai alone, but was more strongly inhibited by 5-aza-CdR in A2780 and CAOV3 than in PEO14 and OAW42 cells (Fig. 5A; Supplementary Fig. S11). No synergy in inhibition of DNA methylation was seen in any of the combination treatments (Fig. 5A). We then performed bisulfite sequencing to examine the DNA methylation levels at the 5′ LTR of HERV-Fc1, which serves as its promoter (illustrated in Fig. 5B), because the expression levels of HERV-Fc1 were further upregulated by combination treatment compared with 5-aza-CdR treatment alone in A2780 cells (Fig. 1B). Consistent with the MethyLight assay results, bisulfite sequencing data showed no robust changes in DNA methylation induced by G9Ai treatment alone or in combination with 5-aza-CdR in A2780 cells (Fig. 5C). Thus, the enzymatic activity of G9A was dispensable for the maintenance or restoration of DNA methylation status at LINE-1 elements or the LTR of HERV-Fc1 after 5-aza-CdR treatment. To summarize, although DNA demethylation was required for upregulation of a group of ERVs such as HERV-Fc1, the synergistic effects of the combination treatment to further upregulate these ERVs did not involve further DNA demethylation by G9Ai.

Reduction of global and gene locus–specific G9A protein and H3K9 dimethylation levels by 5-aza-CdR treatment alone has been reported previously (30–32). Because the G9Ai (UNC0638) acts as a competitive inhibitor, the combination treatment with G9Ai and 5-aza-CdR could further reduce the chromatin-bound G9A protein and H3K9 methylation levels to increase promoter activity synergistically. To test this possibility, we performed chromatin association assays to quantify chromatin-bound G9A protein levels at day 5 after the treatments in A2780 cells. G9Ai treatment alone did not significantly change the chromatin-bound G9A protein levels (Fig. 6A and B), but increases free G9A proteins in both whole-cell and soluble fractions (Supplementary Fig. S12A–S12C). 5-Aza-CdR treatment resulted in a significant reduction of chromatin-bound G9A protein levels, which were further reduced by combination treatment in A2780 cells (Fig. 6A and B), despite the fact that more soluble G9A proteins were found after combination treatment than 5-aza-CdR treatment alone (Supplementary Fig. S12A and S12C).

We next quantified the levels of H3K9 methylation by Western blot in A2780 cells, and we found that both G9Ai and 5-aza-CdR significantly reduced global H3K9me1/2/3 levels, especially for H3K9me2, and the combination treatment further reduced the global levels of these repressive marks (Fig. 6C and D). Interestingly, global H3K9me3 levels were also reduced by the treatments in A2780 cells (Fig. 6C and D). Although the G9A/GLP complex does not catalyze trimethylation of H3K9, as the levels of H3K9me1/2 decreased, other lysine methyltransferase activities might have been reduced by depleting their substrates. Taken together, we observed a global reduction of chromatin-bound G9A protein and H3K9me1/2/3 levels by combination treatment compared with treatment with either compound alone, which might serve as the underlying mechanism for the synergistic effects of combination treatments in A2780 cells.

We also examined the H3K9me1/2 levels in OAW42 cells, in which no synergy was found between G9Ai and 5-aza-CdR. Interestingly, the results showed that 5-aza-CdR treatment alone efficiently reduced H3K9me1/2 levels to less than 20% of those in untreated cells (Supplementary Fig. S13A and S13B), compared with about 75% in A2780 cells (Fig. 6C and D). The combination treatment only slightly reduced the H3K9me1/2 levels compared with those after 5-aza-CdR treatment alone (Supplementary Fig. S13A and S13B). These data suggested that in OAW42 cells, 5-aza-CdR treatment is sufficient to remove most of the repressive H3K9me1/2 marks. The addition of G9Ai therefore might not be effective in further removing these marks globally. A search in our RNA-seq data revealed that the A2780 and CAOV3 cells showed higher levels of G9A/GLP expression, while in the cells that did not respond synergistically, lower levels of G9A/GLP transcripts were found (Supplementary Fig. S13C). These data suggested that the synergy between G9Ai and 5-aza-CdR might be, in part, associated with G9A/GLP expression levels.

The RNA-seq data revealed that a cluster of ERVs could be further upregulated by combination treatments, making it likely that the LTRs of these ERVs were suppressed by H3K9me1/2 in addition to DNA methylation. Although our results demonstrated strong decreases in global H3K9me2 by G9Ai or combination treatment (Fig. 6C and D), it was still not clear whether these changes also occur at ERV LTRs. We therefore performed ChIP assays, followed by quantitative PCR (ChIP-qPCR) to detect the chromatin modification status at the LTRs of HERV-Fc1 and MLT1N2, as two examples in A2780 cells. There was enrichment of the repressive mark H3K9me2 at LTRs of these ERVs compared with the promoters of β-actin (ACTB) (Fig. 6E). Indeed, the repressive H3K9me2 mark was dramatically reduced by G9Ai treatment alone. 5-Aza-CdR treatment resulted in increases, rather than decreases, of H3K9me2 levels at LTRs of these ERVs (Fig. 6E), suggesting an “epigenetic switch” to silencing the ERVs by histone modifications upon loss of DNA methylation (48, 50). H3K9me2 levels were further reduced by the combination treatment at LTRs of these ERVs, associated with an increase in active marks H3K4me3 and H3K27ac (Fig. 6E). Taken together, DNA demethylation by 5-aza-CdR and the additional chromatin alterations at ERV LTRs (decreasing H3K9me2 and increasing H3K4me3 and H3K27ac) induced by the combination treatment likely contributes to the synergistic upregulation of ERVs.

Our study shows that inhibiting both DNA methylation and the histone methyltransferase G9A synergistically induces antitumor effects and enhances the “viral mimicry” response in ovarian cancer cells with high levels of G9A/GLP expression. Importantly, we have shown that G9A was overexpressed in primary ovarian tumors, and was the highest among 33 TCGA cancer types. In addition, overexpression of G9A has been associated with poor prognosis of EOC (29). Therefore, a treatment strategy targeting both G9A and DNA methylation provides a promising treatment option because DNA methylation inhibitors have already been tested in clinical trials with promising outcomes (52, 53). Our dual inhibition approach suggests a potentially new epigenetic therapy combination that may serve as a novel therapeutic strategy for patients with ovarian cancer.

A key mechanism underlying the synergistic effects of G9Ai and 5-aza-CdR on ERV upregulation may involve the “epigenetic switch,” in which a group of ERVs were repressed by G9A upon loss of DNA methylation. This is in line with recent work from our laboratories that there is a gain in repressive marks such as H3K9me2/3 or H3K27me3 after DNA demethylation to maintain the silencing of some ERV LTRs (48–50). Because H3K9me2 levels at LTRs of HERV-Fc1 and MLT1N2 increased after 5-aza-CdR treatment, this histone mark may well be responsible for repressing these ERVs. This is different from the situation at promoters of protein coding genes where 5-aza-CdR effectively reduced H3K9me2 levels and depleted G9A (31, 32). Therefore, the addition of G9Ai after 5-aza-CdR treatment could enhance the expression of these ERVs by further reducing the repressive H3K9me2 marks. It is worth noting that the “epigenetic switch” might be dependent on the expression levels of G9A/GLP or other silencing complexes such as the polycomb repressive complex 2 (PRC2; ref. 50). In addition, different responses to DNA demethylation by 5-aza-CdR might also contribute to the variable synergistic effects observed in these cell lines after the combination treatment.

An interesting finding of this study is that G9A and DNA methylation appears to repress distinct sets of ERVs. This is consistent with our recent finding that DNA methylation tends to predominate in silencing the evolutionarily younger ERVs, which have higher CpG densities, while histone methyltransferases target mainly the middle-aged ERVs with less CpG density (48). Therefore, the combination treatment results in upregulation of more ERVs that include those repressed by G9A or DNA methylation alone. For evolutionary “young” ERVs such as HERV-Fc1 and MLTN1, DNA methylation plays a major silencing role, and demethylation by 5-aza-CdR can induce their expression while G9Ai treatment alone is not sufficient to upregulate these ERVs. However, they were still repressed by G9A after 5-aza-CdR treatment (termed as the “epigenetic switch” mentioned above). Combination treatment simultaneously removed both repressive mechanisms, resulting in further upregulation of these ERVs. In this way, combination treatment with the two inhibitors synergistically strengthens the “viral mimicry” response and leads to increased antitumor effects.

Our findings also show that not all ERVs were equal with respect to their abilities to activate the viral defense pathway. In CAOV3 cells, where G9Ai treatment alone upregulated equivalent numbers of ERVs as 5-aza-CdR (Fig. 4A–C), the levels of induced viral defense genes were lower than those upregulated by 5-aza-CdR treatment (Fig. 3C). This observation reinforces our previous findings that the intermediate-age ERVs silenced mainly by histone methyltransferases did not elicit a strong antiviral response (48). The strength of the viral mimicry response may be determined by the expression of evolutionary “younger” ERVs silenced by DNA methylation. Further studies are needed to characterize the effects of specific ERVs in inducing the antiviral response, which may be crucial for patient responses to the combination epigenetic therapy.

Our results also suggest that G9A plays different roles in various tissues or cell types. For example, it has been shown that G9A is essential in repressing developmental genes at euchromatic regions during early embryogenesis (54). In breast cancer cells, G9A has been shown to be responsible for aberrant silencing of tumor suppressor genes (30). In our study, G9Ai treatment alone did not induce upregulation of tumor suppressor genes. In fact, very few genes were upregulated; instead, a subset of ERVs were upregulated by G9Ai treatment alone in some ovarian cancer cells, highlighting the important role of G9A in suppressing ERVs in ovarian cancer cells.

UNC0638 was designed based on the structure of G9Ai BIX01294 with improved in vitro potency and increased cell membrane permeability (46). UNC0638 exhibited high potency and specificity for inhibiting G9A and GLP with very low off-target toxicities (46). However, its potential for clinical application is limited by poor pharmacokinetics in vivo. The development of next generation G9Ai with comparable potency is needed to improve in vivo pharmacokinetics. Our current preclinical study has provided a rationale for the future clinical application using the combination of inhibitors for both G9A and DNMTs. We predict that this combination would be very helpful for those cancer types that exhibit overexpression of G9A or GLP proteins, especially for those cancers with poor prognosis correlated with higher G9A expression levels, such as ovarian cancer. Our results are limited to the four cell lines tested so that the generality of the results need to be replicated in a larger set. However, they are consistent with recent data we have obtained with knockdowns of histone methyltransferases and point to the need for patient stratification based on the levels of G9A/GLP before combination treatment is initiated.

DNA methylation and the histone modifications H3K9me1/2 are both important for silencing ERVs. Dual inhibition of these processes results in synergistic upregulation of ERVs and induces an antiviral response in some but not all cell lines, serving as a basis for exploring this novel combination treatment in the clinic, especially for patients with ovarian cancer with high levels of H3K9me1/2 histone methyltransferases G9A/GLP expression.

P.A. Jones is a consultant/advisory board member for Zymo Corporation. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Liu, S.L. Thomas, A.K. DeWitt, G. Liang, P. A. Jones

Development of methodology: M. Liu, S.L. Thomas, H. Ohtani

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Liu, S.L. Thomas, A.K. DeWitt

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Liu, S.L. Thomas, A.K. DeWitt, W. Zhou, Z.B. Madaj, G. Liang

Writing, review, and/or revision of the manuscript: M. Liu, S.L. Thomas, A.K. DeWitt, Z.B. Madaj, S.B. Baylin, G. Liang, P. A. Jones

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Liu, S.L. Thomas, A.K. DeWitt, W. Zhou

Study supervision: G. Liang, P. A. Jones

Other (planning of experiments): S.B. Baylin

We thank VARI Core Technologies and Services, including the Flow Cytometry Core, Genomics Core, Bioinformatics and Biostatistics Core, for technical assistance. In particular, we thank the Flow Cytometry Core manager, Rachael Sheridan, PhD, for providing training on the CytoFLEX S analyzer and data analysis. We also thank David Nadziejka (VARI) for technical editing of the article.

This work was supported by R35CA209859 from the NCI to P.A. Jones and G. Liang, by W81XWH14-1-0385 from the U.S. Department of Defense (DoD) to S.B. Baylin and P.A. Jones, and by the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation to S.B. Baylin. Research funding was provided by Van Andel Research Institute through the VARI-SU2C Cancer Epigenetics Dream Team (SU2C-AACR-DT-14-14). Stand Up to Cancer is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.