Selective DNA demethylation by fusion of TDG with a sequence-specific DNA-binding domain

Abstract

Our ability to selectively manipulate gene expression by epigenetic means is limited, as there is no approach for targeted reactivation of epigenetically silenced genes, in contrast to what is available for selective gene silencing. We aimed to develop a tool for selective transcriptional activation by DNA demethylation. Here we present evidence that direct targeting of thymine-DNA-glycosylase (TDG) to specific sequences in the DNA can result in local DNA demethylation at potential regulatory sequences and lead to enhanced gene induction. When TDG was fused to a well-characterized DNA-binding domain [the Rel-homology domain (RHD) of NFκB], we observed decreased DNA methylation and increased transcriptional response to unrelated stimulus of inducible nitric oxide synthase (NOS2). The effect was not seen for control genes lacking either RHD-binding sites or high levels of methylation, nor in control mock-transduced cells. Specific reactivation of epigenetically silenced genes may thus be achievable by this approach, which provides a broadly useful strategy to further our exploration of biological mechanisms and to improve control over the epigenome.

Introduction

DNA methylation is a key player in epigenetic silencing of transcription.¹ Our ability to selectively manipulate epigenetic status of functionally important genes is limited to suppressive methods, including RNA interference and targeted DNA-methyltransferases.² Mechanisms of DNA demethylation remain elusive, with postulates ranging from passive to enzymatic processes, as reviewed in ³. This contributes to our inability to create selective and/or gene-specific instruments to overcome silencing of genes switched off epigenetically. We aimed to develop a tool for selective demethylation of regulatory sequences in the DNA. Such a tool would provide a much needed instrument for causality demonstration in epigenetic studies: e.g., in asthma, dendritic⁴ or other immune cells⁵ show aberrant hypermethylation, but the causal significance of these changes cannot be determined. In the longer-term, it could be useful for cell transdifferentiation, or for immunotherapeutic purposes.

Several putative demethylases have been proposed, including thymine-DNA-glycosylase (TDG).^6-9 We developed a molecular construct in which TDG is targeted to specific DNA sequences to determine whether expression of this construct can lead to 1) local DNA demethylation of key regulatory sequences (e.g., promoter and nearest CpG island) of the target and, 2) enhanced gene induction in a mammalian cell.

In the fusion protein we designed TDG is connected via a short glycine-rich linker to an 'anchor' domain. There is a selection of proteins able to bind DNA sequence-specifically, e.g., many transcription factors use zinc-fingers or similar structures to recognize the DNA.¹⁰ As a proof of principle, we chose naturally occurring rel-homology domain (RHD), a binding domain of NFκB, as the anchor. This domain provides a good selection of well-described NFκB-dependent and -independent genes for test and control purposes.¹⁰ The plasmid also encoded LNGFR for positive selection and other necessary machinery for lentiviral packaging. We transduced 3T3 cells and purified successful transductants using paramagnetic anti-LNGFR-coated nanoparticles. Controls included constructs encoding RHD alone, TDG alone, and mock-transduced cells.

Results and Discussion

Transcriptional screening indicates enhanced induction of message and protein in cells expressing fusion TDG

Promoters containing NFκB responsive elements are often also responsive to IFNγ signaling through STAT1. We anticipated that promoters containing both STAT1 and NF-κB sites would show increased sensitivity to IFNγ in cells expressing our fusion protein (Fig. 1), while genes with only STAT1 sites would respond equally in all groups (Fig. 2). Inducible nitric oxide synthase (Nos2) gene contains both sites; furthermore, the promoter and CpG island are highly methylated in 3T3 cells and the gene is largely silenced,¹² making it a good candidate for de-repression by targeted demethylation. While control groups revealed minimal response to IFNγ stimulus, RHD-TDG expressing cells showed a remarkable 3-fold increased induction (Fig. 3). The response was low at 1, 2 and 4 h, became pronounced at 6 h and remained high at 24 h (curves not shown). ELISA of cell lysates at 6 h demonstrated increased NOS2 protein production (Fig. S1). A similar but less dramatic response was obtained from another gene containing sequences for both NFκB and STAT1 binding, a chemokine Cxcl11 (Fig. S2). The limitation of this target was variability in background expression between the groups; additionally, RHD itself appears to affect expression of Cxcl11. However, this did not preclude observation of increased fold upregulation in the RHD-TDG samples, as expected.

Figure 1. Schematic of the hypothesis: demethylation of CpGs in regulatory sequences in the vicinity of NFκB binding sites by RHD-TDG construct but not by control constructs (e.g., RHD alone) would lead to increased transcription of the target gene. An unrelated stimulus, e.g., STAT1 induced by interferon, would produce stronger upregulation of Nos2 over background, if the promoter is demethylated.

Figure 2. Choice of target genes. Effect was expected in primary target Nos2, secondary target Cxcl11 and auxiliary target LINE-1, but not in control genes. Orange blocks denote NFκB binding sequences, blue blocks denote STAT1 binding sequences; yellow underscores indicate sites interrogated by pyrosequencing. N values represent number of PCR sample measurements.

Figure 3. RHD-TDG increases transcriptional response to IFNγ for Nos2, but not for control genes. *p < 0.01 against RHD control.

Transcription of control genes remained unaffected by the construct

If the increased induction of Nos2 is indeed due to DNA demethylation by TDG, we would expect RHD-TDG to have little or no effect on promoters that are constitutively demethylated, regardless of IFNγ inducibility. Indeed, expression of interferon response factor 1 (Irf1), where the average cytosine methylation in the promoter is only around 3–5% (data not shown) was not significantly affected by the construct, despite both the presence of RHD and STAT1 sites, and healthy induction by IFNγ (Fig. 2). This observation is particularly important since IRF1 is capable of mediating indirect effects of IFNγ on Nos2.¹³ Another control gene, interferon-inducible protein 35 (Ifi35), which has STAT1 sites but does not have NFκB sites, provided a similar strong response to IFNγ independent of the construct (Fig. 3), supporting the model that our construct must bind the promoter in order to affect expression. Finally, a control gene with NFκB but without STAT1, prostaglandin-endoperoxide synthase 2 (Ptgs2), did not respond to IFNγ in any of the groups, suggesting that RHD-TDG does not affect basal expression, but sensitizes the promoter to induction by a specific stimulus. Expression of the construct was not affected by IFNγ stimulation (Fig. S3); purity and viability were equal in all study groups (Fig. S4).

Cells expressing fusion TDG demonstrate decreased DNA methylation in key regulatory sequences of Nos2 and in targeted non-coding elements

To determine whether the increased expression of Nos2 results from DNA demethylation by RHD-TDG, we measured cytosine methylation in the promoter and nearest CpG island by pyrosequencing. We observed statistically significant hypomethylation of a number of sites following RHD-TDG expression (Fig. 4). Interestingly, even a 5–10% reduction in methylation at these sites was associated with a strong transcriptional response, which emphasizes the potential of DNA demethylation for manipulation of gene expression. We have also evaluated methylation in non-coding areas containing NFκB sites, namely the long interspersed nucleotide elements (LINE-1), which has become a customary target in methylation studies because the multiple repeats provide a better signal in the assay. The particular fragment included 5 methylated CpG’s in vicinity of 3 NFκB consensus binding sites, making it a convenient target. RHD-TDG cells had decreased methylation in 4 out of 5 CpG’s (Fig. 3). An interesting topic of further mechanistic studies may be the fact that not all of the CpG's tested revealed decrease in methylation, suggesting that within a targeted fragment there may be stoichiometric or stereologic differences.

Figure 4. RHD-TDG decreases DNA methylation in vicinity of NF-kB binding sites in Nos2 promoter, Nos2 CpG island and in 4 out of 5 tested CpG sites in LINE-1. *p < 0.05 against RHD control.

The mechanism by which TDG achieves cytosine demethylation remains controversial, in particular regarding the role of base-excision repair.^14,15 In this present study, we did not aim to resolve these conflicting models. Instead we propose a practical application of this enzyme, which seems to attract more and more interest in the field, especially given the recent development of a knockout mouse.⁹ Although our RHD-based construct was not aimed to provide a single gene specificity, it does allow proof-of-principle for the utility of DNA-binding domain targeted TDG in overcoming epigenetic silencing. While relying on intrinsic transcription factor binding sites is one option, recent studies describing the design of synthetic, zinc-finger based DNA binding domains^16-18 suggest that high selectivity may be attainable in DNA demethylation, given our demonstration that it is practical to target the TDG to chosen sequences in specific promoters for gene function control.

In summary, murine cells transduced with TDG fused to a sequence-specific DNA binding domain RHD demonstrate decreased methylation of cytosines in regulatory sequences in the vicinity of RHD-binding sites, which was associated with increased transcriptional upregulation of target genes upon unrelated stimulus. Transcription was unaffected for control genes and was minimally affected by control constructs. We demonstrate that direct targeting of this enzyme to specific DNA loci is a promising approach for future mechanistic studies in epigenetics and, more broadly, opens the ability to manipulate the epigenetic status of selected genes in a bidirectional manner.

Materials and Methods

We cloned the Rel Homology Domain of human NFκB p65RelA (residues 1–308) into a derivative of pHAGE-CMV-Ipr1-UBC-LNGFR-W ¹⁹ by PCR cloning into unique XbaI and NotI sites. A poly-glycine floppy linker was added at the 3′ end using synthetic oligonucleotides (sequences GGCCAAGGTGGAGGTGGCTCAGGCGGAGGTCCCGGTGGAGGC (top) and GGCCGCCTCCACCGGGACCTCCGCCTGAGCCACCTCCACCTT (bottom)) to produce pHAGE-RHDg. Two isoforms of murine TDG were cloned from C57/BL6 cDNA, with addition of 5′ NotI and 3′BamHI sites by high-fidelity PCR, and inserted into pHAGE-RHDg. One isoform showed substantially lower activity than the other, therefore serving as internal control (data not shown). The construct targets RHD binding sites in the DNA. Information regarding consensus NFκB and STAT1 binding sequences was obtained from Transfac Database. Figure 1B highlights our choices of target genes for experimental and control purposes. To make the non-targeted constructs, the RHDg moiety was removed from pHAGE-RHDgTDGv1 and pHAGE-RHDgTDGv2 by digestion with XbaI and NotI, blunting with Klenow, and re-ligation. As control for a transduction and transgene expression, mCherry was inserted into a derivative pHAGE-CMV-Ipr1-UBC-LNGFR-W in which the Ipr1 CDS is replaced with an expanded polylinker.

Packaging of lentivirus into 293T cells, transduction into NIH-3T3 cells and selection using paramagnetic anti-LNGFR antibody coated nanoparticles (Miltenyi) were performed as described previously.¹⁹

Purified 3T3 cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with fetal bovine serum, glutamine, penicillin and streptomycin (VWR) in tissue-culture treated 6-well plates (BD). We stimulated the cells with 250 μg/ml of murine recombinant IFNγ (Peprotech) for 1, 2, 4, 6, and 24 h. We harvested the cells by tripsinisation followed by immediate isolation of RNA and DNA using Qiagen RNEasy and DNEasy kits, in complete adherence to the manufacturer’s instructions. RNA was reverse transcribed into cDNA using iScript cDNA Synthesis Kit (Bio-Rad) and amplified in a CFX96 real-time PCR system (Bio-Rad) with respective primers (IDT DNA), see Table S1 for sequences. Depending on the type of polymerase and dye in either the SYBR Green supermix or EvaGreen SsoFast supermix, a typical protocol included a hot start of 95°C for 1–5 min, 35–40 cycles of amplification of 95°C, 55–60°C and 72°C, and a melt curve from 65°C to 95°C in 1 sec. increments. Expression was normalized to β-actin, and fold upregulation was calculated by normalizing IFNγ-stimulated expression to background (delta delta Ct). DNA (500 ng) was converted in a bisulfite reaction using Qiagen EpiTect kit following the manufacturer recommended procedure. PCR amplification with respective primers (IDT DNA, see Table S1 for sequences) continued for 40–49 cycles, depending on the target. PCR product was verified by gel electrophoresis. Remaining product was loaded into a pyrosequencing reaction and analyzed using Pyromark Q96 MD pyrosequencer (Qiagen) in compliance with the manufacturer’s recommended procedure. Some samples were analyzed blindly at our request by a commercial pyrosequencing service (EpigenDx).

We performed ELISA measurements of NOS2 concentration in cell lysates using Quantikine iNOS ELISA kit (R&D).

We performed flow cytometry using FacsCanto II cytometer (BD Biosciences). We used anti-LNGFR antibody (Miltenyi) to confirm success of transduction and selection, and a cell viability kit (Invitrogen) with 7-AAD and Annexin V stains for necrotic and apoptotic cells (Fig S1D).

Experiments were repeated 3 times. In each of the biological repeats, each RNA sample was analyzed at least in triplicates; each DNA and protein sample was analyzed at least in duplicates. Pyrosequencing assays were repeated 3 times. Real-time PCR assays were repeated several times depending on the conditions and primers, contributing to variable number of measurements, reflected in Figure 2. Data were analyzed in Prism 5.0 (GraphPad Software, Inc.) using ANOVA with post-tests.