Cytosine methylation of mitochondrial DNA at CpG sequences impacts Transcription Factor A DNA binding and transcription

Abstract

In eukaryotes, cytosine methylation of nuclear DNA at CpG sequences (^5mCpG) regulates epigenetic inheritance through alterations in chromatin structure. However, mitochondria lack nucleosomal chromatin, therefore the molecular mechanisms by which ^5mCpG influences mitochondria must be different and are as yet unknown. Mitochondrial Transcription Factor A (TFAM) is both the primary DNA-compacting protein in the mitochondrial DNA (mtDNA) nucleoid and a transcription-initiation factor. TFAM must encounter hundreds of CpGs in mtDNA, so the occurrence of ^5mCpG has the potential to impact TFAM-DNA recognition. We used biophysical approaches to determine whether ^5mCpG alters any TFAM-dependent activities. ^5mCpG in the heavy strand promoter (HSP1) increased the binding affinity of TFAM and induced TFAM multimerization with increased cooperativity compared to nonmethylated DNA. However, ^5mCpG had no apparent effect on TFAM-dependent DNA compaction. Additionally, ^5mCpG had a clear and context-dependent effect on transcription initiating from the three mitochondrial promoters. Taken together, our findings demonstrate that ^5mCpG in the mitochondrial promoter region does impact TFAM-dependent activities in vitro.

INTRODUCTION

Reversible methylation at CpG sequences (^5mCpG) plays a crucial role in regulating nuclear gene expression via a variety of mechanisms, but the role ^5mCpG plays in mitochondria remains unclear. The existence of reversible methylation of mitochondrial DNA (mtDNA) has been a highly debated topic for five decades. In recent models of cancer, neurodegenerative diseases, diabetes and aging, reports consistently describe modest percentages of ^5mCpG (between 2–37%), in the mtDNA, in conjunction with altered levels of mitochondrial RNA transcripts and or mtDNA copy number [1–8]. These and other studies using bisulfite sequencing, restriction enzyme digestion, chromatographic analyses, ELISA and Methyl-DNA-immunoprecipitation support the existence of low levels of ^5mCpG in mtDNA (reviewed in [9–11]). However, other studies have raised concerns about the validity of the bisulfite sequencing approach, as only low levels of ^5mCpG were detected in mtDNA [12–16]. Yet the same studies that also used approaches other than bisulfite sequencing generally found low levels of ^5mCpG (between 2–10 %) [13, 16], which correspond to the levels of ^5mCpG in the literature supporting the existence of ^5mCpG in mtDNA. Additional evidence in support of mtDNA methylation comes from the identification of human DNA methyltransferases, DNMT1 [17], DNMT3A [18], DNMT3B [19], and human cytosine demethylation enzymes, including members of the Ten-Eleven Translocation (TET) family of methylcytosine dioxygenases, TET1 and TET2 [7, 20], in mitochondria or associated with the mitochondrial outer membrane. Thus, there is the need to consider the effects of ^5mCpG in the DNA-dependent processes of mtDNA.

CpG methylation in the nucleus participates in the formation of condensed nucleosomal arrays that create heterochromatin states that are important during development and epigenetic inheritance [21]. CpG methylation also has varied effects on direct protein-DNA recognition [22], with protein-DNA interactions that are either inhibited, promoted or unaffected [23]. Unlike nuclear chromatin, the mitochondria lack nucleosomes. Instead, the mitochondrial Transcription Factor A (TFAM, mtTFA) is the primary protective protein component of the mitochondrial nucleoid complex [24–30]. Additionally, TFAM is important for mtDNA maintenance [31], the mtDNA copy number [32, 33], as well as for coordinating mtDNA transcription with mtDNA replication (reviewed in [34]). In its role as a transcription factor, TFAM binds to each of three promoters, heavy strand promoters 1 and 2 (HSP1, 2) and the light strand promoter (LSP), which are located in the noncoding control region of the mtDNA known as the displacement loop (D-loop) (reviewed in [24]).

TFAM binds site-specifically to the mtDNA promoters and non-sequence specifically to the mtDNA, through interactions with the DNA minor groove, to produce a stable 180° bend, “U-turn” structure in the DNA (reviewed in [35]). As a transcription factor, TFAM binds to the three promoters LSP, HSP1 and HSP2, referred to here as the tri-promoter region (Figure 1), and aids in recruitment of the bacteriophage-related RNA polymerase (POLRMT) and the rRNA methyltransferase-related transcription factor 2 (TFB2M) to initiate transcription [11, 36]. TFB1M is an rRNA dimethyltransferase and has a primary role in modulating translation [37–39]. There are 435 CpG sites within the mtDNA including 5 CpGs in the tri-promoter region (Figure 1). ^5mCpG levels in the mtDNA, including the promoter region of the D-loop have been reported to be as little as >2% and as much 35% in various human cells lines and tissue samples [20, 40, 41]. Moreover, ^5mCpG has been observed in the promoter region, including in the HSP1 TFAM binding site as well the LSP and HSP2 promoters [8, 20, 42].

Figure 1. Schematic diagram of the mitochondrial genome (mtDNA).

The circular genome illustrates the heavy strand and light strand gene products for transcripts originating from the LSP, HSP1 and HSP2 (colored in green, gold and blue, respectively). The tRNA genes are shown as black sticks. Promoters, origins of replication, consensus sequence blocks (CSBs) and termination-associated sequences (TAS) are shown in the expanded D-loop region. The expanded promoter region details the LSP (green), HSP1 (gold) and HSP2 (blue) promoters highlighting the interactions with TFAM, TFB2M (B2) and POLRMT (POL). Reported ^5mCpG sites within the promoter region are shown as red dots.

Due to the extensive direct DNA interactions of TFAM throughout the mtDNA, we hypothesized that ^5mCpG would directly impact TFAM-dependent molecular processes. To test this hypothesis, we investigated the effects of ^5mCpG in mtDNA on the DNA binding and transcription-initiation properties of TFAM in vitro. Using fluorescence resonance energy transfer (FRET) assays and electrophoretic mobility shift assays (EMSA), we observed increased affinity and cooperativity of TFAM for the ^5mCpG-HSP1 DNA. ^5mCpG did not significantly influence the ability of TFAM to compact plasmid-length DNA. However, run-off transcription assays revealed altered levels of transcription initiation for both the tri-promoter template and LSP-only templates. Thus, ^5mCpG has effects on TFAM DNA binding and transcription activities, but not DNA compaction, indicating that it impacts site-dependent binding properties of TFAM.

MATERIALS AND METHODS

Preparation of TFAM and DNA

Recombinant human TFAM (aa 42–246) was purified as previously described (Supplementary Figure S1A) [43].

Synthetic DNA was purchased from Integrated DNA Technologies (IDT) (See Supplementary Materials and Methods for sequences). EMSA experiments used HSP1 33 base pair (bp) DNA and 33 bp non-site specific (NS) DNA. For FRET assays HSP1 DNA included donor or acceptor fluorophores at the 3′ end. For the LSP run-off transcription experiments, an 88 bp template (LSP88) comprised the DNA sequence 390–477 from human mtDNA. The oligonucleotides were synthesized with and without the ^5mCpG modification (IDT). Each oligonucleotide was purified using C18 Sep-Pak cartridges (Waters). To produce the 33 bp and 88 bp dsDNA fragments, with and without ^5mCpG, the complementary ssDNA strands were annealed by heating 95°C for 5 minutes and then cooling slowly to room temperature. The DNA was then purified using a DEAE anion exchange column (Tosoh Bioscience) with a gradient of 0 to 1 M NaCl in buffer (50 mM Tris-HCl (pH8.0) and 1 mM EDTA). The purity of each fraction was confirmed on nondenaturing gels (Supplementary Figure S1B–D).

For the compaction assays (EMSA and EM), the pGEMPTEZLSP3 plasmid (pLSP3) containing human mitochondrial DNA sequence 242–825 [44] was expressed in Stellar Competent dam-/dcm- cells purchased from Clontech. The plasmid was methylated at the 5-position of cytosine of CpG sites using M.SssI methyltransferase (New England Biolabs) following the manufacturers protocol. The pLSP3 and ^5mCpG pLSP3 plasmids were then linearized using NdeI restriction enzyme (New England Biotech). The DNA was then purified by phenol and ether extractions followed by ethanol precipitation. The efficiency of CpG methylation was verified by digestion of pLSP3 with restriction enzymes HpaII and MspI (Supplementary Figure S2A).

For the run-off transcription assays, the pLSP3 plasmid was digested with EcoRI-HF (New England Biolabs) to generate the 603 bp “tri-promoter” template (human mtDNA sequence 242–825). The DNA was treated as above to generate the ^5mCpG 603 bp template. The 603 bp DNA fragments were then gel purified and treated as above. The efficiency of ^5mCpG incorporation was verified as above (Supplementary Figure S2B).

Fluorescence resonance energy transfer assays

FRET assays were performed essentially as previously described [45]. Briefly, we used a Horiba Fluorolog 3 spectrofluorimeter that was thermostated to 20 °C. Experiments were conducted with the fluorophore-labeled HSP1 and ^5mCpG-HSP1 DNA each at a concentration of 3.4 nM in FRET Buffer (50 mM HEPES-Na, pH 7.4, 150 mM NaCl, and 1 mM DTT). TFAM was added to the DNA samples at final concentrations of 0 nM, 0.63 nM, 1.25 nM, 1.88 nM, 2.5 nM, 3.13 nM, 3.75 nM, 4.38 nM, 5.00 nM, 5.63 nM, 6.25 nM, 6.88 nM, 7.50 nM, 8.13 nM, 8.75 nM, 10.0 nM, 12.5 nM, 15.0 nM, 20.0 nM, 25.0 nM, 30.0 nM, 35.0 nM, 40.0 nM, 45.0 nM and 50.0 nM. To measure the FRET induced by protein binding to the fluorophore-labeled DNA, we employed the enhanced sensitization of the acceptor method described in detail by Clegg [46]. The FRET effect (FE) = I490/I560, where I490 is the fluorescence emission of TAMRA at 580 nm when excited at 490 nm, and I560 is the fluorescence emission of TAMRA at 580 nm when excited at 560 nm. A plot of normalized FE vs. [TFAM] was used to calculate the TFAM-DNA apparent dissociation constants (K_Dapp) using the ligand-depletion binding model (Equation 1), where DNA is the DNA concentration and x is the total TFAM concentration:

$$Y=\left(\mathit{\text{Bmax}}\right)*\left(\frac{\left(DNA+x+\text{KD}\right)-\surd \left(\left({\left(DNA+x+\text{KD}\right)}^2\right)-\left(4*x*DNA\right)\right)}{2*DNA}\right)$$ Equation (1)

Electrophoretic mobility shift assays

Quantitative EMSA was performed using the HSP1, ^5mCpG-HSP1, NS and ^5mCpG-NS DNA, described above. TFAM was added to the DNA samples (10 nM) in Binding Buffer (20 mM HEPES-Na (pH 7.5), 50 mM KCl, 2 mM MgCl₂ and 2 μg/ml BSA) to final concentrations of 0 nM 1.0 nM, 2.0 nM, 3.0 nM, 4.0 nM, 5.0 nM, 6.0 nM, 7.0 nM, 8.0 nM, 9.0 nM, 10.0 nM, 15.0 nM, 20.0 nM, 25.0 nM and 30.0 nM and incubated for 45 minutes on ice. The EMSA used nondenaturing PAGE (8% acrylamide; 37.5:1 acrylamide:N,N-methylene-bis-acrylamide (BioRad)) with 1X TBE buffer (100 mM Tris, 100 mM boric acid, 2.5 mM EDTA). The gels were prerun at 70 V for 30 min prior to sample loading and then electrophoresed at 70 V for 1 h. The DNA was visualized using Sybr Green I dye (Invitrogen) on a Typhoon 9400 Imager (Molecular Dynamics). The intensity of each band was measured using ImageQuant Software, and after background subtraction, the fraction of DNA bound was calculated, fitted with the Hill equation (Equation 1) and graphed using Prism software.

Stoichiometric EMSA was performed using 100 nM 33 bp DNA (HSP1, ^5mCpG-HSP1, NS or ^5mCpG-NS DNA) and a range of concentrations of TFAM (0 nM, 50 nM, 100 nM, 200 nM, 300 nM, 400 nM, 500 nM, 600 nM, 700 nM and 800 nM). The reactions were incubated on ice for 45 minutes and then loaded onto a pre-run 8% nondenaturing polyacrylamide gel (37.5:1, acrylamide:N,N-methylene-bis-acrylamide, (BioRad)) and electrophoresed for 45 minutes on ice. The DNA was visualized using Sybr Green I dye (Invitrogen) on a Typhoon 8400 Imager (Molecular Dynamics). The intensity of each band was measured using ImageQuant Software, and after background subtraction, the fraction bound to DNA was calculated and graphed using Prism software.

Electron microscopy and DNA compaction assays

Compaction reactions were conducted with 100 nM linearized pLSP3 or ^5mCpG-pLSP3 DNA. The 10 μL reactions in Binding Buffer contained TFAM at concentrations (0 nM, 60 nM, 90 nM, 120 nM or 240 nM) and were incubated at room temperature for 30 minutes. Each sample was split for use both in EMSA and EM. For electron microscopy, each sample was diluted 1:1 with 4 mM Mg-acetate and applied to glow discharged EM grids. Samples were then washed with 4 mM Mg-acetate before staining with 0.02% (wt/vol) uranyl acetate. Grids were imaged on a JEOL 1230 (or JEOL 1200) transmission electron microscope with a Gatan- digital camera at a magnification of 120,000x. For the compaction assays, each sample was electrophoresed on a 0.7 % agarose gel and then visualized using Sybr Green I dye (Invitrogen) on a Typhoon 9400 Imager (Molecular Dynamics). The migration distance of each band was measured using ImageJ software. These values were analyzed by subtracting the migration distance of the free DNA from those of the TFAM:DNA complexes.

In vitro run-off transcription assays

Transcription reactions were performed essentially as in [44, 45, 47]. The samples contained the tri-promoter 603 bp template DNA (3.4 nM) in Reaction Buffer (10 mM Tris-Cl pH8.0, 10 mM MgCl₂, 100 μM DTT, 100 μg/mL BSA), 400 μM ATP, 150 μM CTP, 150 μM GTP, 10 μM UTP, 0.2 μM ³²P UTP and 4 units of RNAaseOut (Invitrogen). The final concentration of NaCl in the transcription reactions is approximately 50 mM, due to the presence of NaCl in the stock solutions of the DNA, NTPs, TFAM, TFB2M, POLRMT. This concentration is in the optimal range for achieving TFAM-dependent mtDNA transcription [48, 49]. The reactions were incubated with 16 nM POLRMT (Enzymax, LLC), 16 nM TFB2M (Enzymax, LLC) and varying concentrations of TFAM (0 nM, 6.8 nM, 16 nM, 34 nM, 51 nM, 68 nM or 136 nM) at 32 °C for 3 hours. All reactions were stopped with the addition of 100 μL 10 mM Tris-HCl pH 8.0, 0.2 M NaCl, 0.5% SDS and 22.5 μg proteinase K. Transcription reactions were then ethanol precipitated and resuspended in 98% formamide, 10 mM EDTA pH 8.0, 0.025% xylene cyanol and 0.025% bromophenol blue. Samples were electrophoresed on an 8% polyacrylamide-7M urea gel. Visualization of transcripts was performed by phosphorimaging using the Typhoon 9400 Imager (Molecular Dynamics). An RNA ladder (RiboRuler Low Range, Thermo Fisher) was used to verify transcription sizes (270 bp for the HSP1, 173 bp for LSP and 186 bp for HSP2). For an internal loading control, the DNA template was quantitated from SybrGreen II-staining of the same gel. Loading was consistent and essentially no corrections were needed. The analyses of the data are described in the Supplementary Materials. Two-way ANOVA analysis was used to determine the statistical significance of the effect of ^5mCpG on transcription.

Transcription reactions for the LSP88 template used the same concentrations and reaction conditions as the tri-promoter reactions. Samples were electrophoresed on a 20% polyacrylamide-7M urea gel. An RNA ladder (Decade Marker, Thermo Fisher) was used to verify transcript sizes. The transcripts were visualized by phosphorimaging as above. They were quantitated and scaled as described in the Supplementary Materials.

Analyses of published structures

The crystal structures of both TFAM bound to the HSP1 (PDB ID 4NOD) [50] and the complex containing the LSP DNA, TFAM, TFB2M and POLRMT (PDB ID 6ERP) [51] were used to map the HSP1 CpG sites. The models was generated using the programs PyMOL [52] for alignments and coloration and Photoshop (Adobe) for final presentation.

RESULTS

^5mCpG methylation of the HSP1 increases the DNA binding affinity of TFAM

^5mCpG has been observed throughout the mitochondrial genome (mtDNA), including in the D-loop, which includes the sequences that direct transcriptional initiation (Figure 1) [1]. The 1,124 nt D-loop contains the three mitochondrial promoters, LSP, HSP1 and HSP2 (Figure 1). HSP1 directs the transcription of the 12S and 16S rRNAs and HSP2 directs all of the protein-encoding genes except for ND6 and 5 tRNA genes, which are transcribed from the LSP [24, 53]. Cytosine methylation was known to directly inhibit [54] and alter the specificity of transcription factor DNA binding [55]. Methylation of both of the CpG sites within the HSP1 TFAM binding site (Figures 1 and 2A) has the potential to influence TFAM-DNA binding.

Figure 2. Binding properties of TFAM to HSP1 and ^5mCpG-HSP1 DNA.

(A) The HSP1 sequence with the ^5mCpG highlighted. Representative fluorescence emission spectra from FRET experiments with 3.4 nM FAM/TAMRA labeled HSP1 DNA and 3.4 nM FAM/TAMRA labeled ^5mCpG-HSP1 DNA with added TFAM (0–50 nM). (C) Binding curves of normalized FRET effect were used to measure the K_Dapp values (n=7; error bars = ± S.D.)

To determine whether the DNA binding affinity of TFAM for HSP1 DNA was altered by the presence of ^5mCpG, we used a fluorescence resonance energy transfer (FRET) assay. The 33 bp DNA fragments of non-methylated HSP1 DNA and ^5mCpG-HSP1 were each labeled with a fluorescein (FAM) donor fluorophore on the 3′ end of one strand and a tetramethylrhodamine (TAMRA) acceptor fluorophore on the other strand [43, 45]. As TFAM was titrated into these samples (Figure 2A), the donor signal at 520 nm decreased and the acceptor signal at 580 nm increased. This strong FRET effect was due to the TFAM-induced bending of the DNA, which reduces the distance between the fluorophores (Figure 2, B). From these data, the apparent binding affinity (K_Dapp) of TFAM for HSP1 was found to be 3.9 ± 0.3 nM and for ^5mCpG-HSP1 it was 1.2 ± 0.3 nM (Figure 2C), revealing a 3-fold increase in the affinity caused by the presence ^5mCpG (Cohen′s d-value of 0.85, indicates a medium effect size). The magnitude of the FRET effect at full protein-DNA occupancy was similar for both templates, indicating no major change in the final DNA end-to-end distance.

Next, we carried out electrophoretic mobility shift assays (EMSA) using the same DNA sequence, but without fluorophore labeling. With addition of increasing concentrations of TFAM, distinct shifted bands were observed for both HSP1 and ^5mCpG-HSP1 DNA (Supplementary Figure S3A and B). The data were quantitated and the binding affinities (K_Dapp) were 5.7 ± 1.8 nM and 2.4 ± 0.7 nM, for the HSP1 and ^5mCpG-HSP1 DNA, respectively (Supplementary Figure S3C). This 2-fold difference is statistically significant and has a Cohen′s d-value of 0.67, indicating a medium effect size. Interestingly, a second shifted band appeared in the EMSA at lower concentrations of TFAM for the ^5mCpG-HSP1 lanes compared to HSP1. Together these results show that TFAM binds with a higher affinity to HSP1 when it is methylated at CpG sites, and points to a difference in the formation of higher order TFAM-DNA complexes.

Oligomerization properties of TFAM are influenced by ^5mCpG

TFAM is known to oligomerize on DNA as part of its mechanism for compacting DNA. TFAM binds to single DNA binding sites inducing a “U-turn” shaped 180° bend [35, 45, 56, 57] which promotes protein-protein interactions that stabilize the higher order complexes [50]. Therefore, we first examined the oligomerization properties of TFAM using stoichiometric EMSA with the HSP1 DNA (Figure 3). At concentrations of DNA above the K_Dapp, TFAM binding to HSP1 DNA gives rise to a population of higher order complexes. The stoichiometric EMSA with DNA:TFAM ratios from 1:0 to 1:8 revealed three major complexes 1–3 (Figure 3A and B). A plot of the fraction bound of each complex as a function of TFAM concentration showed a clear preference of TFAM to form the higher-order complexes at lower DNA:TFAM ratios for the ^5mCpG-HSP1 DNA (compare Figure 3A and B blue, red and green lines). However, as seen from the fraction of free DNA remaining at each concentration (black line), as expected the overall fraction bound was the same for both HSP1 and ^5mCpG-HSP1. These results show that ^5mCpG methylation of the HSP1 provides an environment which promotes the multimerization of TFAM on DNA.

Figure 3. Methylation of HSP1 promotes multimeric TFAM-DNA complexes.

Representative images of stoichiometric EMSA experiments. Increasing concentrations of TFAM (0–800 nM) were used in binding reactions with the (A) HSP1, (B) ^5mCpG-HSP1, (C) non-sequence specific (NS) and (D) ^5mCpG-NS DNA. Graphs show the fraction of free DNA (black line) plotted relative to the left-hand y axis and the fraction bound relative to the right-hand y axis. HSP1 and ^5mCpG-HSP1 were quantitated separately for each of the distinct multimers (from 1 to 3). The NS and ^5mCpG-NS bound products were quantitated as one ‘multimer’, as the complexes did not appear as distinct bands. (n=3; error bars = ± S.D.).

In addition to the recognition of specific promoter sequences, TFAM condenses mtDNA to form the mitochondrial nucleoid [50, 58]. To accomplish this, TFAM coats the mtDNA using a non-sequence specific DNA binding mode that is highly cooperative [45, 50, 58, 59]. Therefore we used a stoichiometric EMSA to determine whether ^5mCpG in a nonspecific (NS) DNA sequence would similarly influence the formation of higher order TFAM-DNA complexes. The designed 33 bp sequence from the mitochondrial 16S RNA gene has CpG sites located at identical positions to that of HSP1. The EMSA with NS and ^5mCpG-NS DNA showed no difference in the formation of higher order TFAM-DNA complexes (Figure 3C and D). In addition, the complexes were not as well defined as they were for the HSP1 DNA. These results suggest that ^5mCpG in short non-sequence specific DNA does not alter TFAM binding, unlike the HSP1 for which higher order complexes are influenced by the CpG methylation state of the DNA.

CpG methylation has no significant impact on TFAM-induced DNA condensation

TFAM condenses the 16,569 bp mtDNA to form nucleoids [27, 57, 58]. Assuming a TFAM binding site size of 30 bp, based on footprinting [60], 552 TFAM molecules would bind to a single mtDNA. However, the TFAM-DNA ratio varies in cells; for example, in HeLa cells between 900–1700 molecules of TFAM can associate with a single mtDNA [26, 61]. Thus, oligomerization of TFAM on a short 33 bp DNA fragment does not mimic the physiologically relevant scenario of mtDNA nucleoid formation. Therefore, we investigated DNA compaction using an established model system for mtDNA nucleoid formation [27] using EMSA compaction assays and electron microscopy. These experiments used a 3,600 bp linearized plasmid (pLSP3) containing 583 bp of the mtDNA D-loop region [44]. The pLSP3 plasmid was purified from a dam^–dcm^–strain of E. coli and then linearized and methylated to completion at CpG sites using SssI methytransferase (Supplementary Figure S2A). Assays were performed over a range of TFAM:DNA bp ratios. At increasing ratios of TFAM:DNA bp, the complexes exhibit decreased electrophoretic mobilities (Figure 4A). Quantitative analyses of the EMSA (Figure 4B) revealed no statistically significant difference between the CpG methylated and nonmethylated DNA in the distance the complexes migrated through the gel. We also examined the samples from the same EMSA experiments using transmission electronic microscopy to illustrate the degree of DNA compaction at each TFAM:DNA ratio (Supplementary Figure S4A and B). These results demonstrate that CpG methylation of DNA promotes TFAM-dependent condensation of DNA into model nucleoid structures as efficiently as nonmethylated DNA.

Figure 4. The effect of ^5mCpG DNA on the DNA-compaction properties of TFAM.

(A) Representative EMSA of the TFAM-induced compaction of linearized pLSP3 and ^5mCpG -pLSP3 DNA. The reactions included DNA alone (lane 1 and 6), and DNA bound to TFAM, at the following ratios: TFAM:DNA = 1:60 bp (lane 2 and 7), TFAM:DNA = 1:45 bp (lane 3 and 8), TFAM:DNA = 1:30 bp (lane 4 and 9), and TFAM:DNA = 1:15 bp (lane 5 and 10), which are equivalent to estimated DNA coverage by TFAM of 50%, 75%, 100% and 200%, respectively. CpG-LSP3 is shown in lanes 1–5 and ^5mCpG-LSP3 is shown in lanes 6–10. (B) Bar graph shows the quantitation of the difference in electrophoretic mobility of each band from (panel A) compared to the free DNA. (n=3; error bars = ± S.D.).

CpG methylation influences mitochondrial transcription in vitro

Mitochondrial transcription initiation requires three proteins, TFAM, POLRMT and TFB2M (reviewed in [24]). TFAM binds to the specific LSP, HSP1 and HSP2 sites, which facilitates the recruitment of the POLRMT and TFB2M to form transcription initiation complexes (Figure 5A) [62]. To test whether ^5mCpG in the promoter regions of LSP, HSP1 and HSP2 alters mitochondrial gene expression, we performed in vitro run-off transcription assays. The pLSP3 plasmid includes the promoter region of the mitochondrial DNA D-loop, corresponding to nucleotides 242–825 of the mtDNA. After excision from pLSP3, the resulting DNA (Figure 5A) is an established tri-promoter template for in vitro run-off transcription assays [44, 45]. Run-off transcription assays were performed with tri-promoter templates that were either nonmethylated or completely methylated at CpG sites (Supplementary Figure S2B).

Figure 5. ^5mCpG influences transcription from a tri-promoter template.

(A) Schematic diagram showing the binding sites of the mitochondrial preinitiation complexes at the HSP1, HSP2 and LSP in the tri-promoter template. HSP2 has a predicted transcription start site, but no reported footprint for TFAM, POLRMT or TFB2M [87–89]. The DNA length and protected sites are to scale and the dashed lines indicate that these complexes are not expected to occupy the promoters simultaneously. Abbreviations for TFB2M and POLRMT are B2 and POL, respectively. The red stars indicate ^5mCpG sites. (B) Representative autoradiogram of in vitro run-off transcription reactions. The increasing TFAM concentration (0–136 nM) is indicated by the triangle. Arrows identify the three transcription products. Shown in panels (C) through (E) for the LSP, HSP1, and HSP2, respectively is the quantitation of transcripts levels for the combined data from raw images that have been scaled to the nonmethylated DNA (as in Supplementary Figure S5 and the Supplementary Materials and Methods). 2-way ANOVA analysis showed that the effect of ^5mCpG on transcript levels was significant for LSP (p < .001) and HSP1 (p < .0001), but not HSP2. (n=3; error bars = ± S.D.).

Transcription occurs in the context of the mitochondrial nucleoid, which is coated with TFAM. Therefore, transcription was monitored for reactions with increasing concentrations of TFAM. As a function of TFAM concentration, all of the templates showed a gradual increase of RNA products from the LSP, HSP1 and HSP2 promoters to a point at which additional TFAM results in decreased production of transcripts (Figure 5B). This is commonly observed and attributed to competition between TFAM binding and preinitiation complex formation in vitro [44, 63, 64]. In the comparison of transcription from the ^5mCpG to the nonmethylated template, there is a striking enhancement of 2.4-fold for HSP1 transcription, which was not observed for HSP2 (Figure 5C–E and Supplementary Figure S5). For the LSP there is approximately a 30% decrease in transcript levels for the ^5mCpG template, evident at lower concentrations of TFAM.

HSP1 is the only promoter that has CpG at well-defined positions in the TFAM binding site. In the LSP, the only CpG is located in the POLRMT/TFB2M binding region, not the TFAM binding site (Figure 1). To isolate this region from the tri-promoter context, we investigated the impact of ^5mCpG on transcription from the LSP in a shorter, 88 bp single promoter template (LSP88) (Supplementary Figure S6A). Interestingly, ^5mCpG enhanced transcription by 2-fold in this context (Supplementary Figures S6B and C). Therefore, the effect of ^5mCpG shows some context dependence in vitro and impacts transcription whether it is located in the TFAM binding site or the POLRMT/TFB2M binding region.

The crystal structures of mitochondrial transcription initiation complexes and TFAM-DNA complexes provides a context for the CpG sequences examined here. The structure of TFAM bound to the HSP1 shows the typical U-turn [50]. Strikingly, the CpG sites are located one base pair away from each of the sites where TFAM intercalates the DNA (Figure 6A). The structure of the mitochondrial transcription initiation complex containing TFAM, TFB2M and POLRMT [51] was produced using the LSP DNA with substitutions designed to promote the formation of an “open-bubble” at the transcription start site (Figure 6B). One cytosine of the CpG is in the double-stranded region 5′ to the open bubble, whereas the other C has been replaced with a T, and is not base-paired. Thus, in the transcription initiation complex structure, this CpG is located precisely at the junction between the duplex DNA and the open-promoter (Figure 6B).

Figure 6. Model showing location of ^5mCpG in HSP1-TFAM and LSP-TFAM-TFB2M-POLRMT structures.

(A) Schematic diagram of HSP1-TFAM structure (PDB ID 4NOD), showing the positions of the CpG sequences. (B) Schematic diagram of the structure of the LSP-TFAM-POLRMT/TFB2M complex (PDB ID 6ERP) to illustrate the CpG positions in a model for the initiation complex. As this structure used an engineered sequence in the position of the CpG, there is a T in place of one C of the CpG. Red spheres represent the DNA nucleotides corresponding to the ^5mC (or T in panel B) and the green circles indicate the site of the TFAM C-terminal tail. (B) Amino acid residues that contact the ^5mC are circled in black

DISCUSSION

In this study we examined the effects of cytosine methylation of CpG on TFAM DNA binding, TFAM-dependent DNA compaction and transcription. Quantitative binding experiments showed that ^5mCpG at two sites within a single HSP1 33 bp TFAM binding site increases the binding affinity for TFAM by a factor of 2. Additionally, as observed by EMSA, there were ^5mCpG -induced differences in the formation of higher order TFAM-DNA complexes in this short HSP1 DNA fragment. However, this effect did not extend to a much longer DNA fragment (3,600 bp) to which TFAM binds primarily non-sequence-specifically. In vitro, ^5mCpG also had an impact on transcription in a promoter-context dependent manner. These findings provide new insights into the impact of ^5mCpG in mtDNA directly on two central functions of TFAM suggesting that ^5mCpG has the potential to influence mitochondrial function in a manner distinct from that in the nucleus.

Unique mechanism for the effects of ^5mCpG on TFAM-dependent activities

The main mechanism of action of ^5mCpG in the nucleus involves the binding of methyl-DNA specific proteins, which in turn regulates access to nucleosomal DNA, and ultimately regulates gene expression (reviewed in [65, 66]). Cytosine methylation is also known to influence the DNA-binding properties of transcription factors (reviewed in [22, 67, 68]). CpG methylation inhibits the DNA binding of methylation sensitive endonucleases and transcription factors, for example homeodomain, bHLH, bZIP, Ets, nuclear receptor and Zn-finger family members [55]. In contrast, some members of the same or different transcription factor families bind with an increased affinity for ^5mCpG DNA, including homeodomain, Zn-finger, bZIP and nuclear receptor proteins. Additionally, ^5mCpG can alter the DNA binding specificity of transcription factors [55]. The likely mechanisms by which ^5mCpG influences these major-groove binding proteins include direct steric hindrance or direct stabilizing hydrophobic interactions with the methyl group [55]. Indeed, structural studies using X-ray crystallography have revealed that the methyl-binding domains (MBDs) of MeCp2 (PDB IDs 3C2I and 5BT2) [69, 70] and MBDs (PDB ID 1IG4, 3VYQ and 6CNQ) [71–73] have stabilizing interactions that involve both hydrophobic as well as hydrophilic interactions, such as hydration at the 5-methyl group of the cytosine.

TFAM does not have the potential to interact directly with the 5-methyl group of cytosine, because it binds in the minor, not major, groove. Additionally, ^5mC is not associated with significant changes in the structure of B-DNA [74, 75]. However, methylation does alter other properties of DNA (reviewed in [22]). The 5-methyl group of thymine stabilizes base stacking relative to uracil as does the ^5mcytosine relative to cytosine [76], and the overall stabilization enthalpy can be strong at −1 kcal/mol [77]. DNA methylation also increases the rigidity of the DNA [78, 79]. For example, nucleosomes contain highly distorted and deformable DNA, but ^5mC reduces nucleosome stability in a manner that is correlated with base-stacking interactions [80]. Further, recent computational analyses provide additional support and indicate that the effects of ^5mC on stacking interactions could be greatly magnified in the context of highly deformed DNA (reviewed in [22]). Therefore, it is highly likely that ^5mC alters stacking and/or deformability of the DNA in the HSP1, which impacts the relative stability of the TFAM-DNA complex.

TFAM as an HMGB protein profoundly bends DNA toward the major groove. This compresses the major groove and the bending is also associated with amino acids that intercalate between adjacent base pairs (base steps) in two regions of the DNA, inducing distortions such as unusually high roll angles (blue in Figure 6A). The locations of the two CpGs in the HSP1 (red in Figure 6A) are both at the base step adjacent to the sites of intercalation (blue in Figure 6A). These segments of DNA are relatively undistorted in the TFAM-DNA crystal structures [35, 50, 57] (red in Figure 6A), but this flanking DNA makes further contacts with TFAM, including with the C-terminal tail. As such, stabilization of this flanking DNA through increased base stacking could lead to the increased affinity of TFAM for ^5mCpG DNA.

Even though there is no CpG in the TFAM binding site, we found that transcription from the LSP single-promoter template was also increased 2-fold. A single CpG is found at the −5 position relative to the transcription start site of the promoter. This dinucleotide interacts extensively with both POLRMT and TFB2M (Figure 6B). Alterations in the stacking of the −4C with −5A that increase the rigidity of the DNA could facilitate strand separation at the −3G position (Figure 6B). However, structural studies with the actual ^5mCpG sequence will be required to substantiate these ideas.

Interestingly, the impact of ^5mCpG does not appear to extend to the non-sequence specific DNA binding properties of TFAM that are important for DNA compaction. The EMSA results (Figure 4) show that TFAM does not adopt well-defined complexes with non-specific sequences of DNA. We previously showed that TFAM bends non-site-specific DNA to a lesser degree than the promoter DNA sequences and that TFAM lacking the C-terminal tail also was impaired in DNA bending [43, 45]. These results, suggest a decreased importance of the positioning of the flanking DNA in non-sequence-specific DNA binding. As such, we suggest that any ^5mCpG-induced DNA stabilization of DNA in that region might have little impact on TFAM. Therefore, it was not surprising that ^5mCpG had no apparent effect DNA compaction by TFAM, because the majority of the binding interactions in the nucleoid are non-sequence specific [58, 81].

Relationship of mtDNA ^5mCpG to mitochondrial transcription

CpG methylation impacts transcription initiation of mtDNA in vitro. Transcripts originating from the HSP1 promoter were increased 2.4-fold, whereas those from LSP were decreased by 28% (Figure 5). This effectively changes the ratio of HSP1 to LSP transcription initiation by a factor of 3. These results are consistent with the increased affinity of TFAM for the ^5mCpG-HSP1 leading to the enhanced recruitment of POLRMT and TFB2M. Also consistent is the decrease of transcripts from the LSP, which occurs in the tri-promoter context. This could be because of a ‘squelching effect’ such that less TFAM is available for promoter recognition. Although there are no CpG sites within the LSP TFAM binding site, methylation of the CpG in the POLRMT/TFB2M binding region (Figure 1) in the LSP-single promoter template showed not repression of transcription, but increased transcript levels (Supplementary Figure S6). Therefore, the effect of ^5mCpG on transcription is context dependent and appears to alter not just TFAM binding, but also the transcription initiation complex. An explanation for such context-dependence of transcription from the LSP could derive from structural features of the promoter regions. For example, if the promoter region were to be occupied by two (or more) transcription complexes, there is the potential for them to interact with each other (Figure 5A). Moreover, there is a pattern of TFAM binding sites between the promoters [82], which could be altered by ^5mCpG creating an environment that is less or more favorable for transcription. We observe changes in TFAM-DNA activities in vitro with fully-methylated promoter DNA. However, the effects of both ^5mCpG elsewhere in the mtDNA and even low levels of ^5mCpG could impact the processes of transcription including elongation and termination, as well as replication, mtDNA copy number and compaction through changes to the relative ratio of utilization of the HSP1 relative to the LSP.

The function of the low ^5mCpG mtDNA in mitochondria in vivo is still unclear. Dozens of studies have correlated the levels of ^5mCpG in the mitochondrial genome to changes in mitochondrial gene expression in cells and in vivo (reviewed in [9, 10]). These studies used different experimental systems and approaches and there is no clear consensus in the literature. With enforced expression of CpG or GpC methyltransferases in mitochondria, a decrease of mitochondrial transcription for Gp^5mC, but not ^5mCpG, has been reported [83]. Other studies find that lower levels of ^5mCpG globally or in the D-loop mtDNA are accompanied by increases in mitochondrial RNA transcript levels [1, 2, 6, 7]. This anti-correlation between CpG mtDNA methylation and transcript levels was also observed when the ^5mCpG levels in the D-loop were increased in disease states [3–5]. In a recent study, both nuclear and mitochondrial DNA transcript levels, mtDNA copy number and cellular functions were quantitated and compared, as ^5mCpG levels were pharmacologically decreased. Here, lower ^5mCpG levels resulted in a two-fold decrease in HSP1/2 transcripts and a slight decrease in LSP ND6 transcripts, and not the anti-correlation previously observed. However, this study reported an increase in mtDNA copy number with reduced ^5mCpG levels [8]. Such an anti-correlation between ^5mCpG levels and mtDNA copy number has consistently been reported [1, 83–86]. Interestingly, a ^5mCpG-dependent shift to lower utilization of the LSP compared to HSP1, has the potential to lower mtDNA copy number, which is consistent with our findings in vitro (Figure 5). Although the function of mitochondrial ^5mCpG is not well understood, understanding how it impacts TFAM-dependent activities in vitro provides a better understanding of ^5mCpG effects on DNA recognition and a foundation and direction for future research.

CONCLUSIONS

The results presented here reveal the impact of CpG methylation of mtDNA promoter DNA on the TFAM-dependent properties of DNA recognition, compaction and transcription initiation. The influence of ^5mCpG on DNA binding and transcription is both sequence and context selective. In contrast, there was no obvious effect of ^5mCpG on TFAM recognition of non-sequence-specific DNA or DNA compaction. Together these results lead to the conclusion that the impact of ^5mCpG on TFAM activities is related to its sequence-selective mode of DNA recognition.