Learn more: PMC Disclaimer | PMC Copyright Notice
DNA Methylation Dictates Histone H3K4 Methylation▿
Abstract
Histone lysine methylation and DNA methylation contribute to transcriptional regulation. We have previously shown that acetylated histones are associated with unmethylated DNA and are nearly absent from the methylated DNA regions by using patch-methylated stable episomes in human cells. The present study further demonstrates that DNA methylation immediately downstream from the transcription start site has a dramatic impact on transcription and that DNA methylation has a larger effect on transcription elongation than on initiation. We also show that dimethylated histone H3 at lysine 4 (H3K4me2) is depleted from regions with DNA methylation and that this effect is not linked to the transcriptional activity in the region. This effect is a local one and does not extend even 200 bp from the methylated DNA regions. Although depleted primarily from the methylated DNA regions, the presence of trimethylated histone H3 at lysine 4 (H3K4me3) may be affected by transcriptional activity as well. The data here suggest that DNA methylation at the junction of transcription initiation and elongation is most critical in transcription suppression and that this effect is mechanistically mediated through chromatin structure. The data also strongly support the model in which DNA methylation and not transcriptional activity dictates a closed chromatin structure, which excludes H3K4me2 and H3K4me3 in the region, as one of the pathways that safeguards the silent state of genes.
Transcriptional repression is often associated with DNA methylation in higher eukaryotes (5). DNA methylation can directly inhibit transcription factor binding or suppress transcription indirectly by recruiting histone deacetylases via methyl-CpG-binding proteins (16). This is clearly an important mechanism for silencing genes. We previously found that methylation of the coding region of a luciferase transcription unit has a larger impact than methylation of the promoter and that the effect of methylation on chromatin structure may play a major role in transcriptional regulation because additional methylation outside of the transcription unit can potentiate the effect of methylation on gene expression (10, 11). We also found that acetylated histones are nearly absent from methylated DNA regions and this association does not propagate along the DNA (11).
Although it has been demonstrated that DNA methylation affects transcription elongation more than initiation in fungi (1, 26), the impact of DNA methylation on transcription may be more complex in mammalian cells. Two studies using a recombinase-mediated transgenic system in mammalian cells proposed that DNA methylation may trigger a closed chromatin structure that blocks transcription initiation when the promoter and coding regions are both methylated (29); additionally, DNA methylation in the intragenic region may lead to histone hypoacetylation and decrease the efficiency of transcription elongation (19). Other studies using nonreplicating plasmids have indicated that the size of the methylated region (14) and methylation adjacent to the promoter (12, 13, 22) may inhibit transcription. While some important implications were made in these studies, the impact of the methylated DNA segments on transcription is difficult to interpret because the size and location of the methylated DNA region may affect transcription differently. While a methylated coding region has little impact on transcription initiation when it is about 1 kb downstream from the promoter, as observed by Lorincz et al. (19), it is conceivable that the same methylated sequence might have a much larger impact on transcription initiation if it were immediately downstream from the promoter.
In Saccharomyces cerevisiae and Drosophila melanogaster, the level of methylated histone H3 at lysine 4 was found to be enriched at the active gene sites (2, 30). Dimethylated and trimethylated histone H3 at lysine 4 (H3K4me2 and H3K4me3, respectively) appeared to be linked to active transcription in chicken erythrocytes and embryos (28, 31) and in human cells (17). H3K4me2 has been found to be present around the active promoters where TFIID and RNA polymerase were also present (15), as well as in the coding sequences of human cells (20). H3K4me2 and H3K4me3 were found at 46% and 66%, respectively, of the start sites of genes on human chromosomes 21 and 22 (3). The presence of H3K4me2 at some locations also appears to be cell type specific, indicating the association of H3K4me2 with markers of active chromosomal regions (3).
Studies of the effect of DNA methylation on local chromatin structure have been more limited. Levels of both H3K4me2 and H3K4me3 were found to be two- to threefold lower in the methylated coding region of the transgene where the depletion of RNA polymerase II (Pol II) was also observed (19). A lack of correlation among gene expression, histone modification, and DNA methylation has been described for some imprinted genes (36). DNA methylation and histone modification are two factors among many in the different pathways and mechanisms for gene suppression.
The present study was designed to determine whether DNA methylation dictates a closed chromatin structure that influences transcription initiation and elongation, depending on the location of the DNA methylation. Methylated DNA patches on the episomes are stably maintained for long intervals after transfection into human cells, as described previously (9, 11). We generated episomes with methylated DNA patches in the promoter, in the reporter coding region, in both the promoter and the reporter coding region, in the first 600 bp of the reporter coding region, and in the last 1.2 kb of the reporter coding region to study the impact of DNA methylation on chromatin structure and gene expression. The sizes of the DNA methylation patches in the promoter alone and in the first 600 bp of the coding region are similar, so we could distinguish whether the differences between methylation in the promoter and in the coding region were due to the size or the location of the methylation. The episome with DNA methylation in the first 600 bp of the coding region and the episome with DNA methylation in the last 1.2 kb of the coding region allowed us to discern whether the location of the DNA methylation patch in the coding region affected transcription differently. We analyzed the presence of H3K4me2, H3K4me3, and Pol II in various regions in correlation with the reporter gene expression of these episomes in human cells.
We found that the size of the DNA methylation patch alone does not determine the extent of transcriptional repression because methylated DNA patches of similar sizes impact transcriptional elongation more than transcriptional initiation. The presence of H3K4me2 correlates with the status of DNA methylation and clearly not with gene expression or Pol II presence. While the level of H3K4me3 also closely correlates with the status of DNA methylation, two exceptions were observed, indicating a contributory role of transcription to the presence of this modified histone. The findings here are consistent with previous findings that methylated H3K4 are enriched in actively transcribed regions. These findings further extend the understanding of how DNA methylation affects the promoter and the coding region. We conclude that DNA methylation in the promoter affects transcription initiation and that DNA methylation has a much larger impact on transcription elongation than on initiation. We also conclude that DNA methylation plays an essential role in the exclusion of methylated H3K4 and dictates a closed chromatin structure. Furthermore, the association of histone modification with DNA methylation has very distinct boundaries that do not spread into regions only 200 bp away. We propose that DNA methylation at the junction of transcriptional initiation and elongation is most critical in transcription suppression and that this effect is mechanistically mediated through chromatin structure.
MATERIALS AND METHODS
Plasmids.
Unique restriction sites were engineered into plasmid pCLH22 (9) in order to clone various sizes of methylation patches at different locations in this plasmid. DNA fragments were purified with a GeneClean kit (ISC BioExpress) after restriction digestion and fractionation by electrophoresis. Unmethylated and in vitro-methylated fragments of the plasmid were ligated to generate patch-methylated constructs as described previously (10, 11) and as detailed below.
Nomenclature of plasmids.
Plasmids with methylation patches of 518 bp to 2.4 kb in length at different locations of the transcription unit were generated using the unique restriction sites available. Plasmids were designated with the addition of “Me” to the name of the patch that is methylated. These include pMeLTR, pMeLuc, pMeLTR/Luc, pMe5′Luc, and pMe3′Luc. The fully unmethylated plasmid is designated pCLH22, and the fully methylated plasmid is designated pMeCLH22.
In vitro DNA methylation.
Plasmid or DNA fragments were methylated in vitro at all CpG sites by using SssI methylase (New England Biolabs) under the conditions recommended by the manufacturer. Following methylation, DNA was dialyzed with nitrocellulose filters (Millipore) after phenol-chloroform extraction. The methylation status of each DNA fragment was confirmed by HhaI restriction enzyme digestion.
In vitro ligation.
The vector preparations were test ligated without the insert to identify any uncut or vector religation products. Only preparations with a background less than 0.01% were used in constructing the patch-methylated plasmids. Methylated and unmethylated portions of the plasmid were ligated in vitro to generate patch-methylated constructs. After ligation, the DNA was phenol-chloroform extracted and dialyzed with nitrocellulose filters. If the Rous sarcoma virus long terminal repeat (RSVLTR) promoter and the luciferase gene remained intact on the unligated vector, these unligated molecules could possibly recircularize in mammalian cells after transfection, thereby generating undesired luciferase expression. Therefore, exonuclease V (US Biochemical) treatment was carried out after ligation of those constructions to eliminate any unligated DNA. Ligated DNA was quantitated by transformation of Escherichia coli by using a known quantity of pCLH22 as a standard.
Cell line and transfection.
The calcium phosphate transfection method (9, 35) was used to transfect 293/EBNA1 cells (9). All transfections were done at least in duplicate in each experiment. All experiments to determine luciferase expression and plasmid quantitation were performed at least twice, using products from independent ligations for confirmation. As described previously (11), each transfection was harvested at least three consecutive times over a 3-week interval, at least 1 week after transfection, to assess any possible changes occurring over time. For chromatin immunoprecipitation (ChIP) assays, transfected cells were selected with 200 μg/ml hygromycin, with the exception of cells transfected with pMeCLH22, which do not express the hygromycin resistance gene due to methylation.
Episome recovery and analysis.
As described previously, 1.25% of the cells were harvested for the luciferase assay; 2.5% of the cells were replated on a 100-mm-diameter tissue culture plate; and the remaining cells were harvested by the Hirt method to recover episomal DNA each time transfected cells reached confluence (9). The same fraction (10%) of the DNA harvested from each transfection was digested with XbaI to linearize the plasmid for DNA quantitation and double digested with XbaI and HhaI to determine the methylation status using Southern blotting. Southern blots were exposed to a phosphorimaging screen, and the radioactivity was quantitated with a PhosphorImager (Bio-Rad GS525 or Bio-Rad FX). DNA quantitation was also confirmed by real-time quantitative PCR (Q-PCR) using at least two different TaqMan probe and primer sets.
Luciferase assay.
An aliquot (1.25%) of transfected cells was harvested and lysed for the luciferase assay. Luciferase activities were analyzed using a Monolight 2020 luminometer (Analytical Luminescence) as described previously (9). After subtraction of the background reading, the luciferase reading was normalized for the amount of DNA from the same harvest, as determined by Southern blotting and/or Q-PCR as described previously (9-11). This ensured that the luciferase expression from the same quantity of plasmids with different methylation states from different transfections were being compared.
Chromatin immunoprecipitation.
ChIP assays were performed by a modification of the protocol described by Braunstein et al. (4) and as described previously (11) with minor modifications. Exponentially growing cultures of 293/EBNA1 cells transfected with patch-methylated stable episomes were fixed with 1% formaldehyde for 10 min at room temperature. Approximately 5 × 106 fixed cells were resuspended in 1 ml radioimmunoprecipitation assay (RIPA) buffer (150 mM NaCl, 50 mM Tris-HCl [pH 8.0], 0.5% sodium deoxycholate, 1% Nonidet P-40, 0.1% sodium dodecyl sulfate [SDS]) that contained mammalian protease inhibitors (Roche). The cell suspension was sonicated and centrifuged as described previously (11). One 120-μl aliquot of soluble chromatin was used for each immunoprecipitation, and one aliquot was reserved as the total chromatin fraction (TCF) without immunoprecipitation steps for Q-PCR analysis. For immunoprecipitation, 1 μg each of anti-dimethyl-histone H3 (Lys4) and anti-trimethyl-histone H3 (Lys4) (Upstate Biotechnology and Abcam) were mixed with 120 μl of soluble chromatin, 6 μg of sheared salmon sperm DNA, and 20 μl of protein G-Sepharose slurry (in RIPA buffer) with rotary mixing at 4°C, overnight. Immunoprecipitation using antibodies against RNA polymerase II (clone 8WG16; Covance) was carried out as described above but mixed with 240 μl of soluble chromatin instead of 120 μl. An aliquot of soluble chromatin was processed without antibodies (no antibody [Ab]) in parallel as negative controls for the immunoprecipitation experiments. The Sepharose beads were collected by centrifugation and washed sequentially for 10 min each at 4°C, once with 0.5 ml RIPA buffer and twice with 0.5 ml phosphate-buffered saline. Immunocomplexes were eluted twice by incubation with 1% SDS in 0.1 M NaHCO3 for 30 min at room temperature, each time with rotary mixing. Eluates were incubated at 65°C for 5 h after the addition of NaCl to a 200 mM final concentration to reverse the formaldehyde cross-links. The TCF aliquot was included from the reverse cross-linking step and processed as the other immunoprecipitation samples. After cross-links were reversed, EDTA was added to a final concentration of 10 mM, Tris-HCl (pH 6.5) was added to a final concentration of 40 mM, and 4 μg proteinase K was added to each tube. After overnight incubation at 37°C, samples were phenol-chloroform extracted and ethanol precipitated with 4 μg of glycogen (Roche). The final DNA pellet was dissolved in 50 μl TE (Tris-EDTA; pH 8.0) for Q-PCR analysis. Five independent immunoprecipitations for each of the anti-H3K4me2, anti-H3K4me3 (four experiments with antibodies from Upstate and one experiment with antibodies from Abcam for these two antibodies), and the anti-RNA polymerase II were carried out for each patch-methylated episome.
Q-PCR and statistical analysis.
Q-PCR was performed with a Bio-Rad iCycler using iQ Supermix (Bio-Rad) and 1 μl of each DNA sample. Fluorescence-labeled TaqMan probes for six regions of the plasmid pCLH22 were synthesized by Biosearch Technologies, and the primers were synthesized by Operon Technologies. Four of the primer and probe sequences were described previously (11). Primers 5′-ACAAGGAGAGAAAAAGCACCGT and 5′-GCCTTCCTAATAATTCACGATCGT and probe 5′-CATGCCGATTGGTGGAAGTAAGGTGG were used for the LTR3 region. Primers 5′-GTCACGCTCGTCGTTTGGT and 5′-TCATGTAACTCGCCTTGATCGT and probe 5′-TGGCTTCATTCAGCTCCGGTTCCC were used for the pBR backbone. Primers 5′-TGAATTCCCCAATGTCAAGCA and 5′-TCGGCACTTTGCATCGG and probe 5′-TTCCGGAATCGGGAGCGCG were used for the hygromycin 5 (Hyg5) region that is 121 bp downstream from the herpes simplex virus thymidine kinase promoter. All Q-PCR assays were carried out using the same two-step program: 95°C for 15 s and 60°C for 1 min for 40 cycles. Within each 96-well plate of Q-PCRs, titrations of known amounts of pCLH22 DNA were included as positive controls and for quantitation. DNA from the TCF and ChIP sample immunoprecipitated with no Ab was included in each 96-well plate of Q-PCRs with ChIP samples immunoprecipitated with antibodies. All Q-PCRs were performed in duplicate. The fraction of immunoprecipitated DNA (%IP) was calculated by dividing the amount of DNA from the sample immunoprecipitated with antibodies (after the background amplification from the no-Ab control was subtracted) by the amount of DNA in the corresponding TCF sample. The %IP of regions LTR1, LTR3, Luc1, Luc2, Hyg, and Hyg5 was then normalized by the %IP of the pBR region of each episome to derive the distribution of H3K4me2 and H3K4me3 on the episome. The %IP of regions LTR1, LTR3, Luc1, and Luc2 from each episome was normalized by the %IP of corresponding regions from pCLH22 to derive the distribution of RNA Pol II in the transcription unit on the episome. A t test was carried out to test the hypothesis that there is no significant difference between the %IP of a region from a patch-methylated episome and the %IP of the same region from pCLH22.
RESULTS
H3K4me2 is present at high levels in the promoter and in the proximal transcribed region and is depleted from stably replicating episomes carrying a high density of DNA methylation.
We previously investigated the relationship between DNA methylation and histone acetylation (11). H3K4me2 and H3K4me3 have been reported to associate with active genes (23, 27, 28). We wondered whether these methylated histones are present on a stable episome and associate with regions of active transcription, as found in the chromosome. pCLH22 is a stable episome that can be maintained long term in 293/EBNA1 cells, and the in vitro methylation can be maintained for up to at least 90 days, as demonstrated previously (9, 10). Human cells harboring the unmethylated and the fully methylated stable episomes pCLH22 and pMeCLH22, respectively, were studied with a ChIP assay using anti-H3K4me2 to explore the presence and distribution of these modified histones on the episome. Seven TaqMan probe and primer sets (Fig. (Fig.1A)1A) were used following the ChIP assay to determine the association of H3K4me2 on different regions of the episome. The “no-antibody” control was used as a negative control for nonspecific pulldown of DNA with protein G-Sepharose beads. The DNA in each PCR was quantitated on the basis of the standard curve generated by using titrations of known amounts of pCLH22 DNA in the same set of PCRs. The percentage of DNA pulled down by the antibodies was calculated by dividing the quantity of DNA in each IP by the quantity of DNA in the TCF, after subtracting the background from the no-antibody control. Because there is no known transcription in the pBR region on the episome of human cells, the %IP of DNA in the pBR region was set at 100. The percentage of DNA immunoprecipitated in other regions was normalized relative to that in the pBR region to assess the association of H3K4me2.
DNA methylation reduces the level of H3K4me2 and H3K4me3 on the stable episome in human cells. (A) Illustration of the episome. Locations of Q-PCR amplicons, LTR1, LTR3, Luc1, Luc2, and pBR, are marked by small bars above the diagram of the episome. The distances between the amplicons in the transcription units are indicated above the labels of the amplicons. The distance between Luc1 and the transcription start site and the distance between Luc2 and the stop codon are indicated under the labels of the regions. (B) Distribution of H3K4me2 on the unmethylated pCLH22 and the fully methylated pMeCLH22. The regions of methylation on the plasmid are depicted by thick horizontal lines hatched with short vertical lines. Long stretches of sequence lacking CpGs within the methylated region are represented by the lack of vertical lines within the thick horizontal line. Q-PCR was performed on the TCF, the no-Ab control, anti-H3K4me2, and anti-H3K4me3 ChIP samples with different TaqMan probe and primer sets. The %IP of regions LTR1, LTR3, Luc1, Luc2, Hyg, and Hyg5 was calculated as described in Materials and Methods and then normalized by the %IP of the pBR region from the same episome to derive the distribution of H3K4me2 on each episome. Histograms with accompanying values represent averages of the normalized %IP from five independent experiments, with standard deviations indicated by error bars. The number in the parentheses above the histogram in the pBR region is the %IP of the pBR region from pMeCLH22 normalized for the corresponding value from pCLH22. (C) Distribution of H3K4me3 on unmethylated pCLH22 and on fully methylated pMeCLH22. Histograms with accompanying values represent averages of the normalized %IP from five independent experiments, with standard deviations indicated by error bars. The number in the parentheses above the histogram in the pBR region is the value of %IP of the pBR region from pMeCLH22 normalized for the corresponding value from pCLH22.
The RSVLTR showed the highest association with H3K4me2, especially at the LTR3 region, which is 191 bp upstream from the transcriptional start site (Fig. (Fig.1B).1B). The Luc1 region showed a similar association (Fig. (Fig.1B),1B), the Hyg region had slightly reduced association (data not shown), and the Luc2 region had a twofold reduction in H3K4me2 association compared with that in the pBR region (Fig. (Fig.1B).1B). The finding that the Luc2 region had more than twofold less H3K4me2 than the Luc1 region is consistent with the finding of Schneider et al. (28) who found that more H3K4me2 is associated with the proximal transcribed region than the downstream regions. In mouse embryonic stem cells or in embryonic fibroblasts, Rougeulle et al. (25) showed that genes subjected to X inactivation and imprinting have a dramatically higher H3K4me2 level in the promoter than in the transcribed regions and that the autosomal genes, and genes that escape X inactivation, do not show such a consistent difference. Schneider et al. (28) demonstrated that the promoter has less association with H3K4me2 than the transcribed region for three of the four genes in the chicken β-globin locus when they are actively transcribed. It appears that the H3K4me2 level is the highest at the junction of the transcription initiation and elongation on the episome in our study. The higher level of H3K4me2 in the promoter than in the transcribed region in our study was not as marked as the X-linked or the imprinted genes (25). It is uncertain what causes the promoter and the coding region to associate with H3K4me2 differently and what the implications of these differences are. It is possible that the promoter strength can affect the association of H3K4me2 with the promoter; this would be consistent with the fact that the RSVLTR, a very strong promoter in 293/EBNA1 cells, showed elevated levels of association with H3K4me2. It is also possible that different factors are recruited to the different promoters, thereby influencing the chromatin modification differently.
Compared with the pBR region of the episome, there was very little H3K4me2 associated with all regions studied on pMeCLH22, which is the episome with a high density of DNA methylation at CpG sites (Fig. (Fig.1B1B and data not shown). It is important to note that the percentage of DNA immunoprecipitated from the pBR region of pMeCLH22 by the anti-H3K4me2 antibody was more than fivefold less than that of pCLH22. Therefore, the level of H3K4me2 was more than 100-fold lower on pMeCLH22 than on pCLH22 in the RSVLTR region and the luciferase coding region. The luciferase gene activity of the fully methylated episome was more than 500-fold lower than that of the unmethylated episome, based on the luciferase assay (9-11). It is clear that H3K4me2 associates with actively transcribed DNA. This result is consistent with the findings by Schneider et al. (28) in which the level of H3K4me2 is increased in actively transcribed genes and that little or no H3K4me2 is found in most of the tissue-specific genes when they are in a silent state, in higher eukaryotes. This finding is also consistent with the observation that methylated H3K4 is associated with transcriptional activation in the chicken β-globin locus (18). The finding that the level of H3K4me2 is substantially reduced on methylated DNA compared with that of unmethylated DNA is consistent with the finding of Lorincz et al. (19).
H3K4me3 distributes differently than H3K4me2 and is depleted from the transcribed region of the fully methylated episome.
To explore the presence and distribution of H3K4me3 on unmethylated and fully methylated stable episomes pCLH22 and pMeCLH22 (in human cells), anti-H3K4me3 was used for ChIP assays, followed by analysis using the seven probe and primer sets described above. The controls, calculations, and normalizations were carried out as described above. It is not possible to compare the levels of H3K4me2 and H3K4me3 on the episome because the efficiencies of the two different antibodies in immunoprecipitation may be different. However, the distributions of H3K4me2 and H3K4me3 across the episome can be compared.
While the distribution of H3K4me3 showed a trend similar to that of H3K4me2 for most of the regions, Luc1 showed the lowest level of H3K4me3 among the seven regions studied (Fig. (Fig.1C1C and data not shown). The level of H3K4me3 was also considerably lower on pMeCLH22 than on pCLH22, except for the pBR region (Fig. (Fig.1C).1C). It is uncertain why H3K4me3 was present at similar levels on pCLH22 and on pMeCLH22 in the pBR region, while none or very little of it was found in other regions of pMeCLH22. Otherwise, the distributions of H3K4me3 were mostly consistent with the findings described by Schneider et al. (28) and Lorincz et al. (19), as discussed above regarding H3K4me2. Therefore, there is a good correspondence between the stable replicating episome and the chromosome.
DNA methylation impacts the presence of H3K4me2 and H3K4me3 on the patch-methylated episomes.
From the experiments described above, it is clear that H3K4me2 and H3K4me3 levels are much reduced on methylated episomes relative to those on unmethylated ones. We were interested in whether the respective distributions of H3K4me2 and H3K3me3 in the transcription unit are affected more by transcription or by DNA methylation. To test this, we carried out ChIP assays using the anti-H3K4me2 and anti-H3K4me3 antibodies on human cells harboring patch-methylated episomes pMeLTR, pMeLTR/Luc, and pMeLuc. The luciferase activity was approximately twofold lower for pMeLTR, which has CpG methylation only in the RSVLTR promoter, than for fully unmethylated pCLH22 (Fig. (Fig.22 and reference 11). The luciferase activity of pMeLTR/Luc, which has CpG methylation in the RSVLTR promoter and the luciferase coding region, was 50-fold lower than that of pCLH22 (Fig. (Fig.22 and reference 11). An eightfold reduction of luciferase activity was detected for pMeLuc, which has CpG methylation in the luciferase coding region only, compared with that of pCLH22.
The position and length of DNA methylation impact gene activity and the levels of H3K4me2 and H3K4me3 on the episomes in human cells. Luciferase activities of the plasmids are shown at the right as luciferase activity relative to that of pCLH22. The luciferase expression was normalized for the amount of DNA harvested from transfected cells by Southern blot analysis. Bars represent the ranges of relative luciferase activity for the same episome from different transfections. The absolute readings of the luciferase activities range from 2 × 106 to 2 × 107 relative light units for pCLH22 from different experiments. At left is shown the distribution of H3K4me2 (black bars) and H3K4me3 (gray bars) on the episomes. The regions of methylation (horizontal lines with short vertical hatch marks) on the episome are shown, with the name of the episome on the left and the size of the methylated patch and the number of methylation sites listed to the right of each line. Histograms represent averages of the normalized %IP from five independent experiments, with standard deviations indicated by error bars. Error bars marked with asterisks indicate that one data point among the five experiments is two- to fourfold higher than the others. We chose to include all data points generated regardless of the clear outliers. A t test was carried out to test the hypothesis that the level of the modified histone is not significantly different between the patch-methylated episome and that of pCLH22. The P values of the t test are listed under each histogram.
The ChIP assay for cells harboring each episome was repeated five times, and Q-PCR was carried out in duplicate with standards and controls for each run. On pMeLTR, the distribution of H3K4me2 in the methylated LTR1 region was the lowest among the seven regions examined and twofold lower than that for the pBR region (Fig. (Fig.22 and data not shown). Although the methylated LTR3 region had slightly more H3K4me2 than the unmethylated Luc1 region on pMeLTR, it was reduced more than twofold from that found in the same region on fully unmethylated pCLH22 (Fig. (Fig.2).2). The H3K4me2 in the unmethylated Luc1 and Luc2 regions stayed at a high level on pMeLTR (Fig. (Fig.2),2), indicating the lack of correlation between the reduction of luciferase gene activity and the level of H3K4me2 in the coding region.
On pMeLTR/Luc, the H3K4me2 level was much reduced in the methylated LTR1 region (more than 16-fold lower than in the pBR region), the LTR3 region (more than 6-fold lower than in the pBR region), the Luc1 region (20-fold lower than in the pBR region), and the Luc2 region (25-fold lower than in the pBR region). Similarly, the H3K4me2 level was much reduced in the methylated Luc1 and Luc2 regions, while the unmethylated LTR1 and LTR3 regions had higher levels of H3K4me2 on pMeLuc (Fig. (Fig.2).2). The levels of H3K4me2 remained high in the Hyg and Hyg5 regions that were unmethylated on all three patch-methylated episomes (data not shown). Patterns of H3K4me3 distribution on these three patch-methylated episomes were similar to those of H3K4me2 (Fig. (Fig.2).2). This finding is consistent with the observation of Lorincz et al. (19) that the methylated region is more than 500 bp downstream of the promoter.
The methylated luciferase gene region did not reduce the level of H3K4me2 or H3K4me3 in the unmethylated promoter of pMeLuc, even though the first methylated CpG was only 170 bp downstream from the LTR3 region. Likewise, promoter methylation on pMeLTR did not appear to reduce the H3K4Me2 or the H3K4Me3 level in the unmethylated coding region, despite the fact that the nearest CpG methylation site was only 115 bp upstream from the Luc1 region and that the luciferase gene activity was reduced twofold. These findings indicate that the chromatin structure associated with DNA methylation does not extend far upstream or downstream and that reduction of gene activity is not clearly correlated with the reduction of H3K4me2 and H3K4me3 levels. Importantly, the data here suggest a strong correlation between DNA methylation and the depletion of H3K4me2 and H3K4me3, even though a direct effect cannot be certain.
DNA methylation immediately downstream from the transcriptional start site has a major impact on gene activity.
The experiments above did not allow us to clearly conclude whether the depletion of H3K4me2 and H3K4me3 is correlated with reporter gene activity, even though the findings suggest that DNA methylation in the coding region plays a major role in determining the presence of H3K4me2 and H3K4me3 locally. We generated two additional patch-methylated episomes, pMe5′Luc and pMe3′Luc, with different sizes and locations for the methylation patch in the luciferase coding region. The luciferase activity of pMe5′Luc, which has CpG methylation in the first 680 bp of the luciferase coding region, was reduced approximately fivefold from that of the fully unmethylated pCLH22 (Fig. (Fig.3).3). In contrast, the luciferase activity of pMe3′Luc, which has CpG methylation in the last 1,220 bp of the luciferase coding region, was very similar to that of the unmethylated pCLH22 (Fig. (Fig.3).3). This result is sufficient to conclude that the length of the patch of DNA methylation in the transcription unit is not the key determinant of transcription and that DNA methylation near the transcription start site has a major impact on gene activity. Compared with the luciferase activity of pMeLTR (Fig. (Fig.2),2), which had a methylated DNA patch of a size similar to that of pMe5′Luc, it is evident that the location of DNA methylation is critical in transcription suppression and that it is possible that DNA methylation impacts transcription elongation more than initiation.
The level of H3K4me2 is correlated with DNA methylation, and the presence of H3K4me3 may be influenced by gene activity in addition to DNA methylation on the episomes in human cells. The luciferase activities of the episomes and the histograms representing the distribution of H3K4me2 and H3K4me3 are depicted as described in the legend to Fig. Fig.22.
The presence of H3K4me2 correlates with DNA methylation, while DNA methylation is not the sole determinant of the H3K4me3 presence on the stable episome.
We carried out ChIP assays using anti-H3K4me2 and anti-H3K4me3 antibodies on human cells harboring the two patch-methylated episomes pMe5′Luc and pMe3′Luc (described above) to investigate whether the presence of H3K4me2 and H3K4me3 correlates with gene activity. Five independent ChIP assay experiments for cells harboring each episome were carried out, and Q-PCR was carried out in duplicate with standards and controls for each run. The levels of H3K4me2 were much reduced in the methylated regions of pMe5′Luc (Luc1 region) and pMe3′Luc (Luc2 region). In contrast, the levels of H3K4me2 in the unmethylated regions of pMe5′Luc (Luc2 region) and pMe3′Luc (Luc1 region) were similar to those in pCLH22 (Fig. (Fig.3).3). These findings clearly indicate that the level of H3K4me2 is correlated with DNA methylation and not with reporter gene activity, since the reporter gene activity of pMe5′Luc is much lower than that of pMe3′Luc.
Levels of H3K4me3 are reduced in both the methylated Luc1 region and the unmethylated Luc2 region of pMe5′Luc as well as in the methylated Luc2 region of pMe3′Luc (Fig. (Fig.3).3). The much reduced level of H3K4me3 found in the unmethylated Luc2 region of pMe5′Luc was surprising, but it was reproduced in all five experiments. This is one of the two instances where the presence of H3K4me2 and H3K4me3 is not concordant (Fig. (Fig.1,1, see the pBR region of pMeCLH22). These findings suggest that the gene activity may have some impact on the presence of H3K4me3 in addition to DNA methylation. The lack of H3K4me2 and H3K4me3 in the 3′ portion of the coding region on pMe3′Luc does not affect gene expression significantly. Hence, H3K4me2 and H3K4me3 are not absolutely required for the passage of RNA Pol II, consistent with the implications of the study by Schneider et al. (28).
The presence of RNA polymerase II in the transcription unit is not correlated with the level of H3K4me2 or H3K4me3.
To further define the relationships among DNA methylation, H3K4 modifications, and transcription, we carried out ChIP assays using antibodies against Pol II to determine the presence of Pol II on the patch-methylated episomes in human cells. Five independent ChIP assay experiments were carried out with cells harboring each of the six patch-methylated and fully unmethylated episomes. Q-PCR was carried out in duplicate with standards and controls for each run as described above. There was no appreciable signal above background level from the LTR1 and the pBR regions in all of the 35 ChIP assays (seven episomes, five experiments). This is consistent with the lack of transcription in the pBR region and the distances of the LTR1 region upstream from the TATA box (310 bp) and the transcription start site of the RSVLTR promoter (339 bp).
Unlike the assays for H3K4, no internal control region can be used to normalize the abundance of Pol II on each episome because no Pol II is in the regions without transcription. The ChIP assay results for Pol II on each patch-methylated episome were normalized relative to the results on fully unmethylated pCLH22, which had unaffected luciferase expression. Compared with unmethylated pCLH22, a more than fivefold reduction of Pol II was observed in the LTR3 region, and a two- to threefold reduction of Pol II was observed in the Luc1 and Luc2 regions of pMeLTR (Fig. (Fig.4).4). As described above, the levels of H3K4me2 and H3K4me3 remained high in the Luc1 and Luc2 regions of this stable episome. The reduction of Pol II in the unmethylated luciferase coding region is most likely due to the inhibition of transcription initiation by the DNA methylation and histone modification in the promoter.
The presence of Pol II in the transcription unit is not correlated with the level of H3K4me2 or H3K4me3. The luciferase activities of the episomes are the same as those shown in Fig. Fig.22 and and3.3. The histograms on the left represent H3K4me2 (black bars), H3K4me3 (gray bars), and Pol II (white bar) on each episome after being normalized to the %IP of the corresponding regions from pCLH22. The accompanying value on top of each bar represents the average of the normalized %IP from five independent ChIP experiments. The data for H3K4me2 and H3K4me3 are from the same experiments as described in the legends to Fig. Fig.22 and and33 but normalized to the %IP of the corresponding regions from pCLH22 instead of the pBR region of the same episome. Error bars indicate standard deviations. No Pol II was detected in the LTR1 and pBR regions of any of the episomes. The histograms on the right show the relative luciferase expression and the total Pol II in the LTR3, Luc1, and Luc2 regions and the total Pol II in the Luc1 and Luc2 regions. Values accompanying the bars for total Pol II are the sums of the regions included in each bar.
Consistent with the nearly 100-fold reduction of luciferase activity for pMeLTR/Luc, a dramatic 30- to 100-fold reduction of Pol II in the LTR3, Luc1, and Luc2 regions of this stable episome was observed (Fig. (Fig.4).4). A 25- to 50-fold reduction of Pol II in the methylated luciferase coding region, including both the Luc1 and the Luc2 regions of pMeLuc, was observed (Fig. (Fig.4).4). Interestingly, a nearly sixfold reduction of Pol II was observed in the LTR3 region, in contrast to the near normal levels of H3K4me2 and H3K4me3, of pMeLuc (Fig. (Fig.4).4). This finding indicates the discordance of H3K4 modification and the transcriptional activity as reflected by the presence of Pol II. It is possible that CpG methylation in the 1.9-kb DNA immediately downstream of the promoter affects the transcription initiation, even though it does not impact H3K4 modification.
There was a 25- to 100-fold reduction of Pol II in the Luc1 and Luc2 regions of pMe5′Luc, while only a small reduction of Pol II was observed in the LTR3 region (Fig. (Fig.4).4). This finding indicates that CpG methylation in the 600 bp of DNA immediately downstream from the promoter does not have a significant impact on the presence of Pol II in the promoter. Despite the nearly normal level of Pol II in the promoter of pMe5′Luc, very little Pol II was present in the coding region only 115 bp downstream from the transcriptional start site. This finding is consistent with the previous finding that DNA methylation in the coding region can reduce the efficiency of Pol II elongation (19). It is also important to note that the level of H3K4me2 was not reduced in the Luc2 region, even though a significant reduction of Pol II was observed in the same region of pMe5′Luc. This strongly indicates that transcriptional activity is not concordant with the presence of H3K4me2, and it is consistent with the observation for pMeLuc described above.
A nearly twofold reduction of Pol II was observed in the unmethylated LTR3 and Luc1 regions of pMe3′Luc, while a more than 10-fold reduction of Pol II was observed in the methylated Luc2 region of the same episome (Fig. (Fig.4).4). This finding adds further support to the interpretation here that DNA methylation can reduce Pol II elongation efficiency. As expected, there was nearly no Pol II found in all three regions of fully methylated pMeCLH22 (Fig. (Fig.44).
Taken together, the findings here strongly indicate that the presence of H3K4me2 correlates with DNA methylation and not the presence of Pol II. While the presence of H3K4me3 is not entirely concordant with DNA methylation, it also does not correlate with the presence of Pol II. These findings also indicate that DNA methylation immediately downstream from the promoter can reduce the Pol II presence at the promoter through mechanisms other than H3K4 modifications. This effect of DNA methylation is potentially size dependent. It is noteworthy that the luciferase gene activity correlates with the combined levels of Pol II in the Luc1 and Luc2 regions better than the combined levels of Pol II in the promoter and the coding region (LTR3, Luc1, and Luc2) (Fig. (Fig.44).
DISCUSSION
Consistent with our previous findings, we found in this study that the location of DNA methylation plays a major role in transcription suppression. We further demonstrate in the current study that methylated DNA patches of similar size impact transcription elongation much more than transcription initiation. The size of the DNA methylation patch immediately downstream of the promoter may play a role in the inhibition of transcription initiation, but it does not affect the presence of H3K4me2 and H3K4me3 in the promoter. We find that DNA methylation immediately downstream from the transcription start site has a larger impact on transcription than DNA methylation in the 3′ end of the coding region. The abundance of Pol II in the coding region reflects the reporter gene activity. Most importantly, we find that the presence of H3K4me2 clearly correlates with the status of DNA methylation in the region and not with gene expression or Pol II abundance. Furthermore, the exclusion of H3K4me2 from methylated DNA regions is a very local effect and is consistent with our previous report of the local effect of DNA methylation on the presence of acetylated histones. While the level of H3K4me3 also closely correlates with the status of DNA methylation, two exceptions were observed, indicating a contributory role of transcription in the presence of this modified histone.
Methylation of the promoter (pMeLTR) and the 600 bp of the 5′ region of the coding region (pMe5′Luc) have very different impacts on reporter gene activity, even though the DNA methylation patch sizes were similar. The level of Pol II was reduced about sevenfold in the promoter of pMeLTR, and that most likely leads to the two- to threefold reduction of Pol II in the coding region. In contrast, Pol II was nearly absent in the coding region of pMe5′Luc despite the relatively high level of Pol II detected in the promoter. These findings strongly suggest that DNA methylation impacts transcriptional initiation less than elongation.
DNA methylation at the 5′ end of the coding region inhibits reporter gene expression much more than DNA methylation at the 3′ portion of the coding region, even though the methylation patch at the 3′ end was twice the size (Fig. (Fig.4,4, compare pMe5′Luc and pMe3′Luc). Also, DNA methylation at the 5′ end of the coding region has an effect on reporter gene expression that is similar to DNA methylation in the entire coding region (Fig. (Fig.4,4, compare pMeLuc and pMe5′Luc). These findings further indicate that the location of the DNA methylation plays a more important role in transcriptional suppression than the size of the DNA methylation patch. It is clear that DNA methylation at the junction of transcriptional initiation and elongation may play a crucial role in gene suppression. DNA methylation in the region surrounding the transcriptional start site has been correlated with the lack of expression of genes in specific tissues or those on the inactivated X chromosome; these include the glutathione S-transferase gene (21), the hypoxanthine phosphoribosyltransferase gene (24), the Maspin gene (8), and the DEAD-box protein 4 (32). It has been proposed previously that methylation at five CpG sites surrounding the MAGE-A1 transcription start site might be required for the suppression of the gene (6). Our findings here may explain the role of DNA methylation in the suppression of these genes.
While DNA methylation in the promoter may have a small impact on transcriptional initiation, DNA methylation at the beginning of the transcriptional start site may inhibit transcription more dramatically. When DNA methylation is present in the region surrounding the transcriptional start site, which includes parts of the promoter and the coding region, transcription is effectively suppressed. The low reporter gene activity from the episome harboring DNA methylation in both the promoter and in the coding region (pMeLTR/Luc) supports this view, even though the size of the methylation patch cannot be completely ruled out as a contributing factor. We know that episomes harboring 120 bp of methylated DNA patch immediately downstream from the transcriptional start site showed similar reporter activity and no reduction of H3K4me2, H3K4me3, and Pol II levels compared with those of the fully unmethylated episome (data not shown). Therefore, a minimum size of DNA methylation (between 120 bp and 600 bp) at a critical region, such as the region of transition for transcriptional initiation and elongation, may efficiently suppress gene activity.
In the current study, DNA methylation of the entire 1.9-kb reporter coding region (pMeLuc) led to a more than eightfold reduction of Pol II, while the levels of H3K4me2 and H3K4me3 remained high in the promoter. However, 600 bp of DNA methylation at the 5′ end of the coding region (pMe5′Luc) did not change the levels of Pol II, H3K4me2, and H3K4me3 in the promoter. Very little Pol II was found in the coding region of pMeLuc and pMe5′Luc, indicating the lack of transcription, particularly elongation. These findings indicate that the size of the DNA methylation patch may affect transcription initiation in the promoter immediately upstream through a mechanism independent of the presence of H3K4me2 and H3K4me3. This conclusion is not inconsistent with the finding of Lorincz et al. (19) that the methylated coding region, at about 1 kb downstream from the promoter, has little impact on transcriptional initiation. It is also noteworthy that an accumulation of Pol II in the promoter was not observed on either pMeLuc or pMe5′Luc, supporting the model in which the reduction of transcriptional elongation efficiency leads to premature Pol II dissociation, a conclusion also favored by Lorincz et al. (19). A much larger impact on reporter gene expression is observed when DNA methylation is present in the first 600 bp of the coding region (pMe5′Luc) than when the 1.2 kb of DNA at the 3′ end of the coding region is methylated (pMe3′Luc). The level of Pol II is much lower in the methylated 3′ region than the unmethylated 5′ region on pMe3′Luc, indicating that DNA methylation affects transcription elongation. The transcription efficiency in the 3′ coding region may play a less important role in overall gene activity, based on the observation that the reporter gene activity of pMe3′Luc is nearly normal (despite the absence of Pol II in the methylated 3′ coding region).
It has been reported that methylated H3K4 is found in the actively transcribed gene region in higher eukaryotes (3, 15, 20, 28) and that the level of H3K4me2 is coupled with transcription (30). It has also been reported that the level of H3K4me2 is two- to threefold lower in the methylated transgene than in the unmethylated transgene (19). However, the connection between DNA methylation, H3K4 methylation, and transcription cannot be discerned from these previous studies. We observed a tight correlation between the depletion of H3K4me2 in the regions of DNA methylation, regardless of the reporter gene activity. The luciferase gene activity from pMe3′Luc is 20% reduced, but the level of H3K4me2 in the methylated 3′ portion of the coding region is fivefold or more lower than those of episomes (pMeLTR, pMe5′Luc, and pCLH22) without DNA methylation in the same region. Conversely, the level of H3K4me2 remains high in the unmethylated DNA regions regardless of the presence of Pol II. For example, the reporter gene activity is more than fivefold suppressed and the Pol II level is more than 20-fold reduced, while the level of H3K4me2 remains high in the unmethylated 3′ coding region of pMe5′Luc. Furthermore, DNA methylation approximately 200 bp downstream does not affect the presence of H3K4me2 in the unmethylated promoter on pMeLuc and pMe5′Luc. This distinct boundary effect strongly supports the view that DNA methylation affects the presence of H3K4me2 directly.
We propose here that DNA methylation dictates a closed chromatin structure that is devoid of H3K4me2 and inhibits transcription and that the presence of H3K4me2 marks an open chromatin structure that would permit transcription if all other conditions for active transcription were fulfilled (Fig. (Fig.5).5). Based on this model, H3K4me2 can be detected where active transcription is observed, and it can also be detected at other regions with open chromatin when no active transcription occurs in these regions. This model is consistent with the observations that H3K4me2 correlates with active chromosomal regions (3) and that H3K4me2 marks monoallelically expressed genes (25). This model would provide an explanation for the detection of H3K4me2 in the inactive β-globin locus in the study by Schneider et al. (28), and it explains the high level of H3K4me2 observed in the unmethylated pBR region on the episomes that had no detectable transcriptional activity in the current study. This model is also consistent with the finding that methylated H3K4 is absent from the methylated DNA in the imprinted gene, even though DNA methylation is not concordant with gene expression (36). This model may provide further mechanistic understanding of the findings in a recent study describing the correlation between histone methylation and transgene expression (37).
Gene suppression by DNA methylation through chromatin-mediated mechanisms. A DNA region that exceeds a threshold of DNA methylation size and density would have a closed chromatin structure, which excludes H3K4me2 and H3K4me3 and other histones that permit an open chromatin structure (within a defined boundary). In addition to other mechanisms for gene regulation, DNA methylation safeguards the silent status of the genes. P, promoter; C, coding region.
Consistent with previous reports in higher eukaryotes (19, 28), the level and presence of H3K4me3 correlate mostly with those of H3K4me2 and DNA methylation status in the current study, with two exceptions. First, H3K4me3 was depleted in the 3′ unmethylated coding region on pMe5′Luc, while the H3K4me2 level remained high in the same region. It is possible that transcriptional activity plays a contributory role in the presence of H3K4me3. While DNA methylation plays a major role in the exclusion of H3K4me3, the lack of transcription elongation determines the exclusion of H3K4me3 in the unmethylated region downstream. Second, a high level of H3K4me3 was detected in the pBR region of the fully methylated pMeCLH22, while H3K4me2 was depleted from the entire episome and H3K4me3 also was nearly absent from other regions of the same episome. This is not inconsistent with the finding of Bernstein et al. (3) that a substantial fraction of H3K4me3 was found to reside outside of the start of known genes. We do not have an explanation at the current time for the presence of H3K4me3 in the methylated pBR region or outside of the gene start sites; it is possible that there is an element in the region that dictates the presence of H3K4me3. This observation suggests that the mechanism and the role of histone modification may be more complex than current understanding permits. Many other factors, such as polycomb and trithorax group proteins, may also participate in determining or maintaining the closed or open chromatin structure. Polycomb group protein has also been reported to associate with DNA methyltransferase activity (34). It is possible that DNA methylation is a downstream event to safeguard the silent status of the region. Although the current study does not address how de novo DNA methylation is targeted, it does demonstrate that the effect of preexisting DNA methylation on the exclusion of H3K4me2 and H3K4me3 is very local, suggesting a direct effect of DNA methylation on histone modification. We are interested in investigating what histone modifications and factors are associated with methylated DNA in human cells.
Using antibodies from two independent sources, we consistently detected H3K9me3 on fully methylated pMeCLH22 in all regions examined (data not shown). However, the relationship among DNA methylation, transcription, and the presence of methylated H3K9 on the patch-methylated episomes appears to be much more complex. Methylated H3K9 has been reported to be associated with condensed heterochromatin, but H3K9 methylation has also been found to be enriched in actively transcribed regions of some genes (7, 33). It is possible that transcriptional elongation plays a significant role in the H3K9 modification. Further analysis of the interplay between DNA methylation and histone modifications is necessary.
Acknowledgments
We thank M. R. Lieber and D. Shibita for critical reading of the manuscript. We also thank J. Rice and M. Stallcup for helpful discussions.
This work was supported by NIH grant RO1 GM54781.
Footnotes
▿Published ahead of print on 22 January 2007.