The FMR1 gene is subject to repeat mediated-gene silencing when the CGG-repeat tract in the 5′ UTR exceeds 200 repeat units. This results in Fragile X syndrome, the most common heritable cause of intellectual disability and a major cause of autism spectrum disorders. The mechanism of gene silencing is not fully understood, and efforts to reverse this gene silencing have had limited success. Here, we show that the level of trimethylation of histone H3 on lysine 27, a hallmark of the activity of EZH2, a component of repressive Polycomb Group (PcG) complexes like PRC2, is increased on reactivation of the silenced allele by either the DNA demethylating agent 5-azadeoxycytidine or the SIRT1 inhibitor splitomicin. The level of H3K27me3 increases and decreases in parallel with the FMR1 mRNA level. Furthermore, reducing the levels of the FMR1 mRNA reduces the accumulation of H3K27me3. This suggests a model for FMR1 gene silencing in which the FMR1 mRNA generated from the reactivated allele acts in cis to repress its own transcription via the recruitment of PcG complexes to the FMR1 locus.
Fragile X syndrome (FXS; MIM #300624) is the most common cause of inherited intellectual disability and autism spectrum disorders (1). It results from mutations in the FMR1 gene that encodes FMRP, a protein important for the regulation of translation of many brain-specific mRNAs (2,3). The most common mutation in FXS is the expansion of a CGG-repeat tract in the 5′ untranslated region of the FMR1 gene that results in repeat-mediated epigenetic silencing via a mechanism that is still not well understood (4,5).
The majority of fragile X (FX) alleles are heavily methylated at the DNA level (4,6) and enriched for other epigenetic marks including histone H3 dimethylated at lysine 9 (H3K9me2), trimethylated at lysine 9 (H3K9me3), trimethylated at lysine 27 (H3K27me3) and histone H4 trimethylated at lysine 20 (H4K20me3) (7–12). While rare unmethylated FX alleles lack cytosine methylation, they are nonetheless enriched for H3K9me2 (13). This indicates that dimethylation of H3K9 precedes or is independent of DNA methylation. This would be consistent with the observation that de novo methylation of many other genes is associated with prior histone methylation (14–18). The silenced FMR1 gene can be partially reactivated by treatment with the DNA methyltransferase inhibitor 5-aza-2′-deoxycytidine (AZA) (7,8). Similar reactivation can also be achieved using splitomicin (SPT), an inhibitor of the SIRT1-mediated deacetylation of H4K16 (11). While these findings have provided us with insights into some aspects of gene silencing, many aspects of the sequence of events leading to gene silencing remain unknown.
To better understand the gene silencing process and to improve the prospects for long-term reactivation of silenced FX alleles, we have characterized the histone marks on the FMR1 gene in FX lymphoblastoid cells before and after treatment with AZA. Our data provide insight into the sequence of events preceding and following DNA methylation and lend support to the idea that the deposition of repressive histone marks like H3K9 di- and trimethylation and H4K20me3 precede or are independent of DNA methylation. Our data also implicate the FMR1 transcript itself in the silencing process via its recruitment of Polycomb Group (PcG) complexes to the reactivated allele.
Resilencing of the reactivated allele occurs before DNA methylation returns
While some FX cell lines express some FMR1 mRNA (19), all the FX cell lines used in this study had FMR1 levels that were below the detection limit of our assay (<0.01% of normal). AZA treatment of these cells resulted in a biphasic pattern of gene reactivation in which the FMR1 mRNA levels reached 6–10% of normal after 3 days of treatment and then increased to a maximum of 26–80% of normal 7 days after the drug was removed (Fig. 1A). Inter-experimental variability was seen in the final level of FMR1 reactivation; however, GM03200B showed consistently lower levels of reactivation than GM04025E. Similar differences in the extent of reactivation in different cell lines have previously been reported that was attributed to some combination of the difference in repeat number and the extent of DNA methylation (20). However, irrespective of the extent of reactivation, transcription of the FMR1 gene was transient, and by 10 days, the FMR1 mRNA levels began to decline, consistently falling below 1% of normal by Day 24 in both cell lines (Fig. 1A). This pattern of reactivation differs from what is seen with SPT where the levels of FMR1 mRNA begin to decline soon after the drug was withdrawn (data not shown). The long-lasting effect of AZA is likely due to it being a cytidine analog that can be incorporated into DNA. Pyrosequencing of DNA isolated from AZA-treated cells was used to analyze the methylation status of 22 CpG residues in the FMR1 promoter (Fig. 1B). This showed that methylation levels continued to decrease after the drug was withdrawn. Furthermore, this data demonstrated that even though the FMR1 mRNA levels had dropped significantly by 17 days, remethylation of the allele had not yet begun. This supports the idea that gene silencing is established before DNA methylation takes place perhaps via the recruitment of other silencing factors to the unmethylated allele. Work by others has shown that methylation is eventually restored after longer times in culture (20,21).
The levels of H3K9me2/3, H4K20me3 and H3K27me3 do not decline on DNA demethylation
Using chromatin immunoprecipitation (ChIP) coupled to real-time qPCR, we examined the levels of all four repressive chromatin marks known to be associated with the silenced FMR1 gene, H3K9me2, H3K9me3, H4K20me3 and H3K27me3 in AZA-treated cells (Figs 2, 3 and 4). The levels of H4K20me3 were similar on both cell lines at Day 0. In contrast, at Day 0, H3K9me2 levels were 80% and H3K9me3 levels were 20% higher on the FMR1 gene in GM04025E cells than GM03200B cells. No significant decrease in any of these marks relative to the untreated cells was seen at any time point. In the case of genes in which DNA methylation precedes the deposition of repressive histone modifications, AZA treatment leads to decreases in the levels of these repressive chromatin marks (22), whereas in the case of genes in which H3K9 methylation precedes DNA methylation, AZA has no effect (14–18). Thus, the failure to see any decrease in the repressive chromatin marks on demethylated FX alleles supports the idea that these marks are deposited prior to or are independent of DNA methylation.
The levels of H3K27me3 track closely with FMR1 mRNA levels
While the levels of some repressive marks increased transiently during AZA treatment, only H3K27me3 increased consistently as the FMR1 mRNA levels increased, with both reaching a maximum at 10 days after the addition of AZA (Fig. 4A). Then, as the level of FMR1 mRNA declined, so the level of H3K27me3 returned to its pretreatment level (Fig. 4A). The H3K27me3 profile for the FX cell lines after treatment with AZA was similar to that of H3K4me2, a mark of active transcription (Fig. 4A). The change in H3K4me2 level concomitant with change in the level of FMR1 mRNA is consistent with the idea that this modification is deposited cotranscriptionally (23).
The highest levels of these two apparently antagonistic modifications coincided with the highest level of transcription. Furthermore, re-ChIP experiments where chromatin fragments that were first immunoprecipitated with an antibody to H3K27me3 and then immunoprecipitated with an antibody to H3K4me2 demonstrated that both of these modifications frequently co-localize on same allele (Fig. 4B). An increase in both H3K4me2 and H3K27me3 was observed in two additional FXS patient cell lines but not in normal cells (Fig. 4C and D). This suggests that the increase in both H3K4me2 and H3K27me3 is a specific characteristic of the reactivated FMR1 gene in FX cells rather than a nonspecific effect of treatment with AZA.
The PcG protein enhancer of zeste 2, EZH2, is the histone methyltransferase catalytic subunit, which along with two other PcG proteins suppressor of zeste 12 (SUZ12) and embryonic ectoderm development (EED), is responsible for carrying out trimethylation of H3K27 (24). As expected, EZH2 levels were also elevated on the reactivated FMR1 gene in cells treated with AZA (Fig. 5A).
The increase in H3K27me3 on the FMR1 gene on gene reactivation is dependent on FMR1 mRNA
A comparable increase in H3K27me3 levels was also seen when the FMR1 gene was partially reactivated with SPT, an inhibitor of the deacetylase SIRT1 (Fig. 5B) that deacetylates H4K16 on FX alleles (11). As SPT treatment does not cause DNA demethylation (11), this supports the idea that the signal for H3K27me3 deposition is not the DNA methylation status per se.
Addition of the transcription inhibitor actinomycin D to AZA-treated cells reduced the level of H3K27me3 accumulation on the reactivated allele (Fig. 6A). This suggested a role for transcription in the deposition of H3K27me3. To test whether the decline in H3K27me3 reflected a specific effect of transcription of the FMR1 locus, we treated FX lymphoblastoid cells with 10 µm AZA for 3 days prior to transfection with siRNA specific for FMR1 mRNA. Lymphoblastoid cells are notoriously difficult to transfect (25,26) and even using the AMAXA Nucleofector® II or Neon™ transfection systems, significant FMR1 knockdown was not achieved with the GM03200B cell line. With the GM04025E cell line, we were able to obtain an average of 40% FMR1 knockdown in two separate experiments. This resulted in a decline in the level of H3K27me3 of 30 ± 2%. To confirm this correlation between the decline in FMR1 mRNA and the drop in H3K27me3 enrichment, we examined the effect of FMR1 siRNA knockdown in an FX fibroblast cell line (GM05848) in which the siRNA-mediated decrease in the FMR1 mRNA levels was easier to achieve. The fibroblast cell line showed the same reactivation/silencing kinetics as the lymphoblastoid cells lines, with the maximum reactivation being achieved in 10 days and with levels of activation comparable with what was seen in GM03200B (Fig. 6B).
Knockdown in these cell lines was associated with a significant decrease in the H3K27me3 levels on the FMR1 gene (Fig. 6C and D). This effect was specific to the FMR1 gene as there was no effect on the levels of H3K27me3 on MyoD, another H3K27me3-enriched gene (Fig. 6D). No FMRP was detected in western blots from GM04025E or GM05848 cells after AZA treatment likely due to the poor translation of transcripts with large numbers of CGG-repeats (data not shown). Thus, the decrease in H3K27me3 seen on FMR1 knockdown is very unlikely to result from a change in FMRP levels. Rather our data support the idea that the FMR1 mRNA is specifically responsible for the increased deposition of H3K27me3 mark on the reactivated FMR1 gene in FX cells.
The increase and decrease in H4K16 acetylation precedes that of the FMR1 mRNA
In addition to an increase in the active mark, H3K4me2, an increase in a second active mark, H4K16Ac, was seen, with H4K16Ac reaching its maximum levels at 3 days in the GM03200B FX cells (Fig. 7A) and somewhere between 3 and 10 days in the case of GM04025E FX cells (Fig. 7B). The rise and fall of the H4K16 acetylation mark precedes slightly the rise and fall of the FMR1 mRNA levels.
Reactivation of FMR1 alleles by treatment with AZA allows us to examine which repressive histone modifications decline in response to DNA demethylation and thus to differentiate repressive events that precede DNA methylation from those that occur downstream. As reactivation is followed by resilencing when AZA is removed, this system also allows us to examine which factors are associated with resilencing of the FX allele.
No significant decline was seen in the level of H3K9me2, H3K9me3, H3K27me3 or H4K20me3 during reactivation of the FMR1 gene with AZA (Figs 2–4). Thus, either these histone modifications precede DNA methylation or they are independent of it. This is consistent with the observation that unmethylated FX alleles are nevertheless enriched for H3K9 methylation (13) and that in unmethylated embryonic stem cells (ESCs), the deposition of H3K9 methylation occurs prior to DNA methylation during differentiation (27). Thus, the preponderance of evidence suggests that the FMR1 gene belongs to the large group of human genes in which DNA methylation is a consequence of earlier histone modification events at residues like H3K9 rather than being an early event in the gene silencing process (14–18). We show here that gene reactivation can occur despite the presence of these repressive marks. This suggests that if these modifications are necessary for FMR1 gene silencing, they are not sufficient. This is consistent with the fact that the FX alleles that are associated with H3K9me2 but show no DNA methylation are not silenced (13).
The idea that H3K9 methylation precedes DNA methylation conflicts with recent data obtained from the differentiation of two additional FX ESC lines (28). In these lines, the FX allele was already at least partially methylated, but H3K9me2 did not appear on the FX allele until after 45–60 days of neuronal differentiation at a time when neuronal maturation was taking place. One way to reconcile these data is if the ESCs used in this study were only partially methylated and that full DNA methylation only occurred much later in differentiated neurons after the deposition of H3K9me2.
The fact that we do not see any increase in H3K9me2 during resilencing in our model system may reflect differences between the initial silencing process and resilencing or the fact that, as not all repressive chromatin marks are removed by AZA treatment, we are only able to observe later events in the silencing process. Whatever the explanation for this difference, the increase in H3K27me3 levels that we observed on the reactivated FMR1 gene (Fig. 4) has important ramifications for our understanding of events associated with resilencing and with efforts to achieve long-term reactivation of the silenced allele. It may also reflect a step downstream of H3K9 dimethylation that occurs on de novo silencing in neurons.
G+C-rich regions with a high density of unmethylated CpG dinucleotides are thought to be good targets for vertebrate PRC2 recruitment (29). As the repeat tract in FX alleles consists of >600 bp of 100% G+C DNA with a methylatable C every third base, such alleles may represent very effective targets for PRC2 when methylation is reduced following AZA treatment. However, the fact that a compound like SPT also increases H3K27me3 without affecting DNA demethylation suggests that the lack of DNA methylation per se may not be the signal for PRC2 recruitment on the reactivated FMR1 alleles (Fig. 5B). Our demonstration that reducing the FMR1 mRNA level reduces the trimethylation of H3K27 suggests a role for an RNA-dependent step in this process.
The CGG-repeats in the FMR1 transcript have been shown to form a variety of stem-loop structures (30,31). These structures are analogous to those thought to be responsible for the recruitment of PcG complexes to other genes (32,33). While this work was in progress, a role for a co-transcriptionally formed RNA:DNA hybrid involving the repeat was proposed for FX gene silencing in an FX ESC model (28). It may be that a RNA:DNA hybrid involving part of the repeat serves to tether the transcript to the FMR1 gene, whereas the stem-loop structures formed by the remaining repeats act to recruit PcG complexes and other chromatin modifiers to the FMR1 locus. A similar role for transcript-mediated gene silencing has been invoked for the antisense transcript that overlaps with the RASSF1A gene (34).
Whatever the role of the FMR1 mRNA in the process, it is known that many genes that undergo subsequent de novo methylation in cancer cells are already marked with H3K27me3, and it has been suggested that this chromatin mark might predispose those genes for later DNA hypermethylation (35–37). EZH2, the PcG component that carries out H3K27 trimethylation, can itself recruit DNMTs and is known to be responsible for the de novo methylation of some CpG island genes (17). Thus, it may be that the FMR1 mRNA-mediated recruitment of PcG complexes to the CpG-rich FX repeat is the trigger for de novo methylation of unmethylated alleles. Components of PcG complexes also interact with a variety of histone deacetylases (38). As the reactivated allele is most active when the levels of H3K27me3 are highest, it suggests that this particular modification is not directly responsible for shutting down transcription. Rather H3K27me3 or the PcG complexes recruit other histone modifying factors that do. As FX gene resilencing is also observed long before DNA methylation is restored (Fig. 1), it may be that DNA methylation occurs relatively late in the process, perhaps to lock in the silenced state. Consistent with this idea is the fact that SIRT1, which we previously showed to be important for FMR1 gene silencing via its deacetylation of H4K16 (11), is associated with both PcG complexes and DNMTs and is required for the remethylation of other transiently reactivated genes (39). Furthermore, it would be consistent with our observation that H4K16 acetylation, which increases very rapidly when the FMR1 gene is demethylated, decreases before the levels of H3K27me3 and the FMR1 transcript do (Fig. 7).
AZA-treated alleles are characterized by enrichment not only for H3K27me3 but also for the active chromatin mark H3K4me2. Thus, the reactivated allele has a bivalent chromatin configuration reminiscent of that reported for a variety of genes in ESCs that can become aberrantly methylated in cancer (40) or with aging (41). By analogy with what is thought to occur in these genes, it may be that transcription of the expanded allele in the early embryo (27) makes the FMR1 gene an aberrant PcG target. Once marked with H3K27me3, the FX locus is then at high risk of being silenced and hypermethylated as differentiation proceeds. Thus, there is a catch-22 situation in FXS where the mRNA produced from the FMR1 gene is the trigger for FMR1 gene silencing. This has implications for attempts currently underway to find ways of reactivating the silenced allele in patient cells. FMR1 mRNA-mediated PcG-recruitment may also provide an explanation for how the FMR1 mRNA triggers de novo gene silencing during embryonic development.
MATERIALS AND METHODS
Cell lines and reagents
Lymphoblastoid cell lines from normal (GM06865, GM06895) and FX males (GM03200B, GM04025E, GM07294, GM09145) and a male FX fibroblast cell line (GM05848) were obtained from Coriell Cell Repository (Camden, NJ, USA). The repeat numbers in GM03200B, GM04025 and GM05848 are 530, 645 and 738, respectively. The precise FMR1 repeat numbers in the GM07294 and GM09145 cell lines are not known, but both are full mutation alleles by Southern blotting. The FMR1 mRNA levels for all of the FX cell lines were below the limit of detection in real-time PCR (0.01% of normal), consistent with the presence of very few, if any, cells with PM alleles and with the complete methylation of the FM alleles (data not shown).
Lymphoblastoid cells were grown in RPMI 1640 supplemented with 10% FBS and 1× antibiotic–antimycotic liquid consisting of penicillin, streptomycin and fungisone (Life Technologies, Carlsbad, CA, USA). Fibroblast cells were maintained in DMEM supplemented with 10% FBS and 1× non-essential amino acids (all from Life Technologies). Cells were treated with 10 µm 5-azadeoxycytidine (AZA, Sigma, St Louis, MO, USA) where indicated. Chromatin immunoprecipitation assay kits and normal mouse and rabbit IgG were purchased from Upstate (Temecula, CA, USA). Antibodies against histone H3K9me2 (ab1220), H3K9me3 (ab8898), H3K27me3 (ab6002) and H3K4me2 (ab7766) were from Abcam (Cambridge, MA, USA). Antibodies against H4K20me3 (39 180) were from Active Motif (Carlsbad, CA, USA). Splitomicin was purchased from Tocris Bioscience (Ellisville, MO, USA) and used at a concentration of 700 µm. DNA methylation at 22 CpG residues in the FMR1 promoter was analyzed by pyrosequencing of bisulfite converted DNA by EpigenDx, Inc., (Hopkinton, MA, USA) using assays ADS1451-FS1 and ADS1451-FS2.
Total RNA was isolated from untreated or cells treated with AZA using Trizol (Life Technologies). Superscript III™ first strand cDNA synthesis system (Invitrogen) was used to reverse-transcribe the RNA as per the manufacturer's instructions. Quantitative real-time PCRs were carried out using StepOne Plus™ Real-time PCR system (Applied Biosystems, Foster City, CA, USA), Taqman™ universal PCR master mix and FMR1 and GUS Taqman probe-primers (Applied Biosystems). For siRNA knockdown experiment in lymphoblastoid cells, Neon™ Transfection System (Life Technologies) was used as per manufacturer's instructions. Briefly, 2 × 107 AZA-treated cells were pelleted and washed with PBS. The cell pellet was re-suspended in 200 µl of buffer R, and 72 µl of 20 µmFMR1 siRNA was added to it (Life Technologies). Ten microliters of cells with siRNA were used for each transfection at 1450 V for 2 × 20 ms. Each 10 µl was then plated in 1 well of a 12-well plate containing 2 ml of medium for a total of 27 wells. Cells were then allowed to grow for another 48 h. Cells were harvested for RNA and chromatin preparation. For fibroblasts cells, siRNA transfections were done using Lipofectamine® RNAiMAX reagent (Life Technologies) as per manufacturer's instructions. Briefly, cells were treated with 10 µm AZA for 3 days and then grown in drug-free medium. On Day 6, cells were split onto four 10-cm dishes. On Day 8, cells were transfected with either FMR1 siRNA or On-Target Plus control siRNA (ThermoFisher Scientific, Waltham, MA, USA) in 6 ml medium without antibiotics. For each transfection, 27 µl of Lipofectamine® RNAiMax reagent was diluted in 450 µl of Opti MEM® (Life Technologies) and 13.5 µl of 20 µm siRNA stock was diluted in 450 µl of Opti MEM® and mixed together and incubated at room temperature for 5 min. The mixture was then added dropwise to the cells, which were grown for 2 more days and harvested for RNA and chromatin preparation at Day 10.
For actinomycin D experiment, cells were treated with 10 µm AZA for 3 days. One day after AZA addition, either DMSO or 5 µm actinomycin D (Sigma) was added to the cells and the cells grown for 48 h. The medium was then changed to AZA-free medium, and fresh actinomycin D was added. The cells were grown for 3 more days and were harvested for RNA and chromatin preparation on Day 6.
Chromatin immunoprecipitation assays
Chromatin immunoprecipitation assays were performed as described before (12) using a ChIP assay kit from EMD-Millipore (Temecula, CA, USA). To prepare chromatin for immunoprecipitation, cells were fixed with 1% formaldehyde for 10 min at room temperature and lysed as per the kit manufacturer's instructions. The chromatin was sonicated to fragments of <500 bp using a Bioruptor® (Diagenode, Denville, NJ, USA). Real-time PCRs on the immunoprecipitated DNAs were carried out using the Power SYBR™ Green PCR master mix (Applied Biosystems). For amplification of the 5′ flanking region, the primer pair Upstream-F (5′ ACAGTGGAATGTAAAGGGTTG 3′) and Upstream-R (5′ GTGTTAAGCACTTGAGGTTCAT 3′) were used. This primer pair amplifies a region 876–736 bp upstream of the 3′ most transcription start site. For amplification of FMR1 exon1, the primer pair Exon1-F (5′ CGCTAGCAGGGCTGAAGAGAA 3′) and Exon1-R (5′ GTACCTTGTAGAAAGCGCCATTGGAG 3′) were used. This primer pair amplifies the region +236 to +312 relative to the transcription start site. GAPDH was amplified with primers hsGAPDH exon1F1 (5′ TCGACAGTCAGCCGCATCT 3′) and hsGAPDH intron1R1 (5′ CTAGCCTCCCGGGTTTCTCT 3′). MyoD was amplified with primers hsMYO-D exon1-F (5′ CCGCCTGAGCAAAGTAAATGA 3′) and hsMYO-D exon1-R (5′ GGCAACCGCTGGTTTGG 3′). For quantitation, the comparative threshold (Ct) method was used. Enrichment over input was calculated and where indicated was normalized to the positive control. For re-ChIP analysis, the chromatin was first immunoprecipitated with antibody against H3K27me3, divided into two aliquots and then subjected to ChIP with either control Ig or antibody against H3K4me2. Enrichment over input was calculated as earlier and the data plotted as the fold increase over control Ig. Student's t-test was used to calculate P-values (GraphPad Software, Inc., La Jolla, CA, USA).
This work was supported by a grant from the Intramural program of the National Institute of Diabetes, Digestive and Kidney Diseases, National Institutes of Health to K.U.
The authors thank Dr Bruce Hayward for his careful reading of this manuscript.
Conflict of Interest statement. None declared.