|Home | About | Journals | Submit | Contact Us | Français|
The DNA hypomethylating drug Decitabine (DAC) reactivates silenced gene expression in cancer and is approved for the treatment of the myelodysplastic syndrome. Gene reactivation after DAC is variable and incompletely understood. Here, we established a cell line system (YB5) derived from the SW48 colon cancer cell line to study DAC-induced reactivation. YB5 contains a hypermethylated CMV promoter driving GFP, and the locus is transcriptionally silent. GFP re-expression can be achieved by DAC treatment, but the expression level of individual cells is heterogeneous. DAC-treated YB5 cells were separated into GFP-positive and -negative sub-populations. By comparing DAC-treated sorted GFP-positive and -negative cells, we found that their methylation levels were similarly decreased, but that histone modifications and histone H3 densities were remarkably different. Despite a similar degree of (incomplete) DNA hypomethylation, GFP-positive cells reverted to an active chromatin structure marked by higher H3K9 acetylation, lower H3K27 tri-methylation, and lower promoter nucleosome density. GFP-negative cells had histone modifications and promoter nucleosome density similar to parental cells. Upon DAC withdrawal, gradual re-silencing and remethylation occurred in both GFP-positive and -negative cells, and the re-silencing correlated with a gradual increase in nucleosome occupancy in GFP-positive cells. These data demonstrate that hypomethylation alone after DAC is insufficient for gene expression induction, and that chromatin resetting to an active state including nucleosome eviction is required for activation of protein expression. Our findings suggest that gene expression is the key in optimizing DAC treatment strategies in the clinic.
Over 50% of all human promoters are contained in CpG islands (1), and many show aberrant hypermethylation in cancer, especially tumor suppressor genes (2, 3). Abrogation of hypermethylation and reversal of epigenetic silencing is recognized as a powerful tool in cancer therapeutics (4). The compound Decitabine (5-aza-2’-deoxycytidine, or DAC) is an approved drug in the treatment of myelodysplastic syndromes (MDS) (5), and is currently in clinical trials in a variety of solid tumors (6–8). However, the precise mechanisms and dynamics of gene reactivation after hypomethylation induction are not well-understood. DAC inhibits DNA methyltransferase activity and passive global hypomethylation is induced when the cells divide (9). Eventually, this hypomethylation restores gene expression at silent promoters (10). The required minimum degree of hypomethylation for gene reactivation is largely unknown. DNA methylation has been associated with an altered chromatin structure, presumed to be related to recruitment of histone modifiers by DNA methylation (2, 11). This has been proposed to explain the observed synergy at the gene expression level between DAC and histone deacetylase (HDAC) inhibitors (12). It has also been suggested that gene reactivation after DAC happens despite persistent or increased binding by polycomb group (PcG) proteins (13), which usually silence gene expression. Part of the difficulty in understanding the precise events after DAC exposure is the marked cell-to-cell and gene-to-gene heterogeneity in gene reactivation. This heterogeneity implies that, in most studies, mixed populations of expressing/non-expressing cells are examined.
To overcome this problem, we developed a cell line system whereby GFP expression is controlled by a methylated CMV promoter. After sorting GFP-positive and -negative cells, we observe that DAC-induced gene expression does not require full demethylation, and that induction of hypomethylation alone is not sufficient to activate gene expression. The essential determinant appears to be chromatin remodeling to an active state after hypomethylation induction. Our findings also provide the molecular basis for the variation in gene expression induction by hypomethylation and suggest the optimal use of DAC in clinics.
A patch methylated construct was modified from pEGFP-N1 (Clontech, CA). By in vitro methylation (SssI methylase, NEB, MA), restriction enzymes (PciI and BglII, NEB) digestion and re-lagation (T4 ligase, NEB), the final construct had the methylation restricted to the pCMV region (Figure 1a). The linearized patch-methylated construct was transfected into SW48 cells using Lipofectamine2000 (Invitrogen, CA). After G418 (Invitrogen) selection, GFP negative cells were selected out using FACSAriaII (BD Biosciences, NJ) for single cell cloning. The isolated single clones were further examined for transgene copy number using real-time PCR (Figure S2) and for genomic insertion site using inverse PCR (Figure 1b).
Both SW48 cells (ATCC, VA) and the derived single cell clones were cultured in IL-15 medium with 10% FBS and 100µg/ml streptomycin-penicillin. Daily treatment with 5-aza-2’-deoxycytidine (Sigma, MO) was adopted for GFP reactivation. GFP positive cell percentages were measured using BD FACS Calibur flow cytometer; GFP cell sorting was conducted using BD FACSAriaII. Post-sorting analysis was performed to assess the purity of the groups. Flow sorting and FACS data were processed using FlowJo (TreeStar, OR).
Extraction and bisulfite conversion of genomic DNA was carried out as described previously (14, 15). Pyrosequencing and bisulfite cloning/sequencing were used to study methylation levels. Primers/conditions are listed in Supplementary Table1.
Total RNA was extracted using Trizol (Invitrogen), the residual genomic DNA was removed using DNA-Free™ kit (Ambion, TX), and cDNA was synthesized using High-Capacity cDNA Kit (Applied Biosystems). Quantification of cDNA was done by real-time PCR using ABI Prism 7900HT system. All cDNA products were amplified with the Universal PCR Master Mix (BioRad, CA) and performed in triplicate. GAPDH was used as a reference gene. Primers/probes are listed in Supplementary Table1. 5'-Rapid amplification of cDNA ends was performed using a 5'-RACE kit (Invitrogen) to determine the TSS of GFP gene. Total RNA from 100nM DAC-treated YB5 cells was used as template and the GFP specific primers are listed in Supplementary Table1.
ChIP analyses were performed as described previously (16). Cells were 1% formaldehyde fixed and lysed followed by sonication shearing using the Biorupter (Diagenode, Belgium). Antibodies used were: anti-histone H3 (ab1791, Abcam, MA), anti-histone H3K9-acetylation (07-352, Millipore, MA), anti-histone H3K27-tri-methylation (07-449, Millipore), and anti-IgG (ab46540, Abcam). The value of each histone mark was determined by H3 and IgG normalization following the equation: Enrichment = 2[Ct(H3)-Ct(Ab)]-2[Ct(H3)-Ct(IgG)]. The value of histone H3 protein enrichment was determined by input control. 1% of chromatin was used as input control. Quantification of ChIP DNA was done by real-time PCR, and primers/probes are listed in Supplementary Table1.
Total histones were prepared by acidic extraction and resolved on 15% SDS-polyacrylamide gels as described (17). Additional antibodies used were: anti-pan-acetylated histone H4 (06-866, Millipore), anti-histone H4 (07-108, Millipore).
We started deriving a DNA methylation reporter assay by transfecting an in vitro methylated CMV-GFP transgene into the colon cancer cell SW48, which has intense hypermethylation of multiple genes characteristic of the CIMP subtype of colon cancers (19). CMV promoter is over 500bp in length and includes 30 CpG sites with a CpG percentage of 6%; the ObsCpG/ExpCpG ratio is 0.89 and the GC content is 50%. Thus, the CMV promoter is a classical CpG island following Gardiner-Garden and Frommer’s criteria. The outline of generating a patch-hypermethylated plasmid and transfection into SW48 is presented in Figure 1a. After selection, sorting and single cell cloning, we tested multiple isolates for the required characteristics (integrated intact transgene, silenced gene expression) and characterized one, YB5, in detail. We used qPCR to determine that the transgene dosage in YB5 genome was one (Supplementary Figure S2). Copy number did not change over a time period of up to 15 months (data not shown). We next used inverse PCR to determine the integration site. The resulting PCR product contained a 774bp-long sequence with 100% homology to position 73061660-73062433 of the minus strand on Chromosome 1 in the UCSC BLAT database (hg18, March 2006) (Figure 1b). Thus, the transgene integrated into an intragenic region of human EST CD655906 on chromosome 1p31.1. We also used a GFP expressing clone, YB11, which contained one copy of CMV-GFP transgene at chromosome 19p13.3 region as a positive control for subsequent experiments.
We used quantitative bisulfite pyrosequencing and bisulfite cloning/sequencing to study the DNA methylation state of the transgene in detail. Bisulfite pyrosequencing revealed that the CMV promoter in the GFP expressing clone YB11 was unmethylated, while the silenced clone YB5 was hypermethylated from −337 bp to +19 bp of the transcription start site (Figure 2a). This region covers the core CMV promoter and contains 22 CpG sites with an average methylation level over 80%. Analysis of early and late cell passages of YB5 demonstrated that the methylation pattern is stable. The hypermethylation pattern was also confirmed by bisulfite cloning/sequencing using another set of PCR primers (Figure 2b). Almost every CpG site had very high levels of DNA methylation, with the exception of two CpG sites that correspond to CREB binding sites indicated by Genomatix Software analysis (München, Germany). Interestingly, we detected no binding of CREB or phospho-CREB to the region in the YB5 cells, while binding was detected in YB11 by ChIP assays (Supplementary Figure S3).
Next, we investigated the impact of CMV hypermethylation on the expression of GFP gene. Using qRT-PCR, we found robust GFP expression in YB11, while no GFP mRNA in YB5 and SW48. Using the hypomethylating agent DAC at different concentrations (from 10 nM to 200 nM), the YB5 GFP gene could be reactivated in a dose-dependent way (Figure 3a). Because DAC can have effects independent of DNA methylation (20), we also tested whether gene reactivation can be achieved by directly inhibiting DNMT1, the main DNA methyltransferase. Using Dnmt1 siRNA knockdown, we found efficient GFP reactivation compared to the scrambled control. By contrast, knockdown of EZH2 had no effect on GFP reactivation (Supplementary Figure S4). Further more, to determine whether the DAC reactivated GFP transcript is driven by the CMV promoter, a 5’RACE experiment was conducted, which confirmed that the GFP mRNA starts from the 3’end of the CMV promoter. Thus, the integrated transgene is silenced by DNA methylation and mimics endogenous tumor suppressor gene silencing in cancer.
We next used FACS flow cytometry to analyze GFP reactivation. As can been seen in Figure 3b, SW48 and YB5 cells stayed at the GFP negative region and YB11 fell into the GFP positive region, while the 100 nM DAC treated YB5 cell exhibited a distribution shift to the GFP positive region. The percentages of GFP positive cells under the gradient DAC treatment are summarized in Figure 3c. The correlation of GFP percentages from FACS analysis with mRNA expression levels determined by qRT-PCR was analyzed and found an R2 value of 0.97 (Figure 3d). Taking advantage of the single cell detection of GFP expression enabled by flow analysis, we observed a large difference in expression levels of individual cells originating from an identical cell population after a demethylating stimulus. Moreover, the proportion of cells positive for GFP did not exceed 40%, even with increasing doses or duration of exposure. Previous studies analyzed DAC treated cells as a uniform population, whereas the single cell analysis captures this very heterogeneous behavior.
To explore the mechanisms underlying heterogeneity in gene expression after DAC treatment, we separated GFP-positive and GFP-negative sub-populations using cell sorting cytometer (Figure 4a). Post-sorting analysis revealed that the purity of sorted populations was about 85% in GFP-positive sub-population and 98% in GFP-negative sub-population. The sorted cells were collected and methylation levels were studied. The first question we asked is whether the difference in gene expression is simply due to lack of hypomethylation in a sub-population of cells, perhaps related to quiescence at the time of exposure. The CMV promoter methylation was first investigated using pyrosequencing and found that the GFP-positive cells demethylated from 81.4+/−0.2% to 45.4+/−0.2% and the GFP-negative cells to 51.1+/−0.1% (Figure 4b). We next used bisulfite cloning/sequencing on these cells and found that DNA methylation is mosaic and overall similar between positive and negative cells (Figure 4c). Besides the CMV promoter, we also examined global methylation status using the Long Interspersed Nucleotide Element-1 (LINE1) repetitive element (22, 23), and we found that global methylation decreased from 58.0+/−0.3% to 39.0+/−0.4% in GFP-positive cells and to 41.9+/−0.1% in GFP-negative cells. Similarly, we found no differences in methylation levels of tumor suppressor genes P16INK4a, CDH13, RASSF1a, RARβ2, and PGRB between sorted sub-populations (Figure 4d, left panel). The expressions of P16INK4a, RASSF1a and PGRB were measured, and P16INK4a and RASSF1a were reactivated by DAC while PGRB was not. Similar to GFP, the expressions of P16INK4a and RASSF1a were higher in GFP-positive cells than -negative cells (Figure 4d, right panel). These data suggest that a decrease in methylation may be necessary but is not sufficient for gene reactivation after DAC; other critical events must be involved. Cell cycle distributions of GFP-positive and -negative cells were analyzed, but no differences were found (data not shown).
To ensure that our results are not strictly due to the presence of hemi-methylated DNA, we repeated the experiment with a one-time DAC treatment, and we still saw partial methylation associated with transcription and a relatively small difference between GFP-positive and -negative cells (Supplementary Figure S5).
Because chromatin structure is also critical to regulate silencing and gene expression in mammalian cells, we examined histone modifications in parental cells and DAC-treated GFP-positive/-negative sub-populations. Several modification marks were investigated using ChIP assays, including histone H3 lysine9-acetylation (H3K9ac), lysine4-trimethylation (H3K4me3), lysine9-trimethylation (H3K9me3) and lysine27-trimethylation (H3K27me3). Three regions along the CMV-GFP locus were studied, including the promoter region (-200bp), transcription start site region (TSS) and GFP coding region (+400bp).
The parental YB5 cells exhibited a closed chromatin structure, devoid of H3K9ac and enriched for H3K27me3, whereas the expressing YB11 cells were just the opposite. Post-DAC, the hypomethylated GFP-negative cells were similar to parental cells, retained low H3K9ac and high H3K27me3 levels, while the GFP-positive cells showed chromatin altering to an active state, with 2.5–5 fold higher level of H3K9ac and 2.5–8 fold lower H3K27me3 comparing to the -negative cells. Though histone H3K4me3 and H3K9me3 were different between YB5 and YB11 cells, they were not found to be distinguishable in GFP-positive and -negative cells. Also, the ChIP assay did not show binding of CREB in either GFP-positive or GFP-negative cells (data not shown). Interestingly, the histone H3 densities at the promoter and TSS regions were found to be very different between GFP-positive and -negative cells. The GFP-positive cells showed H3 loss suggesting promoter nucleosome eviction, while GFP-negative cells retained most of the histone H3 of the parental YB5 cells (Figure 5a). By contrast, we performed western blot analysis and found no differences in global histone marks between GFP-positive and -negative cell populations (Figure 5b).
To verify that the active chromatin state could occur despite residual DNA methylation, we performed bisulfite pyrosequencing on DNA immunoprecipitated with histone H3K9ac and histone H3K27me3 antibodies. We found that the H3K9ac-bound DNA is 31% methylated, and H3K9ac-unbound DNA is 56% methylated; while H3K27me3-bound DNA is 59% methylated, and H3K27me3-unbound DNA is 54% methylated. The methylation level is the average of seven CpG sites located from-104bp to -41bp from the TSS. These data show that at the CMV promoter locus, H3K9-acetylated histones can be present on methylated DNA, but shows a preference to less methylated DNA; H3K27-tri-metylated histones have no particular DNA methylation preference (Supplementary Figure S6).
HDAC inhibitors can increase global histone acetylation; we therefore asked if altering histone marks by treatment with the HDAC inhibitor TSA would promote the expression of the sorted GFP-negative cells. Low doses of TSA showed no effect on the fully methylated YB5 cells. As shown in Figure 5c, 24 hours after sorting, the hypomethylated but initially GFP-negative cells had about 12% GFP-positive cells, likely representing the continuing effects of DAC. TSA treatment increased this number to 46%, which indicated the synergy was achieved at the GFP locus. As expected, in GFP-positive cells, TSA had little effect on boosting gene expression further. Using ChIP assays, we confirmed that TSA treatment elevated histone H3K9ac at the CMV locus, while decreased histone H3K27me3. However, the post-TSA treatment did not affect histone H3 density in GFP-negative cells (data not shown).
After DAC induced hypomethylation, gene remethylation is the norm (24, 25), though the mechanisms of this phenomenon are unknown. It has been proposed that a residual closed chromatin state predisposes to remethylation and that high levels of gene expression may protect against remethylation (26). This important question could be addressed using this model. To do this, we cultured the sorted cells separately and followed DNA methylation over time. The percentage of GFP positive cells over a 48 day period is shown in Figure 6a and the methylation at each time point is shown in Figure 6b. As can be seen, after DAC withdrawal, GFP positive cells decreased in a two phase fashion: a rapid reduction at the first several days (up to day 5) and a slow decrease later on. Remethylation occurs in both populations, and the rate of remethylation is identical in sorted GFP-positive and -negative cells. We also tested the effect of prolonged TSA treatment, but it would not prevent remethylation of either GFP-positive cells or GFP-negative cells (data not shown).
Because gene expression appears to decline more rapidly than DNA remethylation, we examined chromatin changes focusing on the nucleosome occupancy of promoter and TSS regions. Histone H3 densities of GFP-positive and -negative cells at day 2 and day 5 were examined. Combining the day 0 data from Figure 5a, the nucleosome recovery kinetics is shown in Figure 6c. It is found that an increase in histone H3 occurred rapidly after DAC withdrawal, and the speed in GFP-positive cells was faster than that of GFP-negative cells. This nucleosome reassembly in GFP-positive cells appears to be correlated with the rapid phase of GFP re-silencing. Collectively, the DAC-induced chromatin remodeling is short lived. The rate of DNA remethylation was not dependent on the expression levels or chromatin structure post DAC and thus may relate to the residual DNA methylation marks near the CMV-GFP locus.
In this study, we describe a new model system YB5, which contains a hypermethylated and silenced CMV driven GFP gene with stably inherited epigenetic properties established over time. In this model, the promoter is DNA hypermethylated and has a closed chromatin structure characterized by histone H3K9 deacetylation and H3K27 hypermethylation. Expression is suppressed at the transcriptional level and can be restored by using the demethylating agent Decitabine or by inhibiting DNMT1 expression. Thus, this system mimics many of the features of typical gene silencing in mammalian cells, including cancer cells. CMV is a strong CpG island containing promoter, and it can be efficiently silenced as previously reported (27, 28). Because reactivated gene expression can be easily visualized and selected for, this model allowed us to ask important questions regarding the minimal requirements for gene reactivation, as well as to track re-silencing after epigenetic modulation. These questions have previously been investigated primarily from the perspective of entire (mixed) cell populations, though one study used sub-cloning and gene expression selection to ask questions about remethylation (9).
We first asked about the association between hypomethylation induction and gene reactivation. Surprisingly, methylation studies showed similar demethylation levels between GFP positive cells and negative cells, which suggests that pharmacological uptake of DAC is uniform and not rate-limiting in these cells. In sorted cells that had high GFP expression, DNA methylation did not decrease to normal levels (nearly zero) suggesting that a previous hypothesis about heterogeneous reexpression being explained by a mixture of cells demethylated to 0% (and reexpressing) while others remain hypermethylated and silenced is incorrect. In fact, only 6 of 24 alleles sequenced had complete demethylation in these cells with high GFP expression. Thus, only a moderate degree of DNA hypomethylation is required for gene reactivation. In our experiments, we ruled out activation of an alternative host promoter as a possible explanation. However, further experiments may be needed to verify that methylated CpG islands can still have promoter activity if they retain an open chromatin. Bisulfite-sequencing analysis of H3K9ac histones showed about 30% DNA methylation, confirming that the gene can be activated despite residual DNA methylation, but also suggesting that unmethylated alleles are enriched in open chromatin. Moreover, in our data the GFP negative cells after DAC also had decreased DNA methylation levels, demonstrating that DNA hypomethylation, per-se, is not sufficient for gene reactivation. Rather, the main molecular difference between GFP-positive and GFP-negative cells post-DAC was the differential histone modifications and histone H3 densities revealed by ChIP assays. Thus, chromatin resetting is the prime factor determining gene reexpression state after DAC induced DNA demethylation. Interestingly, it has previously been reported that DAC induces nuclease sensitivity changes in the human HPRT gene locus prior to gene reexpression (29), and decreased nucleosome occupancy of hMLH1 promoter after DAC (30), which are consistent with our data.
It is interesting to ask why a relatively small degree of DNA demethylation would naturally, but not uniformly, bring histone changes and chromatin resetting. A possible explanation is that in the absence of DNMTs, the coupled assembly of newly synthesized histone octamers during cellular replication may be disrupted. The nascent histone octamers lose some original repressive histone tail marks, the promoter nucleosome assembly is disabled and the chromatin alters to a locally open structure, resulting in transcription of this strand of DNA (31, 32). Indeed, DNMT1 and DNMT3b have been reported to bind to HDACs, histone methyltransferases and chromatin scaffold proteins, and it is possible that this binding is important to replication of histone marks. It is interesting to note that after withdrawal of DAC, about 12% of sorted GFP negative cells transform to expressing cells after 24 hours in regular growth medium, which indicates the continuing chromatin resetting. This model also explains the observed synergistic effect between DNA demethylating agents and low-dose histone deacetylases inhibitors (12).
The cell sorting strategy also allowed us to address the vexing problem of gene resilencing and remethylation after DAC withdrawal. Using a mixed cell population, it was previously reported that the repressive histone mark H3K27me3 persists or increases after DAC treatment (13), and thus serves as a nidus for resilencing. Our data using purified cells are not consistent with these observations. Rather, we find that resilencing and remethylation is independent of gene expression levels or of local chromatin structure. The kinetics of DNA remethylation is quite flat, which can be cell division related (33). However, loss of expression after DAC withdrawal is very rapid in the first several days (seen at day 2 and day 5). Notably, the speed of histone H3 gains up to day 5 is also prompt, which appears to be coincident with rapid GFP loss. Thus, it is highly likely the first re-silencing is due to the reassembly of nucleosomes, though the driving force of re-packaging DNA is unknown. It is possible that a closed chromatin configuration persists in-cis away from the CMV promoter and underlies gene resilencing.
It remains to be seen how much the data we have generated with this in vitro system are applicable to endogenous gene silencing. All indications are that they would be – the transgene is stably integrated, shows the typical methylation-associated closed chromatin configuration and is reactivated by Decitabine or DNMT1 knock-down with similar kinetics and patterns as endogenous genes. Nevertheless, it is likely that there will be gene or locus specific events that influence reactivation. Indeed, some genes are silenced without detectable H3K27me3 (33), and those may behave differently. Similarly, promoters vary in their CpG density and degree of DNA methylation, and this could influence gene reactivation patterns.
Our data have clinical implications for the use of DNA methylation inhibitors. In treated patients, relatively small decreases in DNA methylation were observed (34, 35), but these were accompanied by significant gene reactivation and clinical responses. Moreover, as would be predicted from the current data, gene reactivation was a better predictor of response than hypomethylation induction (36, 37). Just like the in vitro situation, gene remethylation was observed, and a higher extent of remethylation was associated with resistance to therapy (35).
In summary, we found that DNA hypomethylation is necessary but not sufficient for gene reactivation after DAC. Rather, local chromatin structure resetting, which can happen at a low level of DNA demethylation, is a key determinant of actual gene re-expression. These data have implications for the use of hypomethylating drugs in the clinic. In addition, the YB5 system can be useful for evaluating potential demethylating compounds and epigenetic synergy studies to boost gene reactivation as well as prevent resilencing and remethylation.
This work was supported by National Institutes of Health grants CA100632, CA098006, and CA121104. JPI is an American Cancer Society Clinical Research professor supported by a generous gift from the F. M. Kirby Foundation. DNA sequencing and Flow Cytometry in the respective cores at the M. D. Anderson Cancer Center is supported by Core Grant CA16672 from the National Institutes of Health.