The involvement of epigenetic factors, particularly DNA methylation on the regulation of gene expression has been recognised for quite some time, however it is a process not totally understood. The events which underlie genomic hypomethylation and hypermethylation of tumour suppressor genes in malignant cells and why these types of genes are targeted remain unresolved. HPLC analysis of genomic DNA was performed to assess the level of total methylation content present in several types of CRC cells. Genome wide hypomethylation was observed in all the CRC cell lines analysed when compared with the fibroblast cell line, and the lower levels of global methylation in the LoVo and HT29 cell lines may contribute to the genomic instability observed in these cell lines. This finding is consistent with previous reports that malignant cells have lower levels of genomic DNA methylation when compared with healthy tissues [
23,
24]. The de-methylation observed on a global level did not correlate with that found at specific CGIs, and consequently, expression was observed from genes associated with a hypermethylated promoter. The discrepancy between global and specific gene methylation levels can be explained by the reduced methylation level of Alu elements and LINE methylation. Yang and colleagues [
25] demonstrated Alu elements and LINE methylation is reduced by 16% and 60% respectively in colorectal cancer cell lines treated for 72 h with 5-aza-dC. Repetitive regions are likely to fall within compacted heterochromatin where methyltransferase access to the DNA is limited and as a result these sequences become more readily hypomethylated. Furthermore, methylation reductions in a combination of transposable elements [
5], satellite repeats [
26] and other methylated GC rich areas of the genome which do not form
bona fide CpG islands [
27] may also contribute to this difference.
Global methylation levels and response to 5-aza-dC and TSA
In all cell lines studied significant genome-wide de-methylation was observed, and a greater than 50% reduction was observed in SW48, SW480 and HCT116 cells, indicating significant DNA de-methylation. Amongst the cancer cells, higher pre-treatment levels of global methylation appeared to correlate with a larger decrease in global methylation levels. Even with exposure to a high concentration of 5-aza-dC, global methylation levels following treatment averaged ~1.2% in the colorectal cancer cell lines. This may be interpreted to suggest that if total methylation levels fall below this point, cell death may be induced either through de-methylation, or toxicity from excessive DNMT binding to DNA. Further analysis of DNA with genome wide methylation arrays might reveal regions that are common to all cell lines which are resistant to DNA de-methylation induced by 5-aza-dC. This possibility was not examined in the current study.
By the tenth day of drug free growth, the global methylation of all cell lines had increased and was nearing pre-treatment levels. The rate of re-methylation was steady in the SW480, LoVo, HT29 and fibroblast cell lines, whilst periods of rapid re-methylation were observed in HCT116 and SW48 cell lines (Figure
1). Despite genomic methylation increasing, the rate was reduced dramatically after this point which may suggest levels have, or could, plateau below the original level. This scenario could be indicative of an altered pattern of gene expression following 5-aza-dC treatment. The HCT116 and SW48 cell lines are known to harbour epigenetic changes, and were among the most affected by 5-aza-dC which may suggest regulation of DNA methylation in these cell lines is different from the others studied.
Continual exposure of the HCT116 cells to Trichostatin A was performed in combination with regular 5-aza-dC treatment to investigate whether altering histone acetylation and chromatin conformation could affect the DNA methylation process and subsequent gene expression patterns. TSA is an inhibitor of Histone Deacetylase and leads to histone hyper-acetylation. Histone acetylation is associated with regions of active chromatin [
28] and has been shown to assist the binding of a the TFIIIA transcription factor to chromatin templates [
29]. With this knowledge, an investigation into whether global de-methylation could be enhanced and if re-methylation could be restricted by histone hyper-acetylation was undertaken. Our results show that at a genome wide level, TSA did not enhance the de-methylation process in HCT116 cells, and continual exposure to TSA for ten days did not significantly alter the re-methylation process (Table
2) in the HCT116 cell line. The p-value of 0.06 indicates the largest difference is at the d0 time point, however at this time the TSA treated cell line had a greater level of global methylation, and therefore did not enhance de-methylation. Based on these observations, DNA methylation is not hindered by histone acetylation. This notion does not conflict with previous findings that 5-aza-dC and TSA have a synergistic effect on gene expression [
17,
30] rather it indicates methyltransferase enzymes are not deterred from hyper-acetylated DNA.
Expression arrays were conducted with the aim of identifying reactivated genes that were differentially expressed between the cell lines and could be subjected to bisulfite sequencing analysis. In addition, the influence of TSA on expression patterns could be observed. Combined treatment of TSA with 5-aza-dC did not cause an increase in the number of reactivated genes which is in accordance with its minor influence on genomic methylation levels. Again, this does not conflict with previous reports of a synergistic effect of the two drugs, but indicates DNA methylation plays a greater role than histone acetylation in reactivating silenced genes on a genome wide level [
20]. The synergistic effect of TSA was observed with prolonged expression of 165 genes deemed temporarily reactivated with 5-aza-dC alone, implicating TSA treatment as a valuable tool for maintaining expression of genes reactivated with 5-aza-dC.
The combined use of 5-aza-dC and TSA may be advantageous in overcoming poor outcomes in tumour types that do not respond to 5-aza-dC alone. A 'maintenance' administration of TSA following a 5-aza-dC treatment cycle may assist with prolonged gene expression without the cytotoxic effects of 5-aza-dC. Furthermore, identification of the pre-existing methylation patterns at genes targeted for reactivation could determine whether that gene will respond to treatment and whether a particular patient is suitable for this type of therapy.
Bisulphite sequencing of CpG islands
Bisulphite conversion of DNA followed by PCR and direct sequencing across a CGI permits quantification of the methylation at individual CpG sites and allows for the establishment of a methylation profile of CpG islands. The genes studied displayed varying levels of responsiveness to 5-aza-dC treatment, as observed by a decreased CGI methylation, however the decrease was not consistent with that observed on a genome wide level. The largest decrease in gene specific methylation was in the RNF113B gene. After treatment, CGI methylation levels dropped by over 20% in the HT29 cell line with an associated increase in gene expression. By the tenth day of drug-free growth the CGI methylation returned to pre-treatment levels, which correlated with the return of normal levels of gene expression.
Sequencing of CGIs allowed the detection of several instances where a small cluster of cytosines were hypomethylated amongst an otherwise hypermethylated CpG island in a non-expressed gene. These hypomethylated cytosines appear at a CpG site adjacent to the TSS of a gene, as seen in the MAGEA3 CGI in SW480 cells. Upon culturing the cells for a further four and ten days in drug free media, these cells were found to still be expressing the previously silenced MAGEA3 gene, suggesting the transcriptional status of this gene had been permanently reversed. In comparison, the HCT116 cell line expressed these two genes constitutively. The only common methylation pattern amongst the two cell lines was <50% methylation at the CpG sites 1 bp upstream and 10 bp downstream of the TSS in the MAGEA3 gene. This kind of reactivation was also observed in CDO1, and HSPC105 genes in SW480 cells which also carried hypomethylated CpG sites.
To investigate the possibility that the enduring reactivation of genes was due to continued cytotoxicity or apoptosis, we monitored these levels over the corresponding time period. Both cytotoxicity and apoptosis levels were elevated by 5-aza-dC, but decreased when the drug was removed. This indicates gene expression variation is a result of changes in genomic methylation rather than activation of apoptotic pathways. The fraction of necrotic cells remained higher than in untreated cells which are likely to represent reactivated genes that orchestrate cell death or cells that died in the time prior to the measurement of apoptosis.
Treatment of cells with 5-aza-dC caused the reactivation of numerous genes, although de-methylation within the CGI of specific genes did not correlate with genomic levels (as discussed earlier) or transcription levels. Consequently we observed that 5-aza-dC induced expression can be driven from a largely methylated promoter with localised demethylation at the TSS. One such example is the
CDO1 gene in SW480 cells (Additional File
7: Figure S7), where ~10% de-methylation (ie 10% of all alleles) at three CpG sites caused a greater than 1000-fold up-regulation of CDO1 mRNA.
We envisage the significant increase in expression of CDO1 is inflated due to non-existent expression in untreated cells, and that modest expression is permitted due to hypomethylation in a small proportion of cells at the TSS. As methylation of surrounding CpG sites was largely unaltered, demethylation of three CpG sites in a 50 bp region surrounding the TSS is permissive of transcription, and does not require hypomethylation of the entire allele. A similar finding has been previously reported in
CDKN2A in cervical carcinogenesis [
31] and also in the
CDH1 gene in [
32]. The minor demethylation of
CDH1 in conjunction with large increase of gene expression may suggest that DNA methylation does not repress transcription, and gene up-regulation could be due to either a secondary effect or the involvement of other factors which are also modified by 5-aza-dC treatment.
Clusters of CpG sites were identified in some genes that showed a region of hypomethylation, such as
MAGEA3 in HT29 cells (Additional File
4: Figure S4) and
ZFP3 in the SW480 cells (Additional File
8: Figure S8). Expression of these genes was detected before 5-aza-dC treatment, demonstrating the importance of methylation in the CpG sites around the TSS, and how a region of hypomethylation is permissive of transcription. Methylation of ~50% was detected at some CGIs including
CDKN2A in HCT116 cells. These genes are likely to show mono-allelic methylation which has been reported previously [
33,
34], where all expression is from one non-methylated allele.
The
MLH1 gene is silenced by methylation on both alleles in SW48 cells [
35]. Following 5-aza-dC treatment, de-methylation was induced and the gene was re-expressed and by the tenth day of drug free growth, the gene was still expressed despite methylation returning to pre-treatment levels. The expression level of
MLH1 at day 10 may be due to lower methylation at a CpG site 32 bp downstream from the TSS, which is similar to the long term reactivation in the
CDO1,
MAGEA3 and
HSPC105 genes in SW480 cells. The analysis undertaken suggests that genes thought to be under control of epigenetic modifications such as
CDKN2A and
MLH1 were not shown to have significantly altered patterns of CpG methylation following 5-aza-dC treatment, although an increase in expression was observed.
In non-expressed genes, the identification of regions of hypomethylated cytosine in a generally hypermethylated CpG island raises the question of how these cytosines are maintained in a hypomethylated state. As the region of hypomethylated cytosines in these genes is less than 146 bp - the length of DNA associated with a nucleosome, it would suggest in this scenario at least, an absent nucleosome is not a factor in assisting transcription as previously described [
36]. It would appear 5-aza-dC can induce an irregular nucleosomal conformation that permits expression from methylated genes. It is also possible that 5-aza-dC reduces repressive histone tail marks or initiates the gain of active marks in the area surrounding the hypomethylated cytosine following 5-aza-dC exposure. The methylation of Histone H3 Lysine 9 is a modification associated with inactive chromatin and has been shown to be rapidly lost with 5-aza-dC treatment [
37]. It is possible that a loss of H3K9 methylation may have been induced by 5-aza-dC in the current study, which may allow binding of transcriptional proteins, which in turn enhance transcriptional reactivation.
Hypomethylated CpG sites which exert control on gene expression may have implications for methods such as methylation-specific PCR and array technologies which rely on the methylation status of a small number of CpG sites in order to determine a given genes methylation status. The observation that a single or small group of CpG sites could affect expression may have greater implications should a polymorphism exist at the site. A polymorphism at the CG dinucleotide will deny methyl-group attachment which would be advantageous to individuals with that change by conferring a protective effect against epigenetic inactivation, particularly in genes such as MLH1.
Our results show that 5-aza-dC induces gene expression, but is not necessarily dependant on DNA de-methylation. The pre-existing level of methylation surrounding the transcriptional start site of a gene appears important in long term reactivation. This was demonstrated as the transcriptional status of silenced genes could be reversed with 5-aza-dC and for up to ten days after its removal in genes with hypomethylated cytosines despite minimal CpG de-methylation. In CGIs which exhibited a uniform level of hypermethylation, transcription could be induced for a short term only. The acetylation of histones was not found to alter the de-methylation or re-methylation process and was therefore not expected to cause a change in the gene expression profile. Although constitutively expressed genes demonstrate hypomethylation, 5-aza-dC treatment was found to force expression from genes with hypermethylated CGIs, suggesting that 5-aza-dC is capable of influencing other factors involved with gene expression, such as proteins with a methyl-binding domain or histone modifications.