Cell Culture and Reagent

SKM-1 (Japanese Collection of Research Bioresources Cell Bank, JCRB0118) cells were cultured in RPMI-1640 (Gibco) medium containing 10% fetal bovine serum (JRH Biosciences), 100 U/mL penicillin and 100 µg/mL streptomycin (Meiji). All cells were culture in an environment of saturated humidity, 5% CO2, and 37° C. Stock solutions (1mM) of 5-azacytidine AZA and 5-Aza-2’-deoxycytidine DAC (Wako chemicals) were stored at -80° C and diluted in tissue culture medium to the required concentrations before being added to cells in the exponential phase of growth.

Cell Proliferation assay

SKM-1 cells were seeded into 6-well plates at 1x10⁵ cells, 24 hours before drug treatments. Cells were treated daily with AZA and DAC at final concentrations of 0nM (mock-treated control), 0.1nM, 1nM, 10nM, 100nM, and 1µM for 7 days with medium changes every 24 hours. Three hours before plate reading, 20µl of CellTiter 96 AQueous One Solution Cell Proliferation Assay (MTS) (Promega) was added into a 96-well plate containing 100µl of the treated samples. Cell proliferation was measured by OD 490nm and the relative cell proliferations were normalised by mock-treated control.

Prolonged AZA/DAC treatment and western blotting analysis

Stocked AZA and DAC were added daily to cultures of SKM-1 cells for 4 independent experiments performed at two separate times. For sets 1 and 2, 2x10⁵ cells were cultured in 6-well plates with medium changed every 24 hours. For sets 3 and 4, 1x10⁶ cells were cultured in 90mm Petri dish plates with medium changed every 48 hours. Treatments were stopped at day 28 and the cells were grown in the absence of drugs for 10 days. Samples were collected on days 7, 14, 21, 28 and 38 for both genomic DNA and total mRNA purifications (DNeasy Blood & Tissue Kit and RNeasy Mini Kit, QIAGEN).

Protein extracts were also obtained from sets 3 and 4 after 28 days of treatment. Cells were lysed in RIPA buffer containing Halt Protease and Phosphatase Inhibitor Single-Use Cocktail (Thermo Scientific) for 5 minutes on ice. Whole-cell lysates were prepared in denaturing SDS sample buffer and subjected to 8% SDS-PAGE. Proteins were transferred to Immobilon-P membrane (Millipore, Billerica MA) and the blots were blocked with 5% non-fat dry milk in TBS-T buffer (50 mmol/L Tris-HCl, 200 mmol/L NaCl, 5% Tween 20). We used the following primary antibodies: anti-beta Actin (ab8226, 1:5000 dilution, abcam) as loading control, anti-Dnmt1 (ab13537, 1:1000 dilution, abcam), anti-Dnmt3a (ab13888, 1:1000 dilution, abcam), and anti-Dnmt3b (ab13604, 1:1000 dilution, abcam). Horseradish peroxidase conjugated secondary antibodies (NA931V, 1:10000 dilution, GE Healthcare Life Sciences) were detected using ECL prime western blotting detection reagent (GE Healthcare) and the Light-Capture system (ATTO).

Methylcytosine fractionation assay

Global DNA methylation was assayed essentially as described [17] by comparing microarray signals from fragmented DNA with or without additional digestion with McrBC, an enzyme which cuts DNA between pairs of methylated CpG dinucleotides separated by 40-3000 bases of intervening DNA. Briefly, one microgram of intact genomic DNA from each sample was sonicated to generate DNA fragments sized between 8 and 10kb. The fragmented DNA was then optionally digested with 50 units of McrBC (NEB) for 16 hours at 37 degrees. Gel electrophoresis was used to select fragments ranging from 1 to 3kb and whole-genome amplification PCR (GenomePlex^® Complete Whole Genome Amplification Kit, SIGMA) was performed for 14 cycles with 20ng of eluted DNA samples.

CpG-islands array analysis

DNA labelling and hybridisation was performed according to the supplied protocol (Agilent Microarray Analysis of Methylated DNA Immunoprecipitation Version 1.0). For each cell sample, 2.5 µg of McrBC- and mock-digested DNA were labelled with Cy5 and Cy3 respectively. Equal amounts of labelled samples were mixed and applied to Human CpG Island Microarrays (G4492A, Agilent). Methylation levels were estimated from the log (base 2) of the ratio of the intensity of signal from the undigested to digested DNA. Data was analysed by Agilent Genomic Workbench 5.0 and statistical analyses were performed using Bioconductor [18] and custom R code.

Bisulfite sequencing and pyrosequencing

DNA (1µg) from each sample was bisulfite converted using sodium bisulfite (EpiTect Bisulfite kits, Qiagen) following the conditions suggested by the manufacturers. Gene-primers for amplifying 15 regions from the HOXA cluster, GFI1, KCNC4, RALA, KAZALD1, CCND1, TGIF2, ZNF800, SIRT1, GTF2F1, TYROBP, IL8, and TNF were developed using MethPrimer [19]. PCR was performed using EpiTaq HS enzyme (Takara Bio), according to the manufacturer’s instructions. The PCR products were gel-purified (QIAquick Gel Extraction Kit, Qiagen) and cloned into the TA vector (TOPO-TA cloning, Invitrogen). 10 clones from each sample were selected and the sequences were determined using an ABI PRISM® 3700 Genetic Analyzer (Applied Biosystems). Data was summarised by a web-based software; QUMA (http://quma.cdb.riken.jp/top/quma_main_j.html) [20].

LINE-1 element and GAPDH promoter methylation were estimated using the PyroMark LINE-1 and GAPDH CpG Assay (Qiagen). Primers for pyrosequencing HSPA2, TNF, and TYROBP were designed by PyroMark Assay Design 2.0 software (Qiagen). Pyrosequencing was conducted using a PyroMark Q24 instrument (Qiagen) and methylation levels were quantitated with the PyroMark Q24 1.010 software. Relative peak height differences were used to calculate the percentage of 5-methylcytosines at each given site. Percent methylation within a sample was subsequently determined by averaging across all interrogated CpG sites in the analysis. All primer sequences for bisulfite sequencing are shown in Table S1 in File S1.

Gene expression analysis

One microgram of total mRNA was used in cDNA synthesis with the RT² First strand Kit (Qiagen) or ProtoScript® M-MuLV First Strand cDNA Synthesis Kit (NEB). Hematopoietic Stem Cells and Hematopoiesis PCR Array (PAHS-054A, Qiagen) was used for profiling expression of 84 genes. Quantitative real-time PCR (QPCR) was performed by SYBR Green Mastermix (Applied Biosystems) on an Applied Biosystems 7900 or 7500 Real Time PCR system. Relative gene expression was determined based on the threshold cycles (Ct) of the genes of interest and the internal reference gene GAPDH [21]. Primer sequences for HSPA2, TNF, and TYROBP are shown in table S3 in File S1. For expression array analysis, two micrograms of total RNA were used to prepare biotinylated RNA using the Affymetrix One Cycle Target Preparation Protocol driven by T7-linked oligo(dT) primers. Samples were hybridized overnight to Affymetrix HG U133 Plus 2.0 arrays, scanned and processed using GeneChip Operating Software. Statistical analyses were performed using Bioconductor and custom R code. The eXintegrator system was used to visualise expression data and for selection of probe sets by internal probe co-variance.

Data Analysis

Gene coordinates and CpG island positions were based on the HG18 assembly and were obtained from data files provided by the Agilent Genomic Workbench 5.0 system. CpG island positions were re-defined as regions contiguously (with a max inter probe distance of 250 bp) covered by probes, and overlapping CpG islands as indicated by Agilent data files. The resulting CpG island, gene and probe coordinates were then used to define relationships between genes, CpG islands and probes using a series of Perl scripts (Table S2 in File S1). Probes were assigned to the nearest or overlapping CpG islands. CpG islands in turn, were defined as, 1) upstream: upstream of, but not overlapping with the transcriptional start site (TSS), 2) gene body: lying entirely within the transcribed region, 3) promoter: overlapping with the TSS and the 5' region, 4) downstream: 3' of the transcript, 6) transcriptional termination site (TTS): overlapping the transcribed region and the 3' region, 7) Complete: where the complete transcribed region was contained within a CpG island and 8) Distant: with a distance of more than 3000 bp from the nearest gene (Table S3 in File S1) (This numbering is based on a binary OR combination of: upstream 1, gene body 2, downstream 4 and distant 8). The resulting probe-gene mapping was used to define an Agilent probe to Affymetrix probe mapping. A single Affymetrix probe-set was selected for each gene based on the probe pair covariance estimated by the f-statistic (between sample variance / within probe-set variance) using the eXintegrator system (http://www.cdb.riken.jp/scb/documentation/), and these were mapped to gene ids as provided by Agilent using the DAVID online resource. A total of 13721 Affymetrix probe sets from the HGU133_Plus_2 were mapped to 183160 probes.

The analysis presented for figures 1 and 2 was based on a subset of probes selected by their maximum variance within replicate groups (mock, AZA and DAC treatments 4 samples each). The threshold variance, 2^-2.5, was chosen on the basis of a comparison of the distribution of variances within the replicate and mock replicate groups derived by an arbitrary permutation (Figure S1 and Table S4 in File S1), as well as by a manual inspection of log₂-ratio values. This selected 52915 out of a total of 198302 probes. To identify probes that were demethylated as a result of the treatment we first identified probes from this selection (52915) that were methylated in the control (mock) samples (mean log-ratio > 1.0). De-methylated probes were then identified as probes from this subset that scored higher than 0.5 for a simplified f-statistic (between group variance / sum of within-group variance) and had mean log₂-ratios below 1 in the treated sample. Over or under-representation of different classes (upstream, promoter, gene body, etc.) of CpG islands was tested individually by calculating the probability of the observed overlaps between the island classes and the demethylated probes using Fisher’s exact test as implemented by the R phyper function. Since we tested for both over and under-representation in the 7 different island classes we used a conservative p-value threshold of 0.001 (0.05/14 = 0.0034, but since the tests cannot be considered as independent we reduced this to 0.001, For the full set of p-values see Table S5 in File S1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Prolonged treatments of AZA or DAC in SKM-1 cells.

A, Cell proliferation was assayed using the CellTiter 96 Aqueous One Solution assay kit after treating SKM-1 cells for 7 days with different concentrations of AZA (red dotted line) or DAC (blue dashed line). The percentage of cell proliferation was calculated relative to the rate of proliferation in untreated cells, and obtained from the mean (± SEM) of three independent experiments. B, Probes targeting CGI regions were classified into hyper, intermediate or hypo methylation groups according to their log₂-ratios obtained from microarray analysis (See materials and methods). Most probes were hypomethylated (>80% in all cases). The number of probes classified as intermediate or hypermethylated were reduced after AZA or DAC treatments. C, Principal Component Analysis (PCA) with data from 4 independent treatments (from 52915 probes selected by within-replicate variance).

https://doi.org/10.1371/journal.pone.0071099.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Selection of probes demethylated after AZA or DAC treatments.

A, Schematic for classification of probes. Of 52,915 probes with low within-replicate variance, 2217 probes had mean log₂-ratios larger than one in the control samples indicating more than 50% CG methylation. Probes whose between group variance was at least 0.5 times that of the sum of within group variances and whose log₂-ratios was less than 1.0 after treatment were considered as demethylated. B, Distribution of gene feature classes within demethylated CGI probes. In mock-treated cells, the percentage of methylated probes in each class (log₂-ratios higher than 1.0, 2217 probes) is shown. Demethylated probes were over-represented in promoters (p<2E-14) and under-represented in gene bodies (p<2E-20) in both AZA and DAC treated cells as determined by Fisher’s exact test and considering demethylated probes as being sampled from the control hyper-methylated set of probes.

https://doi.org/10.1371/journal.pone.0071099.g002

Affymetrix expression values used in Figure 3, and Figure S6 and Figure S7 in File S1were obtained using the Bioconductor [22] implementation of the Mas5 method. Plots of expression vs. methylation used the kde2d function of the MASS package [23] Genes that were up or down-regulated as a result of AZA or DAC treatment were identified by a profile similarity search (sorted by mean Euclidean distance to a specified profile) implemented in eXintegrator, combined with a minimal two-fold change in mean expression values between treatment and mock samples.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Correlation of gene expression and promoter CGI methylation.

A) MOCK-MOCK (B), AZA-AZA (C), DAC-DAC (D), AZA-MOCK (E), and DAC-MOCK. Two-dimensional kernel density estimates of expression level (log₂ transformed Mas5 estimates) and methylation levels (mean log₂-ratios for the complete island) for a set of low-density CGIs associated with transcriptional start regions (from a total of 198302 probes). Methylation is on the x-axis, expression level on the y-axis. Top row shows plots for methylation plotted against expression levels within individual samples, bottom row shows expression levels in two treated samples plotted against methylation levels obtained from a single control sample as indicated. All plots are from a single representative sample.

https://doi.org/10.1371/journal.pone.0071099.g003

Prolonged low-dose AZA and DAC treatments induce DNA demethylation

Our experiment was designed to detect specific epigenetic alterations in leukemic cells after low-dose exposure to demethylating agents. We employed a monocytic leukaemia cell line, SKM-1, derived from a leukaemia patient with a historical background of MDS without chromosome 5q deletion [24] for low-dose treatments. It has been well documented that high-dose AZA or DAC treatment of several human cell lines induces DNA demethylation and gene re-activation within a day [25,26]. However, it also greatly reduces cell proliferation, viability, and induces apoptosis and is unlikely to mimic the effect of clinical therapy. To find conditions that may give rise to effects similar to those induced by drug treatment in patients, we set out to determine a treatment plan using dosages compatible with normal proliferation of SKM-1 cells over an extended period of time. Graded concentrations (from 0.1nM to 1mM) of the agents were applied daily to SKM-1 cells in fresh medium. After continuous treatment for a week, cell proliferation ceased at the micro-molar scale but remained constant at lower dosage (Figure 1A). SKM-1 cells were approximately 10 times more sensitive to DAC than AZA but otherwise appeared to respond in a similar manner.

We assigned the optimum concentrations of AZA and DAC for daily treatment of SKM-1 cells as 100 and 10 nM respectively. Four weeks of daily treatment showed no effect on SKM-1 cellular morphology, but was sufficient to reduce the amount of DNMT1 and DNMT3B proteins significantly (Figure S2 in File S1). Both AZA and DAC induced DNA demethylation at LINE-1 repetitive sequences, though neither could reduce the basal methylation level of the GAPDH promoter (Figure S3 in File S1).

We also measured the expression of a number of hematopoietic markers using PCR arrays. Both AZA and DAC treatment altered the expression of approximately half of these markers (such as CD4 and CD34) (Figure S4A and Figure S4B in File S1) and their effects were highly similar (Figure S4C in File S1). In contrast to prolonged low-dosage treatments, a short exposure (3 days) of micro-molar AZA (10mM) and DAC (1mM) resulted mainly in gene down-regulation (Figure S4D and Figure S4E in File S1) and greatly suppressed cell proliferation (Figure 1A). These observations indicate that this (low-dosage) treatment plan has the potential to mimic the standard clinical treatment for MDS patients.

Demethylation pattern is regional and gene-specific

Changes in CGI methylation in tumours has been hypothesized to constitute a distinct phenotype (“CpG island methylator phenotype” or CIMP) [27,28] and demethylation of tumour suppressor CGIs was reported in MDS patients after treatments with AZA or DAC [12,25,29,30]. In order to analyse DNA methylation throughout the genome, we employed a method based on the ability of the McrBC enzyme to cut DNA at methylated CpG positions separated by 40 to 3000 bases. Digestion of fragmented DNA with this enzyme followed by comparative hybridisation to micro-arrays with non-digested DNA has been shown to be an effective means of detecting differentially methylated regions (DMR) [17]. Following McrBC digestion labelled probes were hybridized on microarrays containing probes tiled across all CGIs (See materials and methods). To correlate DNA methylation with microarray data, we used bisulfite sequencing to quantify DNA methylation at a selection of CGI probes lying in the HOXA cluster. A fifty percent level of methylation corresponded approximately to a log₂ ratio (log₂-ratios of signal from undigested to McrBC digested DNA) of 1 (Figure S5A in File S1).

To reduce the influence of noise from badly performing probes we first selected a set of 52915 probes (from a total of 198302) that showed low within replicate variance. These probes were then classified into methylation high (log₂-ratios > 1.5), low (log₂-ratios<0.5), and medium (log₂-ratios from 0.5 to 1.5) classes. Most (86.9%) CGIs fell into the methylation low category, and only a small fraction of regions were highly methylated (1.9%) (Figure 1B).

We mapped CpG islands according to their overlap or proximity to gene features (see material and methods). The percentage of highly methylated probes around promoters was significantly under-represented (p<6E-66) whereas probes associated with the gene bodies were over-represented (p<1E-20) (Figure S5B in File S1). Next, we aimed to identify probes demethylated as a result of treatment in all experiments (two sets of experiments using slightly different culture conditions, see materials and methods). Prolonged AZA and DAC treatment reduced the number of methylation high and medium probes by half (Figure 1B) (with DAC showing a slightly stronger effect for the medium group). Principal component analysis (PCA) indicated that AZA and DAC treatments had a global effect on CGI methylation with treated samples clustered away from those in mock-treated controls (Figure 1C).

Since DNMT inhibition results in a decrease in methylation across the genome it is possible that this may affect the accuracy of array-based estimates of methylation through implicit or explicit normalisation procedures. A comparison of log₂-ratios for treated and control samples indicated that a reduction in methylation occurs at the vast majority of methylated regions (Figure S6A and Figure S6B in File S1) and that AZA and DAC have very similar effects (Figure S6C in File S1). However, probes with low log₂-ratios in the control samples generally showed higher log₂-ratios in treated samples. To determine the cause of this, we performed bisulfite sequencing for all 12 samples for regions showing an increase, and ones showing a decrease in methylation after treatment (Figure S7 in File S1). This indicated that increases in log₂-ratios after treatment at regions hypomethylated in control samples do not represent increases in methylation, and are likely caused by inappropriate normalization. More pleasingly however, this analysis shows a strong linear relationship between percent methylation and log₂-ratios for regions with more than 10% of methylation. Importantly this relationship is identical across all samples thus validating our primary data.

From the selected 52915 probes, only 2217 CGI probes had log₂-ratios higher than 1.0 (suggesting more than 50% methylation) in control samples (Figure 2A). Of these, a total of 880 and 803 probes were demethylated by at least one of, or both, drugs respectively (Table S6 in File S1). Probes representing promoter CGIs were over-represented (p<2E-14) whereas probes associated with gene bodies were under-represented (p<2E-20) (Figure 2B and Table S5 in File S1) in the identified sets. In summary, our result shows that low-dose AZA and DAC treatment can effectively induce CGI demethylation at promoters, while methylation is maintained within gene bodies.

AZA and DAC treatments result in sustained gene activation

We next examined the correlation between expression and methylation levels. We performed transcriptome analyses for both mock and drug treated SKM-1 cells. The level of methylation in individual islands was summarised by the mean log₂-ratios, and these were plotted against expression levels. Since individual genes can overlap multiple CGIs we divided the CGIs into classes depending on their overlap with gene features as described above and made separate plots for each class (Figure S8 in File S1). In the control cells, a clear anti-correlation between gene expression and methylation was observed for CGIs overlapping promoter elements. This correlation was stronger for promoter CGIs with low CG content (Figure S9 in File S1), which may be due to the general paucity of highly methylated high CG density CGIs. The data also suggested that relationships between expression levels and DNA methylation exist at non-promoter CGIs (eg. gene body, upstream / downstream). However, these relationships were not as robust with observations depending on the summary statistics used (Figure S10 in File S1), and apparently restricted to subsets of islands within each class rather than generally true for the full set of islands.

Interestingly, this anti-correlation was lost or markedly reduced in AZA and DAC treated cells (Figure 3B, Figure 3C, and Figure S11 and Figure S12 in File S1). However, expression levels within AZA and DAC treated cells were still anti-correlated against promoter methylation levels in control cells (Figure 3D, Figure 3E). This strongly suggests that promoter CGI demethylation was not generally sufficient to modify expression patterns, and emphasizes the roles of other means of maintaining cell state.

Although a correlation between CGI demethylation and up-regulation of gene expression was not generally observed, we identified a small number of genes where expression appeared to change following demethylation (Table 1). It should be emphasised that more than half (110/190) of the genes whose expression was more than two times higher in AZA and DAC treated samples (Table S7 in File S1) were not associated with CGIs, and no array based methylation data was obtained for these genes. However, DNA demethylation was detected in the non-CGI promoters of the top three up-regulated genes (TYROBP, IL8, and TNF) by bisulfite sequencing (Figure S13 in File S1). These data indicate that prolonged low dose treatments are capable of demethylating CpG sites at non-CGI promoters and that this may have an effect on gene expression.

We selected three candidates showing a ten-fold increase in expression after treatment; HSPA2, TNF, and TYROBP, to further characterize the action of the drug treatment. DNA methylation and expression profiles were determined for AZA-treated cells collected at days 7, 14, 21, and 28. Partial demethylation was detected by pyrosequencing at week one, and the methylation levels decreased gradually (Figure 4) throughout the treatment. Gene expression levels showed an inverse correlation to the methylation pattern throughout the assay. More importantly, these modifications were maintained in the absence of drugs for ten days.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Time-course methylation and expression analysis for HSPA2, TNF, and TYROBP in AZA treated SKM-1 cells.

Top, CG methylation as determined by pyrosequencing. Y-axes indicate mean percentage of methylation at 3 (HSPA2), 5 (TNF), and 4 CpG sites (TYROBP). Bottom, qRT-PCR data showing relative gene expression normalized to the housekeeping gene GAPDH. No drug was applied to the cells from day 28 to day 38. Both expression and methylation data are means (± SEM) from four independent experiments obtained from mock- (green line) and AZA- (red dashed line) treated cells.

https://doi.org/10.1371/journal.pone.0071099.g004