2.1 Patient information

Samples of patients with chronic myelogenous leukemia (CML), acute promyelocytic leukemia (APL), gastrointestinal stromal tumor (GIST) and acute lymphoblastic leukemia (ALL) were obtained from established tissue banks at University of Southern California and M. D. Anderson Cancer Center. All samples were collected according to the institutional review board of the University of Southern California and M. D. Anderson Cancer Center in accordance with institutional guidelines. A brief description of the study samples, such as gender, age, and specimen type, is provided in Supplementary Data 1. Treatment with the DNA methyltransferase inhibitor, 5-azacytidine, has been described previously [29]. CD34⁺ hematopoietic stem cells isolated from umbilical cords and transfected with the BCR-ABL gene (Ph⁺ CD34⁺) were generously provided by Dr. Ravi Bhatia [6]. Briefly, the controls expressed green fluorescent protein (GFP) alone, and the cell fractions were collected according to low or high BCR-ABL expression. The cells were cultured for 7 or 14 days before harvest to examine DNA methylation.

2.2 Genomic DNA extraction and DNA methylation analysis

DNA was extracted using phenol-chloroform and stored at -80°C until required. DNA extracted from the samples was modified with sodium bisulfite using the 96 Zymo EZ DNA Methylation kit (Zymo Research, Orange, CA, USA). DNA methylation status was assessed using the GoldenGate Assay for Methylation Cancer Panel I from Illumina (La Jolla, CA, USA) and performed at the University of Southern California Epigenome Center, as described elsewhere [30]. The commercially available DNA Methylation Cancer Panel I interrogates 1,505 independent CpG sites. Data were reported as beta values, with the methylation status of the interrogated CpG site calculated as the ratio of the fluorescent signal from methylated allele relative to the sum of both methylated and unmethylated alleles. β values with detection p-values > 0.01 were removed from further analysis. The raw DNA methylation data will be available upon request. We used genomic DNA prepared following whole-genome amplification (WGA) as an ‘unmethylated’ control, and MSssI-treated DNA as a fully ‘methylated’ control. WGA kits (REPLI-g, Qiagen, Doncaster, Australia) were used to amplify DNA from normal blood as recommended, and these were bisulfite-treated as described above. The methylation reactions were carried out in 1x M.SssI buffer with 160µM S-adenosyl methionine (SAM) (supplied with M.SssI by New England Biolabs, Ipswich, MA, USA), and DNA was treated three times with M.SssI prior to bisulfite treatment

2.3 Bisulfite-PCR pyrosequencing measurements

Bisulfite-PCR pyrosequencing was performed for LINE-1 assay [26]. We used the PyroGold Reagent kit (Biotage, Uppsala, Sweden) on a Pyrosequencing 96HS as previously described [31]. In brief, 10 μL PCR products for each sequencing reaction were immobilized onto streptavidin-coated beads (Streptavidin Sepharose HP, GE Healthcare Biosciences, Pittsburgh, PA, USA) in binding buffer (10 mM Tris-HCl, pH 7.6, 2 M NaCl, 1 mM EDTA, 0.1% Tween 20) for 10 min. The biotinylated template was purified with the Pyrosequencing vacuum prep tool (Biotage, Uppsala, Sweden) and incubated with 10pmol/reaction individually with each sequencing primer in annealing buffer (20 mM Tris-acetate, pH 7.6 and 2 mM MgAc2). The DNA strands were denatured at 80°C for 2 min and reannealed at room temperature for 10 min. Sequencing was performed according to manufacturer’s instructions. The allele frequency (% cytosine or % thymidine) was calculated from the peak height, analyzed with the allele quantification module in the PSQ 96 HS software (Biotage, Uppsala, Sweden). Percentage of methylation was determined by the percentage of cytosine-to-thymidine conversion (methylation= % cytosine/(% cytosine + % thymidine)).

2.4 Data analysis

Data from the Illumina GoldenGate Methylation microarrays were analyzed using the open-source programs Cluster 3.0 and Java Treeview. Cluster 3.0 was used to perform an unsupervised hierarchical cluster analysis by applying complete linkage clustering with a Euclidean distance metric. The basic statistic and scatter plots were done using JMP, Version 7 (SAS Institute Inc., Cary, NC, USA). X-chromosome linked loci, loci that possibly contain single nucleotide polymorphisms (SNPs) and/ or repetitive elements were eliminated to increase the strength of analysis [31].

3.1 DNA methylation analysis of CML, APL and GIST

We examined DNA methylation using the Illumina GoldenGate Methylation Assay, which simultaneously analyzes methylation at 1,505 CpG sites in 807 genes, in 36 patients with CML, 18 patients with APL, five patients with GIST and six healthy controls. We analyzed 972 of these loci, as 533 loci are located on the X-chromosome or have previously been described to contain SNPs or repeat sequences that affect the DNA methylation analysis [31]. An unsupervised two-dimensional cluster analysis of these homogenous cancers demonstrated clear clustering by disease type, consistent with the theory that aberrant DNA methylation is not random but highly consistent within a specific cancer type, and that the DNA methylation profile is unique within a specific cancer (Figure 1). We found a high degree of correlation between DNA methylation profiles within each cancer (Pearson’s correlation CML= 0.927, APL= 0.915, and GIST= 0.892), and the correlation between cancers was lower (Pearson’s correlation= 0.852).

Figure 1

DNA methylation analysis of homogenous cancers. Two-dimensional unsupervised cluster analysis of DNA methylation in 36 CML, 18 APL and five GIST patients, and six samples of peripheral blood from healthy controls. DNA methylation at 972 CpG sites are shown, with probes corresponding to sites on the X-chromosome and those known to contain SNPs were sequestered and not used for downstream analysis. Red arrows indicate clusters of DNA methylation that are specific for each disease. Blue represents low levels of DNA methylation (beta value close to 0) and yellow represents high level of DNA methylation (beta value close to 1).

Unique DNA methylation profiles existed for CML, APL, and GIST by visual inspection (Figure 1, red arrows). The leukemia-specific methylated loci and the potential function of genes differentially methylated in leukemia were also analyzed by Ingenuity pathway analysis and listed in Supplementary Data 2. The DNA methylation profiles in these tumors enabled clustering by disease type, and the DNA methylation profiles were shared by individuals with the same disease. Therefore, DNA methylation profiles were similar between specific cancers, but different between CML, APL and GIST cancers.

3.2 The Philadelphia Chromosome and DNA methylation in leukemia

In order to understand the DNA methylation changes driven by t(9,22) in CML, we examined the DNA methylation profiles of CML and Ph⁺ ALL, a lymphoid malignancy which is also characterized by t(9:22). In addition, we examined CD34⁺ hematopoietic stem cells isolated from umbilical cords and transfected with the BCR-ABL gene (Ph⁺ CD34⁺)[6]. Two-dimensional unsupervised cluster analysis was performed using DNA methylation data from CML, ALL (Ph⁺ and Ph^-), and CD34⁺ (Ph^- and Ph⁺) (Figure 2) [6]. Here again, we found clear clustering by disease and cell type, but not by the presence or absence of BCR-ABL. Partial discrimination between CML and Ph⁺ ALL was possible despite the shared BCR-ABL genomic translocation, and CML did not cluster with Ph⁺ CD34⁺. Ph⁺ ALL and Ph^- ALL showed a distinct pattern of DNA methylation independently of the presence of the Philadelphia chromosome.

Figure 2

DNA methylation profiles of Ph⁺ leukemias. Two-dimensional unsupervised cluster analysis of DNA methylation profiles associated with the Philadelphia chromosome (CML, Ph⁺CD34⁺, and Ph⁺ ALL) and counterparts (Ph^- CD34⁺ and Ph^- ALL). Blue represents low DNA methylation (beta value close to 0), and yellow represents high DNA methylation levels (beta value close to 1). Different leukemia types are represented by different color bars on the top of heatmap.

Next, a consensus DNA methylation profile for each disease was determined by calculating the mean DNA methylation beta value at each locus. The hierarchical clustering tree of mean methylation of all samples also supported the notion that the myeloid malignancy, CML (three phases based on clinical characteristics; chronic phase, accelerated phase, and blast crisis phase), is more closely related to the other myeloid malignancy, APL than Ph⁺ CD34⁺ cells or Ph⁺ ALL (Figure 3). The difference in correlation was subtle, nonetheless consistent with the cluster analysis, and the highest correlation shown with CML was APL (r=0.9475), and not Ph⁺ CD34⁺ hematopoietic stem cells (r=0.9343) or Ph⁺ ALL (r=0.9416). Interestingly there was a highly consistent DNA methylation profile between Ph⁺ and Ph^- ALL, and between Ph⁺ CD34⁺ and Ph CD34⁺ cells, despite differences in Philadelphia chromosome status (r=0.9621, p=2.2x10^-16 and r=0.9633, p=1x10^-10 respectively) (Figure 4).

Figure 3

The hierarchical tree of leukemias with mean DNA methylation. Relationship of DNA methylation profile by clustering based analysis with different leukemias, GIST, Ph^+/- CD34⁺ hematopoietic stem cells, and healthy controls. A close relationship is demonstrated for the different phases of CML and myeloid leukemias (CML and APL), lymphoid leukemias (Ph⁺ ALL and Ph^- ALL) and hematopoietic stem cells (Ph^+/- CD34⁺). Relationships identified by clustering are independent of BCR-ABL status.

Figure 4

Direct comparison of DNA methylation profiles for different leukemias. DNA methylation profiles were compared for various leukemias (CML, APL, Ph⁺ ALL, Ph^- ALL, Ph^- CD34⁺, and Ph⁺ CD34⁺ cells). Scatter plots were generated for direct comparisons of mean methylation beta values. (A) CML with APL (r=0.9475), (B) CML with Ph⁺ ALL (r=0.9416), (C) CML with Ph⁺ CD34⁺ cells (r=0.9343) (D) Ph⁺ALL with Ph^- ALL (r=0.9621), and (E) Ph⁺CD34⁺ cells with Ph^- CD34⁺ cells (r=0.9633), and (F) CML with Ph^- ALL (r=0.9200).

3.3 DNA methylation changes with CML progression

The clinical course of untreated CML typically follows an ordered progression for almost all patients from an early chronic phase (CP) through an accelerated phase (AP) to terminal blast crisis phase (BP). The consistency of this progression among patients likely reflects the homogeneity of the disease, and also enabled us to examine DNA methylation changes involved in the progression of CML. Thus, we determined the DNA methylation changes found in different phases of CML. Subset analysis of our CML samples by chronic phase (n=23), accelerated phase (n=5), and blast phase (n=8) revealed 84 loci that clearly showed increased DNA methylation (mean β>0.01 and p value < 0.05) and 66 loci that showed a decrease in DNA methylation (β <0.01 and p value < 0.05) with CML progression (Figure 5A). The list of genes and Ingenuity pathway analysis are listed in Supplementary Data 3. Interestingly, we had a single patient for whom we had a series of blood samples taken at the time of diagnosis (chronic phase) and from the accelerated and blast phases. As a further validation of these identified loci, we plotted DNA methylation levels at 20 hyper-and hypomethylated loci in an independent set of serial samples from this single patient at the chronic, accelerated and blast phases (Figure 5B). DNA methylation changes of hypermethylation or hypomethylation with CML progression from 36 individual CML patients were well validated in the serial samples from a single patient.

Figure 5

DNA methylation changes associated with CML progression. (A) DNA methylation levels at 84 loci showed a significant increase while 66 loci showed a significant decrease during progression of CML from chronic phase (bars) to accelerated phase (blue dots), and blast crisis phase (red dots). (B) Examination of 20 hypermethylated (left) and 20 hypomethylated (right) CML progression related loci with serial samples from a single patient.

3.4 The effect of azacitidine (5-azacytidine, Vidaza^®) treatment upon DNA methylation changes in CML patient

We next wanted to determine the plasticity of the disease-specific DNA methylation changes in CML and changes involved in CML progression. We studied serial DNA methylation changes in a CML patient who became resistant to imatinib mesylate and developed blast crisis. The patient was then treated in a clinical trial with the DNA methyltransferase inhibitor, azacitidine (5-azacytidine, Vidaza^®) for three courses and responded with hematological improvement of white blood cell count (Clinical trial reported previously by Bernstein et al. [29] Patient #2, Table 1). Eventually, a more potent tyrosine kinase inhibitor, dasatinib (Sprycel^®), became available and the patient’s therapy was switched. Serial peripheral blood samples were collected and the genome-wide changes in DNA methylation profile were studied by Illumina GoldenGate Methylation Assay and by bisulfite PCR-pyrosequencing to assess the LINE-1 DNA repetitive element in samples from the chronic phase, blast phase, azacitidine treatment and dasatinib treatment.

The patient was a 33-year-old female who was originally diagnosed with CML in 1999. The patient was treated with hydrea and then interferon, and finally imatinib mesylate in 2001, when it became available. The patient’s best response to therapy was only a partial cytogenetic response with 31% of cells being Ph⁺ by fluorescence in situ hybridization (FISH) after 18 months of therapy (March of 2003). The patient presented with a markedly elevated white blood cell (WBC) count of 306,000 cells/ mm³ and was found to be in myeloid blast crisis. The patient was started on the DNA methyltransferase inhibitor, azacitidine as part of a clinical trial at a dose of 25 mg/m² intravenously for five consecutive days in Course 1 (28 day courses), 75 mg/m² intravenously for Course 2, and 75 mg/m² subcutaneously for Course 3 [29]. There was a dramatic decrease in the patient’s WBC count to normal by Course 2 of therapy (342,800 cells/mm³ to 5,200 cells/ mm³) (Figure 6A). This corresponded with a subtle but significant decrease in global DNA methylation, as assessed by measuring DNA methylation of the LINE-1 repetitive element by pyrosequencing (Pearson’s correlation 0.67, p value=0.049), and a decrease in the average DNA methylation of 1,006 loci as measured by Illumina GoldenGate Methylation Assay (including the original 972 loci analyzed in this study and those loci located on the X-chromosome loci). The maximum decrease in LINE-1 DNA methylation was 2.5% after treatment with azacitidine. It took three courses of treatment to reach this point, and this was consistent with the average gene-specific change at the 1,006 loci, which showed a mean beta value decrease of 0.03 after three courses of treatment (Figure 6B). According to visual inspection, changes in DNA methylation (the average of promoter gene methylation and global DNA methylation) occurred prior to changes in WBC levels (azacitidine Course 1 Day1, Course 2 Day 3, and off trial). The mean DNA methylation value of the CML progression loci (hypo- or hyper-) also responded to the DNA methyltransferase inhibitor drug (Figure 6C). The loci that were shown to be increasingly methylated during CML progression became less methylated during azacitidine treatment (mean beta value 0.61 to 0.48), and those loci at which methylation decreases during CML progression became increasingly methylated with azacitidine treatment (mean beta value 0.14 to 0.42) (Figure 6C).

Figure 6

Clinical response and DNA methylation changes induced by azacitidine (Vidaza^®) treatment in a patient with CML. A) WBC count, B) DNA methylation of the LINE-1 repetitive element, and the mean DNA methylation of 1,007 CpG sites for Illumina GoldenGate analysis, and C) CML progression hyper-and hypomethylated loci following a course of azacitidine treatment.