Concurrent DNA Hypermethylation of Multiple Genes in Acute Myeloid Leukemia

Methylation patterns are established in a tissue-specific manner early in mammalian development, mediated by a combination of demethylation and de novo methylation of cytosine residues primarily at CpG dinucleotides (reviewed in Ref. 1). The methylation pattern is maintained through subsequent cell divisions by the action of a DNA MTase³ enzyme (2). DNA methylation patterns are often altered in cancer cells, with associated increases in the level of the DNA MTase enzyme (3, 4, 5, 6), widespread genomic hypomethylation, and simultaneous regional increases in DNA methylation patterns (7, 8) having been reported. Aberrant hypermethylation in cancer cells often occurs in the CpG-rich promoter regions (CpG islands) of many tumor suppressor genes and is associated with gene inactivation (reviewed in Ref. 9). Approximately half of the mammalian genes have CpG-rich promoter regions, and although most CpG dinucleotides are methylated in the normal cell, the CpG dinucleotides of CpG islands are essentially unmethylated in all tissues (10). What protects CpG islands from methylation, or moreover, what triggers hypermethylation in the cancer cell is unknown.

To understand further the process of aberrant hypermethylation in cancer, we have chosen to study DNA methylation patterns in leukemia. We have demonstrated previously that the levels of DNA MTase in leukemia are elevated 4–5-fold (6). In addition, DNA methylation patterns are known to be altered in leukemia; a generalized hypomethylation has been reported for B-cell chronic lymphocytic leukemia (11), as well as specific regions of hypomethylation such as in the Bcl-2 oncogene (12), and tumor necrosis factor α and β genes (13). In contrast, hypermethylation of the calcitonin gene has been observed in acute and chronic leukemias and lymphomas (14, 15) and in patients with myelodysplastic syndrome (16, 17). Hypermethylation is also seen in the cyclin-dependent kinase inhibitor gene p15 in patients with acute leukemias (18, 19). The ER gene is hypermethylated in acute and chronic leukaemias, as well as in lymphomas (20). The HIC1 gene also has been reported to be frequently methylated in acute lymphoblastic leukemia but infrequently methylated in AMLs. Therefore, there is mounting evidence that a number of individual genes can be abnormally methylated in leukemia. However, most of the previous studies were performed in separate patient groups by restriction enzyme analysis, and consequently, the methylation data are limited to a few CpG sites in individual genes. Hence, the extent of hypermethylation measured previously in leukemia may not have been representative.

The mechanism responsible for abnormal methylation in cancer is unclear. In this study, we address whether aberrant methylation is limited to single target genes that provide the cell with a growth advantage, similar to a random mutation, or whether multiple genes are susceptible to hypermethylation in the same cell, thereby implying a general deregulation of CpG island methylation in cancer. We have used sodium bisulfite genomic sequencing (21, 22, 23) to re-examine in more detail the patterns of aberrant de novo methylation in leukemia from the same patient cohort. In comparison to Southern analysis or MSP (24), bisulfite sequencing permits analysis of every cytosine in the CpG-rich promoter regions of each target gene. In particular, we address whether critical CpG sites are commonly hypermethylated and whether CpG island methylation is confined to a single locus or multiple loci in the one cohort of patients.

Bone marrow samples were aspirated from 20 patients presenting with AML. The patients (11 males and 9 females) were between 22 and 88 years of age, with a median age of 53 years. The proportion of marrow blasts was >70% in all cases. The bone marrow used as a normal control was aspirated from the sternal cavity from nine patients who were undergoing cardiac surgery and who had given prior informed consent. The patients (seven males and two females) were between 19 and 74 years of age, with a median age of 61 years, and no evidence of leukemia. The study was approved by the Ethics Committee of Royal Prince Alfred Hospital (protocol number X93-0073).

DNA was isolated using TriZOL Reagent (Life Technologies, Inc.) from bone marrow cells that had been lysed in hypotonic lysis buffer. Bisulfite genomic sequencing was used to analyze the methylation patterns. The bisulfite reaction was carried out for 16 h at 55°C on 1–2 μg of HindIII-digested patient DNA, under conditions described by Clark et al. (22). After bisulfite conversion, the DNA was ethanol precipitated, dried, resuspended in 100 μl of TE buffer [10 mm Tris-HCl (pH 8) and 1 mm EDTA] and stored at −20°C. The primers used for amplification of sodium bisulfite-converted DNA are summarized in Table 1. Where direct sequencing was to be performed, the (−21)M13 universal primer sequence was incorporated into the 5′ end of inner primer 2. Nested PCR amplifications were performed on 1–3 μl of bisulfite-treated genomic DNA in a reaction mix containing 200 μm of each of the four deoxynucleotide triphosphates and 2 units of AmpliTaq DNA polymerase (Perkin-Elmer). The reactions were performed in either 25 μl of reaction mixtures containing 50 mm KCl, 10 mm Tris-HCl (pH 8.3), and 100 ng of each primer or in 50-μl reactions containing 67 mm Tris, 16.6 mm ammonium sulfate, 1.7 mg/ml BSA, and 10 mm β-mercaptoethanol in TE buffer and 300 ng of each primer. The volume of the PCR for each gene is as noted in Table 1. Reactions were cycled in a Hybaid DNA Thermal Cycler with variable MgCl₂ concentrations as detailed in Table 1. The standard PCR reaction was performed by the following cycling conditions: 96°C/3 min for 1 cycle; 95°C/1 min, a₁°C/2 min, 72°C/3 min, for 5 cycles; 95°C/1 min, a₂°C/2 min, 72°C/2 min, for 23 cycles; and 72°C/4 min for 1 cycle, where a₁ and a₂ are annealing temperatures 1 and 2, respectively, described in Table 1. All of the primers used were shown to amplify methylated and unmethylated DNA without major bias under these PCR conditions (25).

For automated direct sequencing, the PCR products were amplified using the internal forward primer that included the (−21)M13 universal primer sequence. Direct PCR sequencing reactions were performed using a PRISM Dye Primer Cycle sequencing kit (−21 M13 Fwd) with AmpliTaq FS (Perkin-Elmer) on an automated 373A DNA Sequencer (ABI). Sequencing reactions were performed as recommended by the manufacturer. The amount of methylcytosine of each CpG dinucleotide was quantitated by comparing the peak height of the cytosine signal with the peak height of the cytosine plus thymine signal. When using direct sequencing, it is often difficult to assess methylation levels <15% due to the variable sequencing background signal.

We have used bisulfite genomic sequencing to characterize the overall methylation profile of the CpG islands associated with calcitonin, ER, E-cadherin, p15, p16, Rb, GST-pi, and HIC1 in the bone marrow of patients with AML. We have shown previously that the expression of DNA MTase was increased in a number of the AML patients analyzed in this study (6). Twenty patients with de novo AML, spanning ages 22–88 years, and nine normal controls, spanning ages 19–74 years, were assayed. The bone marrow from the AML patients contained >70% blast cells. Therefore, due to the low amount of normal cell contamination, the methylation levels that are presented represent minimal values. The genes that we sequenced were selected on the basis of being hypermethylated either in leukemia or other cancers or on the basis on our preliminary Southern analysis data that indicated the genes were hypermethylated in the AML samples (data not shown). We chose to analyze the methylation data by direct PCR sequencing because this provides a semiquantitative estimate of the methylation levels in the sample, and individual CpG sites could be assessed in the one reaction. A summary of the genomic sequencing results for each of the eight individual genes analyzed is presented below, followed by an overall summary of the methylation state of the genes for each patient.

It has been well established that the calcitonin gene, which lies on chromosome 11p15.2, is a “hot spot” for methylation in leukemia. The CpG rich 5′ region of the calcitonin gene was found previously to be methylated in 78–95% of acute leukemias by Southern analysis (16, 26) or genomic sequencing (27). To correlate the methylation profile of the calcitonin gene in AML with the methylation profile of the other target genes, we used bisulfite genomic sequencing. Fig. 1 shows the sequence map of the 19 CpG sites in the central CpG rich island close to the transcription start site that were sequenced. There was a low level (<25%) of methylation at one to two CpG sites in approximately half the normal bone marrow samples, and these methylated sites were localized downstream to the start of transcription. In contrast, 12 of 17 (71%) of the AML patients had substantial methylation across the entire CpG-rich region. However, the methylation patterns in the AML samples was heterogeneous, varying from extensive methylation at high levels (e.g., R76 and R86) to low level methylation at four to six CpG sites (e.g., R94 and R125). The remaining AML patients (5 of 17, 29%) displayed methylation profiles similar to the normal patterns (e.g., R63 and R79). There does not appear to be any critical sites that are exclusively methylated in each patient; however, the region encompassing CpG sites 1–6, which are the sites closest to the transcription start site, were found to be the most heavily methylated (50–100%). Of the 19 CpG sites studied, only one site (CpG 15) encompassed an informative site for Southern analysis (HpaII site). However, this site was only methylated in 5 of 17 (29%) of the samples, 1 of which (R63) had a normal methylation profile. Therefore, using bisulfite sequencing enabled a more detailed and informative description of the methylation profile of the calcitonin gene.

The ER gene, located on chromosome 6q25.1, has been shown by Southern blot analysis to be methylated in 50–90% of acute and chronic leukemias (20). This study assayed the methylation state of a single NotI site within the first exon of the gene. To determine the extent of hypermethylation at other CpG sites, we have used bisulfite genomic sequencing to measure the methylation status of 22 CpG sites in the CpG-rich promoter region (Fig. 2). We found no evidence of methylation in this region of the ER in any of the normal bone marrow samples analyzed. In comparison, we found a low level of methylation spanning the 22 CpG sites in the bone marrow cells of 6 of 11 (54%) patients with AML. The methylation profile was heterogeneous, but CpG sites from 15 to 20 were most commonly methylated. The methylation state of CpG sites 14 and 15, which incorporate the NotI site used for Southern analysis, therefore, is indicative of the methylation profile across the region. Interestingly, in comparison to the calcitonin methylation profile, the methylation level for each CpG in the ER gene was <50%, possibly indicating hypermethylation of only one allele.

The E-cadherin gene, located on chromosome 16q22.1, has not previously had its methylation status determined in leukemia. However, hypermethylation has been reported in carcinoma cell lines, including prostate and breast, by Southern analysis (28, 29) and MSP (30). Moreover, hypermethylation of this region was shown to correlate with lack of expression (28, 30). In this study, we sequenced 29 CpG sites that are located in the CpG-rich promoter and part of exon 1 (Fig. 3). All of the normal bone marrow samples, except for N34, displayed no methylation in this region. N34 had 5 of 29 CpG sites methylated at a low level (<25%). In contrast, 9 of 13 (69%) bone marrow samples from patients with AML displayed extensive methylation of the E-cadherin promoter region (Fig. 3). Again, the methylation patterns were found to be heterogeneous, ranging from no detectable methylation (R74 and R75) to patients methylated in each determinable CpG site (R76 and R99). Although the methylation patterns are heterogeneous, the degree of methylation is higher in the 3′ region (50–100%).

p15 and p16 are cyclin-dependent kinase inhibitors proximate to each other on chromosome 9p21. Southern blot analysis and MSP have found p15 to be inactivated through hypermethylation of its associated CpG island in 50–80% of patients with acute leukaemia (18). Interestingly, this aberrant hypermethylation was found to be limited to p15 and was found not to extend to p16.

To enable a more complete analysis of these genes, we determined the methylation status of 47 CpG sites that encompass the transcription start site and exon 1 of p15 (Fig. 4). This sequence also includes the region where hypermethylation has been associated with transcriptional silencing (31). We found in some of the normal bone marrow samples a low level of methylation (<25%) at one to five CpG sites downstream to the start of transcription. In comparison, we found hypermethylation of the p15 gene in 13 of 19 (68%) of AML patients studied. Again, the methylation patterns are heterogeneous, ranging from no methylation (R79 and R94) to sparse methylation (R59 and R60) and to methylation encompassing the entire region (R38 and R102). In general, methylation spanned the entire region analyzed but was highly mosaic in the individual patients. A similar finding of variegation in the methylation profile of p15 was also found in AML by Dodge et al. (19). As for calcitonin, the level of methylation was often high (75–100%) at individual CpG sites, indicating that either both alleles were methylated or that one allele was heavily methylated and the other lost.

For the p16 gene, we analyzed 28 CpG sites in a region spanning the promoter and exon 1 (Fig. 5). Methylation in this region has been correlated to transcriptional silencing (32, 33, 34). We found a low level (<25%) of methylation at one to five CpG sites in three normal samples. In comparison, we found 6 of 20 (30%) patients with AML had an equally low level of methylation (<25-<50%) at individual CpG sites, but methylation was found to be more extensive, spanning 7–14 CpG sites across the region analyzed. Methylation was heterogeneous, and there were no particular CpG sites that were more commonly methylated. Methylation of the p16 gene has not been reported before in leukemia. However, it should be noted that Southern analysis would not have detected the low number of individual sites that are methylated, and MSP analysis may not have detected methylation because the methylation profile was quite heterogeneous. It is not clear whether this small amount of p16 promoter methylation is linked to transcriptional silencing, but the methylation pattern is clearly different in the leukemic cells versus the normal bone marrow cells.

The HIC1 gene located on chromosome 17p13.3 is unusual in that the entire gene is contained in a CpG-rich region. Previous methylation studies have shown that hypermethylation of the five NotI sites spanning the gene are linked with HIC1 gene silencing in neoplastic cells (35). A number of leukemia subtypes have also been shown to display HIC1 hypermethylation, including 10% of patients with AML at presentation (36). To determine whether the HIC1 gene was in fact more extensively methylated in AML, which is not assessable by NotI digestion, we measured the methylation pattern of 45 CpG sites by bisulfite genomic sequencing in a region within the central CpG-rich region of the gene spanning intron 2 and exon 3 (Fig. 6). The region sequenced was chosen because it contained no NotI sites and was central to the HIC1 CpG island. We found significant methylation of the HIC1 gene in all of the normal and AML DNA samples sequenced. Heterogeneity in the methylation patterns was found in all samples; however, in contrast to the AML methylation profile, we found that methylation in the normal cells was limited to exon 3 and was not detected in intron 2. Interestingly, hypermethylation was observed in 10 of 12 (83%) AML samples in both intron 2 and exon 3. Therefore, methylation of intron 2 appears to be specific to the leukemic cells. As with the methylation profile of the other genes analyzed, we found that the methylation patterns in the intron 2 sequence varied significantly among AML patients, with some patients having no detectable methylation (R99) and some patients being methylated in each determinable methylation site (R36).

The Rb gene is located on chromosome 13q14 and is commonly hypermethylated in retinoblastoma tumors (37, 38), and the GST-Pi gene located on chromosome 11q13 is commonly methylated in prostate and breast tumors (39, 40, 41). The methylation pattern for both of these genes has not been studied in detail before in leukemia. To determine whether either Rb or GST-Pi is methylated in our set of AML samples, we assayed for methylation of the promoter region using bisulfite analysis. The 5′ region of the Rb gene containing 27 CpG sites (38) was amplified after sodium bisulfite conversion and sequenced by direct PCR sequencing. No methylation was detected at any of the 27 CpG sites for the normal or the bone marrow samples from the 13 AML patients tested (data not shown). The 5′ region of the GST-Pi gene, containing 38 CpG sites (39), was amplified following sodium bisulfite conversion and methylation tested by BstU I digestion. There was no methylation evident at any of the 6 BstU I sites within the promoter sequence in the normal or in the bone marrow samples from 12 AML patients tested (data not shown). Lack of methylation of the 5′ CpG region of GST-Pi was also confirmed by methylation-sensitive PCR (data not shown). Therefore, unlike the other target genes we studied that we found were frequently methylated in AML, the CpG islands spanning Rb and GST-Pi appear to be protected from hypermethylation in AML, although they are predominantly methylated in other specific cancer types.

By studying the methylation profile of a number of target genes from the same cohort of patients, we have been able to assess whether hypermethylation is limited to single target genes or whether multiple genes are susceptible to hypermethylation in the one cell. Fig. 7,A summarizes the methylation patterns for the eight target genes studied in the normal and AML bone marrow samples. To aid in the interpretation of the results, we scored the gene as significantly methylated only if the samples were methylated at any one CpG site at levels >25% or if the samples were methylated in at least 25% of determinable sites. It is clear that except for methylation in the calcitonin gene that was found in one normal sample (N47) and HIC1 which showed differential methylation in exon 2, the rest of the genes studied from normal bone marrow samples were essentially unmethylated in the CpG-rich island regions analyzed. In contrast, the methylation profile in the AML patients was extensive. Moreover, hypermethylation was not limited to a single target gene but often was found in multiple genes in each patient studied. In fact, 15 of 20 (75%) AML patients displayed concordant methylation in at least two genes. Interestingly, the subset of genes methylated varied in each patient. For example, in R128, only p15 was methylated; in R103, p15 and ER were both methylated; and in R14, p15 and calcitonin were both methylated. There are also patients with multiple genes methylated; for example in R36, calcitonin, E-cadherin, p15, p16, and HIC1 were methylated, and in R38, all of the genes tested were methylated except for GST-Pi and Rb. In fact, GST-Pi and Rb were not methylated in any of the AML samples tested. Fig. 7 B shows that >50% of the patients were methylated in the calcitonin, E-cadherin, ER, p15, and HIC1 genes, and >80% of the patients were methylated in HIC1. However, it is of interest to note that no single gene was always methylated.

To determine whether there is a correlation between methylation profiles and DNA methyltransferase levels, we compared the methylation patterns from eight patients to the expression of DNA MTase as measured by competitive reverse transcription-PCR (6). We previously have shown that DNA MTase is increased 2.5–10-fold in AML, but there appears to be no apparent correlation between DNA MTase expression and the number of target genes methylated (Fig. 7 A). For example, in patient, R99 DNA MTase levels were found to be 10.1-fold higher than the average normal DNA MTase levels, and three genes were found to be hypermethylated, whereas six genes were hypermethylated in patient R38 that had a 5.6-fold increase in DNA MTase levels. In addition, there appears to be no apparent correlation between the number of genes methylated and the age or sex of the patient. Our results obtained in leukemia cells suggest that aberrant methylation in malignancy is due to a general deregulation of the methylation machinery in the cell rather than a specific targeting of single genes. However, the methylation pattern of all CpG island-associated genes are not affected, because there appears to be an individual variability in the range of genes hypermethylated in the same cell type.

Although alterations in DNA methylation patterns are commonly found in essentially all cancers (9), little is known about the underlying mechanism. In this study, we asked whether multiple genes are simultaneously methylated in the same cancer cell and whether the methylation pattern is cancer cell type specific. We analyzed the methylation pattern across the CpG-rich promoter regions of eight tumor-related genes in the same cohort of AML patients using bisulfite analysis. Methylation studies to date generally have been limited to the study of the few CpG dinucleotides that lie in the restriction enzyme recognition site and therefore provide little information about the nature of de novo methylation. Furthermore, the methylation studies for different genes often were performed on different cohorts of patients, making it difficult to assess concurrent methylation within the one sample. By using bisulfite sequencing, we could address whether aberrant de novo methylation was targeted to specific CpG sites or was extensive across the CpG island promoter regions associated with tumor-related genes.

The main conclusion from our study is that aberrant methylation is not confined to single target genes in AML but rather can occur concurrently across many different genes spanning different chromosomes in an individual patient. The HIC1 gene (spanning intron 2) was the most commonly methylated, with >80% of the AML samples tested showing hypermethylation. Calcitonin, E-cadherin, ER, and p15 were also commonly methylated, with >50% of the AML samples showing hypermethylation. In comparison, p16 was less frequently methylated, and Rb and GST-Pi were found to be never methylated in the AML samples tested. Interestingly, the methylation portfolio of each AML patient was remarkably different from each other, with quite a variation in the combination of genes selected. In addition, there was no obvious correlation between ages of the patients with AML and the number of genes or subset of genes methylated. In particular, the methylation of the ER did not appear to be age specific in either the normal or cancer samples, as reported by Issa et al. (42) and Ahuja et al. (43) for colon cancer. This difference may be due to the fact that different tissues were examined in the latter study, or our sample size was limited. However, we have taken care to ensure that the median ages between normal and AML samples were matched to differentiate AML-specific methylation from age-related methylation.

It is clear from our results that, although there is a heterogeneity in the profile of genes methylated within the AML patients, there also appears to be cancer cell type-specific differences. For example, Rb and GST-Pi were both found to be unmethylated in the AML patients, whereas these genes are frequently methylated in retinoblastoma tumors and prostate tumors, respectively (38, 39, 44). p16 was also methylated at a low level in ∼30% of samples in this study, but p16 is frequently methylated in lung cancer and melanoma (45). Liang et al. (46), using a genome scanning analysis of three different tumors, also reported tissue-specific variation in the methylation patterns between the tumors. The fact that multiple genes are frequently methylated in cancer suggests that the mechanism that normally protects CpG islands from methylation is defective in the cancer cell. The fact that different combinations of a subset of genes can be methylated in the one cell type suggests that methylation is stochastic. The set of genes that are commonly methylated possibly reflects the genes that, when silenced, provide the cell with a selective growth advantage in different cell types. However, the cancer specificity observed may reflect the different set of tissue-specific factors found in the different cell types.

The mechanism that protects CpG islands from methylation is unclear, and what triggers hypermethylation in cancer is still unresolved. An increase in DNA MTase was proposed to be important in the initiation and progression of cancer and in establishing altered DNA methylation patterns (3, 4, 5, 9). However, it is clear from our results that simply an increase in DNA MTase is insufficient to explain the altered methylation patterns. The AML samples screened had DNA MTase levels 2.5–10-fold higher than normal bone marrow (6), but the methylation patterns of these patients were variable and bore no obvious relationship to the level of DNA MTase in the cell. It is, however, possible that other methylation enzymes, such as dnmt3a and dnmt3b (47), may play a role in abnormal methylation in leukemia.

Active transcription also has been proposed as a mechanism to protect CpG islands from methylation (48). It is possible that in the cancer cell, a subset of CpG island promoters are silenced, permitting subsequent de novo methylation. This is an attractive hypothesis and may help to explain how hypermethylation appears to be stochastic but also limited to a specific subset of genes for the different cancer types. Unfortunately, because the samples we tested contained a low level of normal cell contamination, it was not possible to accurately quantitate the level of expression for each gene in the sample and demonstrate that the genes were indeed silenced. It is also unclear from our results whether the variable methylation patterns we observed influenced the level of gene transcription within the individual genes tested. However methylation, as determined by Southern analysis, has been associated with inactivation of expression in many of the genes we analyzed (28, 30, 32, 33, 34, 35, 42, 49). We suggested previously that transient CpNpG methylation of Sp1 sites in the promoter regions of CpG islands may be one mechanism that may lead to gene silencing and subsequent de novo methylation of CpG islands (50). However, in this analysis, we found no evidence of CpNpG methylation at detectable levels. Because our data were obtained by direct PCR sequencing, methylation levels <15% are difficult to assess above background.

We chose to analyze the individual methylation patterns for each gene by sequencing to gain an insight into the mechanism responsible for the specific methylation profiles. We found that for each gene, the methylation profile across the CpG island was heterogeneous, and no one CpG site appeared to be always methylated. This is similar to our previous findings of mosaic methylation found in both cancer and normal cells (38, 39, 51, 52). It therefore appears that the de novo methylation process is stochastic and supports the hypothesis that it is the density of methylation across a region that is important rather than methylation of individual target sites. Interestingly, for the HIC1 gene, we did note a marked boundary of methylation that occurred between intron 2 and exon 3 in the normal cell that was lacking in the cancer cell. The existence of such a methylation boundary, which is clearly lost in the cancer cell, may provide clues in the future as to the mechanism that normally protects sequences from methylation in the normal cell (53).

The results of this study have allowed us to answer some important questions about de novo DNA methylation in cancer cells: (a) although a deregulation of the methylation machinery is implicated in abnormal hypermethylation, an increase in DNA MTase alone is not sufficient to mediate this hypermethylation; (b) hypermethylation in AML is not confined to a single locus but rather is a multiple loci event covering CpG-rich gene regions; and (c) the aberrant de novo methylation found in AML appears to be a stochastic process acting at individual CpG sites within different CpG islands combined with a cell type selectivity. Moreover, the cancer specificity may reflect the different set of factors found in different cell types. The next step is to determine what the factors are protecting CpG islands from methylation in the normal cell and what triggers the breakdown of the protective process in the malignant cell.

We thank Cheryl Paul for the excellent automated sequencing and Dr. Doug Millar for advice and helpful discussions.