Extensive Intra- and Interindividual Heterogeneity of ^p15INK4B Methylation in Acute Myeloid Leukemia¹

Cell cycle regulation and the genetics of cancer have become closely connected in recent years. A number of studies have focused on p15^INK4B and p16^INK4A, two neighboring genes at chromosome 9p21, the products of which specifically inhibit cyclin D/CDK4³ and cyclin D/CDK6 complexes and, thereby, regulate the cell cycle negatively (1). Evidence suggesting that p16^INK4A is a tumor suppressor gene involved in the development of several human cancers includes p16^INK4A deletions and point mutations in tumors and tumor cell lines (2, 3), and germ-line p16^INK4A mutations in patients with familial melanoma (4). In contrast, point mutations in p15^INK4B have not been identified in either sporadic tumors (5) or cancer-prone families (6), and deletion of p15^INK4B is almost consistently associated with concomitant deletion of p16^INK4A. Recently, however, the candidacy of p15^INK4B as a tumor suppressor gene was strongly supported by the demonstration that it is selectively inactivated by hypermethylation in some gliomas and in both acute lymphoblastic leukemia and AML (7).

AML patients have traditionally been divided into subclasses on the basis of the morphology of their leukemic blasts in accordance with the FAB classification (8). Increasing attention has, however, been paid to the series of genetic aberrations in myeloid progenitor cells, thought to initiate clonal development in AML. Several types of genetic changes may contribute to the leukemic phenotype, including balanced translocations and alterations of proto-oncogenes and tumor suppressor genes. Lately, transcriptional silencing of tumor suppressor genes by hypermethylation of promoter CpG islands has emerged as another significant event in AML. In addition to p15^INK4B, common targets for hypermethylation in AML include the estrogen receptor gene (9) and HIC-1 (10, 11).

DNA methylation usually occurs at cytosine residues contained in symmetrical CpG dinucleotides. In normal tissues, isolated CpG dinucleotides in bulk chromatin are often methylated, whereas cytosines in CpG islands are unmethylated. In neoplasia, this pattern of methylation is commonly reversed. The mechanisms underlying the generation and maintenance of de novo methylation have, however, remained a subject of considerable debate. Recent studies of fragile-X syndrome (12) and retinoblastoma (13) have suggested that methylation of CpG islands may vary among individuals and be subject to dynamic changes.

To study in detail the patterns of cytosine methylation at the CpG island of the p15^INK4B gene in AML, we devised a novel method that is based on (a) treatment of genomic DNA with sodium bisulfite, (b) amplification of both methylated and unmethylated sequences by PCR, and (c) resolution of differentially methylated sequences by DGGE (14). Treatment of DNA with sodium bisulfite converts unmethylated cytosines to uracils, whereas methylated cytosines remain unreactive (15, 16). Following this treatment, methylated and unmethylated sequences are predicted to differ in thermal stability due to their different GC contents and can be physically separated by electrophoresis in a gel containing an increasing gradient of chemical denaturants (17). The produced gel provides a detailed visual display of the methylation status in complex cell populations and a simple means for isolating clonotypic epigenotypes. Here, we demonstrate the use of bisulfite-DGGE to reveal a highly complex and variable pattern of p15^INK4B methylation in AML.

For this study, MNCs from bone marrow and peripheral blood from 65 AML patients diagnosed between 1989 and 1992 were examined. MNCs had been cryopreserved in 10% DMSO and 10% FCS. In all samples, the percentage of leukemic blasts exceeded 95%. Blood and bone marrow specimens were used interchangeably because pilot experiments showed no difference between cell sources for individual patients. The patients were categorized according to the FAB classification as follows: 3 M0, 18 M1, 16 M2, 18 M4, 8 M5, and 2 M6. As normal controls, peripheral blood MNCs from 10 healthy volunteers were assayed.

MNCs isolated from peripheral blood from three normal individuals were separated into monocytic and lymphocytic fractions using magnetic cell sorting (MACS; Miltenyi Biotec GmbH, Bergisch Gladbach, Germany), according to the manufacturer’s guidelines. In brief, following incubation with an anti-CD14 monoclonal antibody (clone TÜK4; Miltenyi Biotec GmbH), cells were separated on MiniMACS columns (Miltenyi Biotec GmbH). The monocyte fractions contained >95% CD14-positive cells and <0.5% T lymphocytes. The lymphocyte fractions contained <1% monocytes.

Genomic DNA was isolated from MNCs, using the Puregene DNA Isolation Kit (Gentra Systems Inc., Minneapolis, MN), according to the manufacturer’s instructions. Four μg of genomic DNA were treated with sodium bisulfite as described previously (18). The MOLT-4 T-lymphoblastic leukemia and the HL-60 promyelocytic leukemia continuous cell lines were included in each bisulfite treatment as positive and negative controls, respectively.

Primers specific for bisulfite-treated sense strand DNA were constructed. Generation of melt maps and calculation of dissociation constants were performed using the MELT87 program (19). Primers were selected to amplify a region located between positions −47 and 215 relative to the transcriptional starting point, which contains the area within the p15^INK4B promoter and exon 1 analyzed previously (20). GC-clamps were included in the primer sequences to allow subsequent DGGE analysis. Primers used for amplification of the p15^INK4B gene were: p15-F, 5′-[CCGCC]-TTTTTTGTAGGTTGGTTTTTTATTTTG-3′ (forward primer); and p15-R, 5′-[CCCGCCGCCCGCCGCTCGCCCGCCGCGCCCCTGCCCGCCGCCCCCGCCCG]-AAACTAAACTCAACTTCATTACCCTC-3′ (reverse primer). Nucleotides in brackets represent GC-clamps. PCR was carried out in a final volume of 25 μl containing 100–200 ng of bisulfite-treated DNA, 20 mm Tris-HCl (pH 7.5), 100 mm KCl, 1.5 mm MgCl₂, 1 mm DTT, 0.1 mm EDTA, 0.5% Tween 20 (v/v), 0.5% NP40 (v/v), 0.2 mm cresol red, 12% sucrose, 0.2 mm each dNTP, 0.4 μm each primer, and 0.65 units of Expand High Fidelity enzyme (Boehringer Mannheim, Mannheim, Germany). PCR was initiated by hot-start, followed by 39 cycles of 94°C for 30 s, 55°C for 30 s, and 72°C for 30 s and a final extension at 72°C for 5 min. PCR was performed in a GeneAmp PCR System 9600 (Perkin-Elmer Corp., Foster City, CA).

MSP was performed essentially as described previously (20). Briefly, the PCR for p15-M and p15-U were performed in a final volume of 25 μl. Each reaction contained 100–200 ng of bisulfite-treated DNA, 10 mm Tris-HCl (pH 8.3), 50 mm KCl, 1.5 mm MgCl₂, 0.001% gelatin, 0.2 mm cresol red, 12% sucrose, 0.2 mm each dNTP, 0.5 μm each primer, and 0.75 units of AmpliTaq Gold (Perkin-Elmer). PCR conditions were as follows: an initial denaturing step of 10 min at 95°C; 40 cycles of 94°C for 30 s, 60°C (p15-M) or 56°C (p15-U) for 30 s, and 72°C for 30 s; and a final incubation at 72°C for 5 min.

Fifteen μl of the GC-clamped PCR product were loaded onto a 6% polyacrylamide gel containing a 10–65% gradient of urea and formamide. Gels were run at 160 V for 5 h in 1× Tris acetate/EDTA buffer kept at a constant temperature of 58°C. After electrophoresis, gels were stained in 1× Tris acetate/EDTA buffer containing ethidium bromide (2 μg/ml) and photographed under UV transillumination.

Bands were excised from denaturing gradient gels and incubated overnight in 100 μl of distilled water. One μl of a 100-fold dilution was reamplified using primers without GC-clamps. Primer sequences were as follows: p15-F-Seq, 5′-TTTTTTGTAGGTTGGTTTTTTATTTTG-3′; and p15-R-Seq, 5′-AAACTAAACTCAACTTCATTACCCTC-3′. Nucleotides and primers were removed from PCR products using the Wizard PCR Preps Purification System (Promega Corp., Madison, WI), according to the manufacturer’s directions, and eluted in 30 μl of distilled water. Two μl of purified PCR product were sequenced on a 377 ABI Prism automatic sequencer (Applied Biosystems, Foster City, CA) using Thermo Sequenase dye terminator cycle sequencing premix kit (Amersham Life Science Inc., Cleveland, OH), according to the manufacturer’s instructions. Both strands were sequenced using the above primers.

The 207-bp region of the p15^INK4B promoter analyzed in this study is depicted in Fig. 1,A. It contains 27 CpG dinucleotides and 44 non-CpG cytosines and encompasses the primer-binding sites for the MSP assay described by Herman et al. (20). Primers specific for the sense strand of bisulfite-reacted DNA were designed to amplify this region without discriminating between methylated and unmethylated sequences. To optimize the resolution of sequence variants by DGGE, the primers were extended by G+C-rich sequences (termed GC-clamps; Ref. 21). Computerized simulation of DNA melting (19) predicted that a GC-clamped PCR product containing the entirely unmethylated p15^INK4B region has a two-domain melting profile with a T_m of the lower domain of 66.9°C (Fig. 1,B, solid line). In contrast, the fully methylated sequence results in a lower domain T_m of 73.7°C (Fig. 1,B, dashed line), and a partially methylated sequence containing 10 methylated cytosine residues shows an intermediate T_m (70.1°C; Fig. 1 B, dotted line). Thus, an increase in the content of methylated cytosines in native DNA is reflected by a dramatic increase in the thermal stability of the bisulfite-reacted species, providing the basis for separating methylated and unmethylated sequences by DGGE.

The use of bisulfite-DGGE to resolve different p15^INK4B epigenotypes is demonstrated in Fig. 1 C. The MOLT-4 cell line was shown by genomic sequence analysis (22) to be methylated at all of the 27 putative methylation sites within the target sequence (data not shown) and accordingly served as a positive control for methylation. In contrast, the cell line HL-60 is entirely unmethylated and served as a negative control. Parallel bisulfite-DGGE analysis of the two cell lines showed a single well-focused band for HL-60 and a slightly broader band positioned lower in the gel for MOLT-4 (Fig. 1,C). The observed band broadening can probably be ascribed to the melting behavior of the amplification product. As shown in the melt map (Fig. 1,B, dashed line), the melting profile of the fully methylated sequence is characterized by the presence of a short intramolecular sequence of low thermal stability, which may cause a significant increase in band width (23). Analysis of DNA isolated from blood MNCs from a normal volunteer showed a band migrating to a position corresponding to that of the negative control (Fig. 1 C), suggesting that p15^INK4B in normal blood cells is unmethylated.

We analyzed the methylation status of p15^INK4B in blood or bone marrow MNCs from 65 patients with AML at diagnosis. An example of the data obtained by bisulfite-DGGE analysis is shown in Fig. 2. Samples were scored as positive for p15^INK4B methylation when distinct bands or DNA smears below the negative control were observed after ethidium bromide staining of the gel, indicative of sequences with a higher GC content due to methylated cytosines not modified during the bisulfite treatment. Using this approach, p15^INK4B promoter methylation was detected in 46 of the patients (71%). Generally, a high number of different bands was observed both within and between patients, suggesting a high degree of intra- and interindividual heterogeneity of p15^INK4B methylation. The front line of bands and smears in each individual patient was generally positioned higher in the gel than the band originating from the fully methylated control cell line. Considering the theoretical association between degree of methylation and distance of migration in a denaturing gradient gel, this observation implies partial p15^INK4B methylation in the vast majority of malignant cells.

To demonstrate that bands below the negative control in a denaturing gradient gel represent DNA amplified from partially methylated sequences and to examine whether a correlation exists between band migration and the number of methylated sites, we recovered DNA from individual bands representing samples 4-2, 4-8, 4-22, and MOLT-4 from the gel and reamplified. Fig. 3,A shows a DGGE analysis of the reamplified PCR products, arranged according to migration distance. An example of the sequence analysis of individual bands is shown in Fig. 3,B. Cytosines not present in a CpG dinucleotide were invariably converted to uracils by the bisulfite treatment and appeared as thymidines after PCR. The number of unconverted cytosines revealed by the sequence analysis are indicated above each lane in Fig.3,A, showing a clear correlation between the distance of band migration and the extent of p15^INK4B promoter methylation. Fig. 3 C lists the precise locations of methylated cytosines in the six samples, showing that methylation was not specifically confined to any of the 27 CpG sites within the investigated region.

We also examined the methylation status of p15^INK4B in the 65 AML patients by the more sensitive MSP method (Fig. 4). Sixty-one (94%) of the samples were found to be methylated at the p15^INK4B promoter when evaluated by MSP, in contrast to only 46 (71%) when analyzed by bisulfite-DGGE. To address the basis for this discrepancy, we applied both methods to peripheral blood MNCs from 10 normal volunteers. Although none of these individuals was positive for p15^INK4B methylation with bisulfite-DGGE (data not shown), all 10 were found to be methylation positive by MSP (Fig. 5,A). Sequence analysis of PCR products generated with MSP showed the presence of multiple unconverted cytosines in the region between the primer binding sites (data not shown), excluding the possibility of nonspecific amplification. This indicates that a small fraction of normal cells are methylated in the promoter of the p15^INK4B gene. To identify these methylated normal cells, we separated MNCs from peripheral blood from three normal individuals into lymphocytes and monocytes by immunomagnetic bead isolation. The results from the bisulfite-DGGE and MSP analyses of DNA from these fractions are shown in Fig. 5 B. None of the fractions showed evidence of methylated p15^INK4B alleles when evaluated by bisulfite-DGGE. The MSP analysis, on the other hand, was clearly positive for all three lymphocyte fractions but negative for the monocyte fractions.

To examine the relation between the extent of p15^INK4B methylation and different FAB classes, we compared the methylation status of two patient groups, one with the myeloid subtypes M1 and M2 and another with the monocytic subtypes M4 and M5 (Table 1). When analyzed by bisulfite-DGGE, MNCs from patients in the immature group were more frequently p15^INK4B methylated than MNCs from patients in the more differentiated group (P < 0.025, χ² test).

Detection and characterization of aberrantly methylated alleles of specific genes in neoplastic tissues have previously relied upon two different methodological principles. The first of these principles is to determine the methylation status of single CpG sites by using either methylation-sensitive restriction enzymes or PCR primers designed to specifically amplify methylated sequences after treatment of DNA with bisulfite (MSP). The major disadvantage of this approach is that partial methylation involving only sites outside recognition or primer binding sequences will not be detected. The second principle is to collectively amplify methylated and unmethylated alleles from a sample of bisulfite-treated DNA followed by assessment of the methylation status of all CpG sites within the amplified region by sequence analysis. Although powerful, detection of methylated alleles in a tumor sample by bisulfite genomic sequencing may be complicated by the contamination of samples with nonneoplastic cells. Furthermore, the exact methylation profile of individual alleles can only be determined through the inclusion of an extensive and laborious subcloning step.

To circumvent these drawbacks, we have devised an electrophoretic method that displays, in great detail, the composite methylation profile of a specific gene. This method combines nondiscriminatory amplification of methylated and unmethylated sequences using bisulfite-treated DNA as template and resolution of differentially methylated alleles by DGGE. The principle is that an increase in DNA methylation, i.e., an increase in the numbers of CpG sites involved, is associated with an increase in GC content and, hence, in thermal stability. Using a region of the p15^INK4B promoter CpG island as a model, we demonstrated that a fully methylated sequence and the corresponding unmethylated sequence after treatment of DNA with sodium bisulfite differ in T_m by >6°C and can be separated by several centimeters in a standard denaturing gradient gel. A particularly attractive feature of this method is that sequences with different degrees of methylation will be retarded at different positions in the gel, implying that epigenetic clonotypes will be revealed as distinct bands that may be recovered for sequence analysis without the need for subcloning.

We have used this method to characterize the patterns of p15^INK4B methylation in 65 patients with AML. Overall, band patterns indicative of promoter hypermethylation were observed in 71% of the samples. This frequency is lower than but not statistically different from that reported by Herman et al. (7), who used methylation-sensitive enzymes and Southern hybridization. Taken together, these two independent sets of data suggest that p15^INK4B hypermethylation is a feature shared by the majority of AML patients. A highly surprising finding in this study was, however, the exceedingly variable patterns of p15^INK4B methylation both within and between patients. Most samples displayed a DNA smear by bisulfite-DGGE analysis with only a few distinct bands, suggesting the presence of an indefinite number of differently methylated sequences and only occasional clonotypic epigenotypes. Sequence analysis of individual bands recovered from denaturing gradient gels showed that the majority of individual p15^INK4B epigenotypes consisted of partially methylated sequences and that methylation was not specifically confined to any of the 27 CpG sites within the investigated region. Because no association between various degrees of p15^INK4B methylation and levels of p15^INK4B expression has yet been established, the biological consequences of the complex methylation patterns in AML patients remain unknown. One possibility is that the majority of all epigenotypes are associated with loss of p15^INK4B expression, either because full methylation is not required for silencing of the p15^INK4B gene or because CpG sites critical for gene silencing are located outside the investigated region and are invariably methylated. Another possibility is that different subclones in AML express different levels of p15^INK4B, which may modulate the malignant phenotype.

Although the origin of the extensive p15^INK4B methylation mosaicism in AML remains elusive, four possible mechanisms may be considered. (a) The original neoplastic clone or an early subclone may initially become fully methylated but gradually lose methylation over time due to epigenetic infidelity during replication. (b) An early subclone may initially acquire methylation at specific CpG sites, resulting in growth advantage and clonal selection, followed by subsequent spread of methylation in individual cells. (c) Multiple subclones may become individually and stably de novo methylated but to different degrees. (d) Methylation and demethylation are dynamic and stochastic processes with kinetics influenced by chromatin structure and binding of protein factors.

Only few previous studies have provided detailed characterization of methylation of putative tumor suppressor genes in tumor samples. Using a broad panel of quantitative and qualitative methods for the assessment of methylation status, Gonzalgo et al. (24) showed that the 5′ CpG island of the p16^INK4A gene is either fully methylated or unmethylated in malignant melanoma. In contrast, Stirzaker et al. (13) found, by quantitative genomic sequence analysis and subcloning experiments, that the extent of methylation of the Rb gene in primary retinoblastoma tumors may vary considerably both within and between patients. These studies, together with the present data, suggest that the dynamics of DNA methylation may differ between different genes, tumor types, different patients with the same tumor type, and even different tumor cells within the same individual. This variation necessitates great caution in choice of methodology for accurately determining the levels and patterns of methylation.

Another surprising finding in the present study was that p15^INK4B methylated cells could be consistently demonstrated in blood samples from healthy donors. After MNCs were separated from normal blood samples into monocytic and lymphocytic fractions, aberrantly methylated p15^INK4B alleles were demonstrated in the lymphocytic fractions by MSP but not by DGGE. No methylation was observed in the monocytic fractions. Recently, evidence of p16^INK4A hypermethylation in normal breast tissue and normal colon mucosa (25, 26) and p15^INK4B hypermethylation in normal colon mucosa (25) was presented. However, hypermethylation of the gene regions examined in these studies may not correlate with transcriptional silencing. Although the biological significance of p15^INK4B methylation in a subpopulation of normal lymphocytes, therefore, remains elusive, the background of p15^INK4B-methylated alleles in normal blood may constitute a significant problem for some PCR-based diagnostic procedures. In this study, 94% of the samples were found to be p15^INK4B methylated by MSP, which can detect 1 methylated allele in 1000 unmethylated alleles (17). In contrast, only 71% of the samples were found positive for p15^INK4B methylation by the less sensitive bisulfite-DGGE. The higher rate of methylation positive samples observed with MSP may, at least in part, be ascribed to admixture of methylated nonmalignant cells. Alternatively, the sensitivity of ethidium bromide staining and visual inspection of denaturing gradient gels, as performed in the this study, may be too low to detect cases in which only a small proportion of leukemic blasts carry methylated p15^INK4B alleles.

Changes in DNA methylation have emerged as a common feature in both solid tumors and hematological malignancies (27). Within the group of lymphoid malignancies, recent studies have demonstrated that both the p15^INK4B and p16^INK4A genes are frequently hypermethylated in multiple myeloma (28) and Burkitt’s lymphoma (29), whereas p15^INK4B is selectively methylated in T acute lymphoblastic leukemia (30). In the myeloid disorders, myelodysplastic syndrome patients with p15^INK4B hypermethylation have been found to have an adverse prognosis (31). Moreover, because p15^INK4B methylation was associated with a high number of marrow myeloblasts, this would constitute another finding in favor of methylation as a late, but important event in carcinogenesis.

A multitude of prognostic factors have been suggested for AML ranging from those related to tumor burden (e.g., leukocyte count at diagnosis or extramedullary involvement) over cytochemical staining and immunophenotyping to more recent ones related to molecular biology. At present, it is generally acknowledged that cytogenetic lesions are of the utmost importance for the outcome of AML patients (for review see Ref. 32). This is the first study to address the frequencies of p15^INK4B promoter methylation within FAB subgroups. The finding that methylation is seen less often in the monocytic variants M4 and M5 than in the M1 and M2 myeloid ones is interesting but somewhat puzzling. Thus, although the prognosis of M4 patients is favorable compared to AML as a whole, possibly due to the high frequency of the inv16 abnormality in these patients (33), that of M5 is worse due to a high preponderance for extramedullary involvement. M1 and M2 patients are heterogeneous in terms of prognosis but generally believed to display high relapse rates consistent with p15^INK4B hypermethylation. Collectively, these data could argue against p15^INK4B methylation as a major contributor to the development of AML. However, it might still be an important factor in leukemogenesis in AML patients, in whom other factors are less prone to induce an overt malignant phenotype. More patients tested longitudinally are needed to address this question, and we are presently pursuing this line of investigation in a multivariate study with our newly developed multiplex PCR assay (34) to obtain molecular diagnosis in a higher fraction of patients.

We thank Karin Brændstrup for excellent technical assistance and Tommy Byskov Lund for review of the manuscript.