Genome-Wide Profiling of CpG Methylation Identifies Novel Targets of Aberrant Hypermethylation in Myeloid Leukemia

The methylation of CpG islands is associated with transcriptional repression and, in cancer, leads to the abnormal silencing of tumor suppressor genes. Because aberrant hypermethylation may be used as a marker for disease, a sensitive method for the global detection of DNA methylation events is of particular importance. We describe a novel and robust technique, called methyl-CpG immunoprecipitation, which allows the unbiased genome-wide profiling of CpG methylation in limited DNA samples. The approach is based on a recombinant, antibody-like protein that efficiently binds native CpG-methylated DNA. In combination with CpG island microarrays, the technique was used to identify >100 genes with aberrantly methylated CpG islands in three myeloid leukemia cell lines. Interestingly, within all hypermethylation targets, genes involved in transcriptional regulation were significantly overrepresented. More than half of the identified genes were absent in microarray expression studies in either leukemia or normal monocytes, indicating that hypermethylation in cancer may be largely independent of the transcriptional status of the affected gene. Most individually tested genes were also hypermethylated in primary blast cells from acute myeloid leukemia patients, suggesting that our approach can identify novel potential disease markers. The technique may prove useful for genome-wide comparative methylation analysis not only in malignancies.


Introduction
The transcriptional state of a gene is largely determined by its local epigenetic code.Pathologic changes in chromatin structure or DNA methylation that lead to abnormal expression or repression of genes are now being recognized as important contributing factors in developmental diseases, cancer, and aging (1)(2)(3)(4)(5).
In cancer, the epigenetic silencing of tumor suppressor genes seems to be a common, nonrandom, and tumor type-specific event that often coincides with the aberrant methylation of CpG dinucleotides in so-called CpG islands.These regions of high CpG content are mostly unmethylated in normal cells and frequently contain gene promoters and transcription start sites (2,(6)(7)(8).
Hypermethylation of CpG islands seems to be a tumor typespecific event (9,10), and current efforts are concentrating on finding ways to exploit the diagnostic and therapeutic implications of the abnormalities (11,12).A comprehensive knowledge of the methylation profile of a given tumor may provide important information for risk assessment, diagnosis, monitoring, and treatment (6,13).
The investigation of aberrant CpG island methylation has primarily taken a candidate gene approach.Assessment of the clinical potential of hypermethylation profiles and the identification of relevant marker genes, however, require means and methods to detect hypermethylation on a genome-wide level.
Here, we apply a novel and relatively simple technique that allows the generation of unbiased, genome-wide profiles of normal or aberrant CpG island methylation.The procedure is independent of methylation-sensitive restriction endonucleases or bisulfite treatment, which are of limited use for genome-wide profiling strategies (14).Our approach is based on a recombinant methyl-CpG-binding, antibody-like protein that was engineered to efficiently bind CpGmethylated DNA fragments in a so-called methyl-CpG immunoprecipitation (MCIp) assay.In combination with appropriate DNA microarrays, MCIp-enriched DNA fragments can be used to detect hypermethylation events on a genome-wide scale.The efficacy of this system was tested by the profiling of three leukemia cell lines using a 12K CpG island microarray, which led to the identification of many novel genes that are hypermethylated in human myeloid leukemia.The novel profiling technique provides a powerful tool to identify changes in DNA methylation as well as novel marker genes (Stratagene, Heidelberg, Germany).The PCR product was directly cloned into NotI/NheI sites of signal-pIg plus vector (Ingenious, R&D Systems, Abingdon, United Kingdom) and sequence verified.An ApaI/NheI fragment of pIg/MBD-Fc, containing the MBD of human MBD2 fused to the Fc tail of human IgG1, was subcloned into ApaI/SpeI sites of pMTBiP/V5-His B (Invitrogen, Karlsruhe, Germany), resulting in pMTBip/MBD-Fc.
DNA preparation.Genomic DNA from various cell types was prepared using the Blood and Cell Culture DNA Midi kit (Qiagen, Hilden, Germany).At least 1 Ag genomic DNA was digested using MseI or a combination of MseI and Csp6I.Completion of the digest was controlled using agarose gel electrophoresis, and digested DNA was quantified using PicoGreen dsDNA quantitation reagent (Molecular Probes, Karlsruhe, Germany).
Protein expression.Drosophila S2 cells (4 Â 10 6 ) per 60-mm cell culture dish were stably transfected with a mixture of 1.5 Ag pMTBip/MBD-Fc and 0.3 Ag pCoHygro (Invitrogen) using Effectene transfection reagent (Qiagen) according to the manufacturer's protocol.For large-scale production, 1 Â 10 8 to 5 Â 10 8 cells were cultured in 100 to 200 mL Insect-Xpress without FCS and hygromycin in 2,000 mL roller bottles for 2 days before the addition of 0.5 mmol/L CuSO 4 .Culture medium was harvested every 4 to 7 days, and cell culture supernatants were purified using a protein A-Sepharose column.
Reverse Southwestern blot.DNA fragments were separated by agarose gel electrophoresis and directly blotted (without prior denaturation) onto a nylon membrane using a capillary transfer system equivalent to traditional Southern blotting procedures.DNA was UV cross-linked after transfer and blocked overnight with 5% fat-free powdered milk in TBST [20 mmol/L Tris-HCl (pH 7.4), 150 mmol/L NaCl, 0.1% Tween 20].Blots were washed thrice in TBST for 10 minutes at room temperature.Blots were then incubated with MBD-Fc (1:20,000 dilution) in TBST with 5% milk powder for 1 hour at room temperature, washed thrice (TBST, 10 minutes, room temperature), and incubated with horseradish peroxidase (HRP)-conjugated anti-human Fc (1:10,000) in TBST with 5% milk powder for 1 hour at room temperature.After three more washes (TBST, 10 minutes, room temperature), bands were detected using enhanced chemiluminescence (ECL; Amersham, Little Chalfont, United Kingdom).
Methyl-CpG immunoprecipitation.Purified MBD-Fc protein (usually 7.5 Ag) was added to 50 AL protein A-Sepharose 4 Fast Flow beads (Amersham) in 1 mL TBS and rotated overnight on a rotator at 4jC.
On the next day, MBD-Fc beads were transferred into SpinX columns (Sigma-Aldrich) and spin washed twice with buffer A [20 mmol/L Tris-HCl (pH 8), 2 mmol/L MgCl 2 , 0.5 mmol/L EDTA, 300 mmol/L NaCl, 0.1% NP40].Digested DNA (usually 300 ng) was added to the washed MBD-Fc beads in 350 AL buffer A and rotated in sealed SpinX columns for 3 hours on a rotator at 4jC.Beads were centrifuged to recover unbound DNA (300 mmol/L fraction) and subsequently washed with increasing NaCl concentrations (400-600 mmol/L).Flow through of each wash step was either discarded or collected in separate tubes for further analyses.Highly CpG-methylated DNA was eluted with 350 AL buffer E [20 mmol/L Tris-HCl (pH 8), 2 mmol/L MgCl 2 , 0.5 mmol/L EDTA, 1,000 mmol/L NaCl, 0.1% NP40] into a separate tube.Eluted DNA was desalted using QIAquick PCR purification kit (Qiagen).In parallel, 300 ng digested DNA was resuspended in 350 AL buffer D and desalted using QIAquick PCR purification kit.DNA preparations were quantified using PicoGreen dsDNA quantitation reagent.
Real-time genomic PCR.Enrichment of a specific MseI fragment in the MCIp eluate or in MCIp amplicons was detected and quantified relative to the genomic input by real-time LightCycler PCR using the QuantiTect kit (Qiagen) according to the manufacturer's instructions.Primer sequences are given in the Supplementary Table S2.Cycling parameters were as follows: denaturation, 95jC, 15 minutes and amplification, 95jC, 15 seconds, 57jC, 20 seconds, and 72jC, 25 seconds for 42 cycles.The product size was initially controlled by agarose gel electrophoresis, and melting curves were analyzed to control for specificity of the PCRs.
Real-time reverse transcription-PCR.Total RNA was prepared using the RNeasy Midi kit.Contaminating genomic DNA was removed from the samples using TURBO DNA-free kit (Ambion, Huntingdon, United Kingdom) before 1 Ag total RNA was reverse transcribed using Moloney murine leukemia virus reverse transcriptase, RNase H minus, point mutant (Promega, Mannheim, Germany).Real-time PCR was done using the LightCycler (Roche, Mannheim, Germany) with the QuantiTect kit according to the manufacturer's instructions.Primers used are given in Supplementary Table S4.Cycling parameters were as above.
CpG island microarray analysis.To generate fluorescently labeled DNA for CpG island microarray hybridization, MseI-compatible unidirectional ligation-mediated PCR (LMPCR) linker (LMPCR_S-L: 5 ¶-GCGGTGACC-CGGGAGATCTCTTAAG-3 ¶ and LMPCR_AS-L: 5 ¶-TACTTAAGAGATC-3 ¶, 20 Amol/L) was ligated to the MCIp-eluted DNA and in a separate reaction to an equal amount of input DNA (0.5 AL linker/ng DNA) in 60 AL reactions using 1,200 units T4 ligase (New England Biolabs, Frankfurt am Main, Germany) at 16jC overnight.Linker-ligated DNA was desalted using QIAquick PCR purification kit.Amplification of linker-ligated DNA preparations was done using LMPCR primer (5 ¶-GTGACCCGGGA-GATCTCTTAAG-3 ¶) and Taq polymerase (Roche) in the presence of 1.3 mol/L betaine.Amplicons were desalted using QIAquick PCR purification kit and quantified (PicoGreen dsDNA quantitation reagent).Labeling and hybridization of MCIp amplicons were done by the Kompetenzzentrum fu ¨r fluoreszente Bioanatytik (KFB; Regensburg, Germany) according to the protocol provided by the CpG island microarray manufacturer (University Health Network Microarray Centre, Toronto, Ontario, Canada) with modifications.Briefly, normal and tumor MCIp amplicons (4 Ag) were directly labeled with Cy5-dUTP and Cy3-dUTP, respectively, using the BioPrime Array Comparative Genomic Hybridization Genomic Labeling System (Invitrogen).Each fluorescently labeled and purified DNA amplicon (4 Ag) in 300 AL digoxygenin Easy Hyp solution (Roche) supplemented with 25 Ag Cot-1 DNA (Invitrogen) and 30 Ag yeast tRNA was hybridized to human CpG 12K arrays (HCGI12K, University Health Network Microarray Centre) in Gene Frames measuring 60 Â 21 mm (ABgene, Hamburg, Germany) at 37jC for overnight.Slides were washed thrice in 1Â SSC, 0.1% SDS at 50jC for 10 minutes.After two more rinses with 0.1Â SSC, slides were dried and scanned using the Affymetrix 428 scanner (Wooburn Green, United Kingdom).Images were analyzed using the ImaGene 5.6 and Gene Sight Lite software (BioDiscovery, Inc., El Segundo, CA).Locally weighted scatterplot smoothing normalization was used to normalize Cy3 and Cy5 signals.Clones that produced reproducible differential signals on the CpG island microarray were sequenced by the University Health Network Microarray Centre.
Affymetrix microarray analysis.RNA from KG-1, U937, and THP-1 cells as well as from freshly isolated human blood monocytes of a health donor was analyzed using Affymetrix HG-U133_Plus_2 arrays.Hybridization, cRNA labeling, and data handling were done by the KFB.
Sodium bisulfite sequencing.Modification of DNA with sodium bisulfite was done as described previously (16) with modifications (17).Bisulfite-treated DNA was amplified in individual PCRs using the primers given in Supplementary Table S3.PCR products (representing the sense strand) were cloned using the TOPO cloning kit (Invitrogen), and several individual clones were sequenced.
Databases for sequence analysis.Mapping of CpG island clones was done using the University of California at Santa Cruz genome browser (http://genome.ucsc.edu/).Expression profiles for myeloid cell types other than monocytes were obtained from Genomics Institute of the Novartis Research Foundation SymAtlas (http://symatlas.gnf.org/SymAtlas/).
Gene Expression Omnibus accession numbers.Microarray data have been deposited with the Gene Expression Omnibus database (accession numbers for Affymetrix gene arrays: GSM73641, GSM73642, GSM73644, and GSM73645; accession numbers for CpG island microarrays: GSM91623-30).

Results
Design, generation, and properties of a recombinant antibody-like MBD-Fc.To enable an antibody-like detection of double-stranded CpG-methylated DNA, we constructed a vector encoding a fusion protein comprising the MBD of human MBD2, a flexible linker polypeptide, and the Fc portion of human IgG1.The MBD-Fc polypeptide was expressed under the control of a metalinducible promoter in Drosophila S2 cells and collected from the supernatant via protein A affinity chromatography.The purified protein had the expected molecular weight of 40 kDa.To test whether MBD-Fc was able to detect CpG-methylated DNA, we blotted in vitro-methylated PCR fragments with different CpG density onto a nylon membrane using a capillary transfer system equivalent to traditional Southern blotting, however, without denaturing the DNA before blotting.As shown in Fig. 1A, using standard immunoblot conditions and MBD-Fc as equivalent to the primary antibody, methylated DNA could be detected on nylon membranes in a linear fashion.
We next tested MBD-Fc binding of in vitro generated and differentially methylated DNA fragments in an immunoprecipitation-like approach.PCR fragments of human promoters with varying CpG density were generated using PCR and in vitro CpG methylated using SssI or left unmethylated as described in Supplementary Data.The mixture of methylated and unmethylated DNA fragments was bound to MBD-Fc protein A-Sepharose beads and eluted using increasing concentrations of NaCl.Fractions were collected, spin purified, and subjected to agarose gel electrophoresis.As shown in Supplementary Fig. S1, the affinity of a methylated fragment increased with the density of methylated CpG dinucleotide, with unmethylated DNA eluting at relatively low salt concentrations and highly methylated DNA eluting at high salt concentrations.Variation of the amount of input DNA within the binding capacity of the MBD-Fc polypeptide did not significantly change the elution profile.However, the salt-dependent affinity of methylated DNA was contingent on the density of the MBD-Fc fusion protein on the protein A-Sepharose beads.To clarify whether this approach allows the detection of differential methylation of a single gene locus, a CpG island promoter fragment was cloned into a CpG-free vector.The DNA was partially in vitro methylated, fractionated by MCIp, and subjected to bisulfite sequencing.As shown in Supplementary Fig. S2, partially in vitro-methylated DNA fragments were separable according to their methylation degree in a fractionating approach using increasing concentrations of NaCl.
Combination of MCIp and real-time PCR to detect the methylation status of specific CpG island promoters.We assumed that MCIp may be used to discriminate methylated and unmethylated DNA fragments from genomic DNA.To explore this type of application on single-gene level, we used the above procedure to precipitate MseI-restricted genomic DNA of in vitro SssI methylated and untreated normal DNA from monocytes of a healthy donor.MseI was chosen for DNA fragmentation because it is known to preferentially cut in regions of low CpG content while leaving many CpG islands uncut (18).
The salt concentration-dependent enrichment of four different CpG island promoters and a promoter with low CpG density was determined in SssI-methylated and untreated DNA relative to the input DNA using LightCycler real-time PCR.As a positive control for DNA methylation, we used the SNRPN gene promoter that is subject to maternal imprinting with one of its two copies being methylated also in normal cells (19).In normal DNA, the two differentially methylated allele fragments of SNRPN were enriched in two separate fractions (Fig. 1B).Only one enriched fraction was observed with SssI-methylated DNA.For CDKN2B gene (also known as p15 INK4b ), which is frequently methylated in leukemia cells (Fig. 1C; refs.[20][21][22], the fragment was detected mainly in a low salt fraction from normal DNA and in the high salt fraction from SssI-methylated DNA.Similar results were obtained for the human estrogen receptor 1 (ESR1) and human TLR2 genes (data not shown).The profiles of methylated and unmethylated DNA at the CHI3L1 locus were significantly different from those of the abovetested CpG island promoters (Fig. 1D).Most of the untreated CHI3L1 fragment was recovered at lower NaCl concentrations, and a shift was observed toward higher NaCl concentrations when the DNA was SssI methylated.The average difference between SssItreated and untreated monocyte DNA at the CHI3L1 locus is f5 to 6 (out of 12) methylated CpG residues (data not shown), suggesting that the fractionated approach is able to discriminate relatively small differences in CpG methylation.Analysis of the above elution profiles suggests that (a) a 200-to 300-fold enrichment of stronger over less methylated genomic fragments can be obtained in either low or high salt fractions, (b) fragments with low CpG density are largely excluded from the high salt fraction, and (c) the fractionated MCIp approach may allow the resolution of relatively small differences in CpG methylation density.
To test whether MCIp can detect aberrant hypermethylation in tumor samples, DNA from three leukemia cell lines, KG-1 (AML), U937 (histiocytic malignancy, monocytic), and THP-1 (acute monocytic leukemia), as well as from monocytes of a healthy donor were analyzed for SNRNP, CDKN2B, ESR1, and TLR2 promoter enrichment in the high salt fraction (Fig. 2A).The TLR2 gene promoter was enriched in KG-1 and U937 cells but not in THP-1 or normal cells.The methylation pattern of TLR2 was confirmed by bisulfite sequencing (data not shown; ref. 17).Results for CDKN2B (methylated in KG-1 and unmethylated in U937) and ESR1 (methylated in KG-1) were in line with previously published studies (20,22,23).None of the three MseI fragments were significantly enriched in the DNA from normal cells.In concordance with its imprinting-related methylation status, the SNRPN gene promoter was significantly enriched in all leukemia cell lines as well as in normal cells.These experiments established that the high salt MCIp fraction specifically enriches CpG-rich genomic DNA fragments with a high degree of CpG methylation.
To test the sensitivity of our approach, decreasing amounts of U937 DNA were analyzed using the MCIp approach.The enrichment of TLR2 (strong methylation) and CDKN2B gene fragments (no methylation) was determined by LightCycler realtime PCR.As shown in Fig. 2B, a significant enrichment of the TLR2 fragment was achieved using as little as 1 ng genomic DNA fragments (equivalent to f150 tumor cells) for the MCIp procedure.
Samples derived from tumors may contain significant numbers of normal cells, which are expected to be unmethylated at most CpG islands.To test the linearity of methyl-CpG detection with respect to cell purity, MCIp was done using mixtures of DNA from normal blood cells and from the leukemia cell line KG-1 showing high levels of CpG island methylation at several promoters.As shown in Fig. 2C, the TLR2 promoter fragment was only detected in samples containing KG-1 DNA and the signal gradually increased with the proportion of methylated DNA in the sample.Similar results were obtained for the ESR1 locus (data not shown).
Combination of MCIp and CpG island microarray analysis to generate unbiased genome-wide promoter methylation profiles.To achieve a genome-wide analysis of CpG island DNA methylation, we combined MCIp with CpG island microarray hybridization.A small amount of MseI-restricted DNA (300 ng) of three leukemia cell lines (KG-1, U937, and THP-1) and of normal human peripheral blood monocytes was subjected to MCIp.The isolated DNA was amplified using LMPCR.
The resulting amplicons were directly labeled with Cy5 (normal DNA) and Cy3 (leukemia cell DNA), and each leukemia sample was cohybridized with the normal control sample to CpG island microarrays (University Health Network Microarray Centre; Supplementary Fig. S3).This array contains 12,192 CpG island clones from a MseI CpG DNA library that was originally prepared by MeCP2 column purification of nonmethylated CpG island fragments (18).Representative scatter plots of microarray hybridizations are presented in Supplementary Fig. S3C.In hybridizations using the amplicons from tumor cell lines, signals corresponding to both hypomethylated and hypermethylated fragments were observed.Because we were interested in tumor-specific epigenetic silencing, we focused on the analysis of hypermethylated fragments.To identify CpG island fragments that were affected by hypermethylation, results of three independent MCIp experiments (using two different MBD-Fc preparations and three independent DNA preparations) were analyzed in conjunction.Hybridization signals that were consistently different (>2-fold enriched in the leukemia sample) in at least one cell line were selected for further analysis.
In total, THP-1 cells showed 277 differential hybridization signals, U937 cells showed 454, and KG-1 cells showed 330.One hundred ninety-one of 535 spots in total were unambiguously annotated and located in the close proximity (approximately F3,000 bp) to predicted transcriptional start sites and were chosen for further analysis.Several sequences were represented more than once on the CpG island microarray.The final, nonredundant list of differentially represented CpG island DNA fragments contained 131 entries that were in the close proximity to 134 genes (Table 1).At least nine of the listed genes [LMX1A (24), TFAP2A (25), CR2 (26), DCC (27), MYOD1 (28), DLEC1 (29), AKAP12 (30), SSIAH2 (LOC283514), and FOXF1 (31)] have been identified previously as targets of hypermethylation in cancer.
The hypermethylated genes listed in Table 1 are involved in many biological functions.Most strikingly, half of the genes with an assigned molecular function (46 of 89) are involved in DNA binding and transcriptional regulation.
Experimental validation of microarray results.A representative number of gene fragments that were identified using combined MCIp chip analyses were selected for further validation.LightCycler real-time PCR was used to measure the MCIp enrichment of 29 candidate genes (Fig. 3; Supplementary Figs.S4  and S5).Out of these, 26 gene fragments were enriched in a manner comparable with the results obtained by microarray analysis.To  validate MCIp-detected methylation differences using an independent approach, the methylation status of six CpG island fragments [JUN, RAB3C, MAFB, KLF11, ZNF516, and SSIAH2 (LOC283514)] was additionally determined using bisulfite sequencing.As shown in Supplementary Fig. S6, the degree of methylation as determined by bisulfite sequencing correlated well with the results obtained by MCIp.In several cases, the MseI fragment represented on the CpG island microarray did not include the proximal promoter.Because transcription factors may have a particular role in leukemogenesis, DNA fragments that included transcriptional start sites of the four transcription factor genes (MAFB, KLF11, JUN, and ZNF516) were additionally analyzed by MCIp.Whereas JUN promoter fragments were not significantly methylated in any of the samples (data not shown), MAFB, KLF11, and ZNF516 promoter fragments also showed significant methylation (Supplementary Fig. S5).
Global comparison of CpG island methylation and mRNA expression.Complementary mRNA expression data of the above cell lines were generated by microarray analysis using the human HG-133 Plus 2 Array (Affymetrix).A side by side comparison of CpG methylation and mRNA expression data is presented in Table 1.Interestingly, more than half of the genes (69 of 125) were undetectable ('absent') by microarray analysis in all samples.In cases where significant signals were detected, the mRNA expression of genes that were found to be hypermethylated in tumor cell lines was often low relative to normal monocytes (e.g., PLA2G7, FNBP1, MAFB, or ZNF516).In some cases, the degree of methylation showed no correlation with gene expression detected by microarray analysis (e.g., HOXA10, EPAS1, or SYNJ2), suggesting that CpG methylation in those cases may target regions not relevant for transcription.
Quantitative real-time PCR analysis of transcripts for five of the previously tested genes with hypermethylated CpG islands (JUN, MAFB, KLF11, SSIAH2, and ZNF516) confirmed their downregulation in leukemia cell lines relative to human blood monocytes (data not shown) and showed a significant derepression by treatment with decitabine (5-aza-2 ¶-deoxycytidine) in U937 cells (Figure 4).The effect of demethylation was most striking for MAFB and SSIAH2 (data not shown) that were induced up to 100-fold in treated cells.
Aberrant hypermethylation in AML.Tumor cell lines represent in vitro models of primary tumors, which may have acquired additional alterations on both genetic and epigenetic levels.To test whether genes that were found to be hypermethylated in the leukemia cell lines are also affected in primary tumors, we analyzed DNA of blast cells obtained from 12 AML patients for hypermethylation at promoter fragments of CDKN2B, MAFB (promoter  and upstream fragment), KLF11, ZNF516, KLF5, PFC, and SSIAH2.With the exception of ZNF516 (data not shown), a significant number of patients (ranging from two to nine) were markedly hypermethylated at each of the above loci (Fig. 5; Supplementary Fig. S6).These results suggest that the CpG island fragments identified in tumor cell lines can also be subject to hypermethylation in primary tumor cells and may represent novel biological markers for leukemia.Notably, the youngest patient (P20) was hypermethylated at all loci tested, whereas the oldest patient (P07) was only significantly methylated at the PFC CpG island, indicating that methylation of the above-tested CpG fragments does not correlate with aging.

Discussion
We describe a novel reagent and applications that allow the rapid and sensitive screening of DNA methylation.The central technique, called MCIp, consists of the binding of methylated DNA fragments to the bivalent, antibody-like fusion protein MBD-Fc (a methyl-binding domain fused to a Fc tail) in an immunoprecipitation-like approach.Enriched methylated DNA fragments can be efficiently detected on single-gene level or throughout the genome.The power of these novel techniques was shown by the identification of large number of genes that are affected by aberrant hypermethylation in myeloid leukemias.
Comparison with existing protocols.At present, many techniques are known, which are used for the detection of the CpG methylation of single known candidate gene loci (14).Commonly used assays rely on two basic approaches to distinguish methylated and unmethylated DNA: digestion with methylationsensitive restriction enzymes or bisulfite treatment of DNA.Methods allowing the analysis of the CpG methylation throughout the genome are less well established.In most techniques [e.g., restriction landmark genomic scanning (RLGS; ref. 32) or methylated CpG island amplification (33,34)], methylationsensitive restriction enzymes are used as a component of the method.Here, a major disadvantage is that the analyses only inform on the methylation status of the cytosine residues, which have been recognized by the methylation-sensitive restriction enzymes used.In addition, the selection of the restriction enzymes automatically limits the number of detectable sequences-a global analysis of CpG methylation is therefore not achieved.
The use of naturally occurring MBD proteins to separate methylated and unmethylated DNA fragments are known for more than a decade.Already in 1994, the laboratory of A. Bird developed a method for enriching methylated DNA fragments by means of affinity chromatography using recombinant MeCP2 (18).The technique has been used, improved, and combined with further techniques by other laboratories (35,36).A disadvantage of MeCP2 affinity chromatography is the large amount of genomic DNA required (50-100 Ag) and the relatively time-consuming procedure.In addition, a recent report by Klose et al. (37) showed that MeCP2 requires an A/T run adjacent to the methylated CpG dinucleotide for efficient DNA binding, suggesting that MeCP2 affinity chromatography will be biased toward certain CpG motifs.No binding requirements or preferences of MBD2 were detected in this and in previous studies.
We believe that the high methyl-CpG affinity of MBD2 (38), combined with the bivalent, antibody-like structure of the recombinant MBD-Fc polypeptide, largely increases its binding capacity, enabling the efficient retention of DNA fragments in dependence on their methylation degree.The properties of the recombinant MBD-Fc polypeptide allow its application in smallscale assays requiring only little amounts of DNA.This may actually permit the profiling of DNA methylation of candidate genes from very limited cell numbers, including biopsy samples or cells collected by laser-mediated microdissection.The MCIp approach presented in this report requires only little amounts (>300 ng) of genomic DNA for a complete genome-wide methylation profile when combined with an unspecific LMPCR amplification step (the latter step may be omitted if sufficient starting material is available; data not shown).An unmethylated DNA fragment may be 200-to 500-fold depleted, and up to 80% of a highly methylated fragment may be recovered in the high salt MCIp fraction, showing the high affinity of our recombinant polypeptide.In addition to enzymatic restriction, the DNA may also be fragmented by sonication, resulting in a similar enrichment of methylated fragments in the high salt fraction (data not shown).
A recent paper by Weber et al. (31) describes a related approach using a 5-methylcytosine (5mC) antibody and a denaturing step before the immunoprecipitation of DNA fragments.Their analysis revealed only a small set of promoters being methylated differentially in a normal and a transformed cell line, suggesting that aberrant methylation of CpG island promoters in malignancy might be less frequent than previously hypothesized.In contrast to

Cancer Research
Cancer Res 2006; 66: (12).June 15, 2006 their observations, we detected a much higher percentage of differentially methylated genes, much in line with previous estimates, using the same CpG island microarray platform.This may reflect an inherent property of the cell lines used and, however, may also point to a lesser sensitivity of the 5mC antibody approach compared with our fractionated MCIp approach.
Hypermethylated genes in leukemia cell lines.Our profiling of three leukemia cell lines identified a large number of gene fragments that are likely to be methylated in neoplastic cells.To our knowledge, this study provides one of the largest published collections of potentially hypermethylated genes in cancer cells.
Interestingly, most genes that were identified as hypermethylated in leukemia cell lines showed extremely low or undetectable mRNA expression levels in our microarray experiments.A comparison with published expression profiles for human bone marrow, CD33-positive bone marrow cells, as well as mature myeloid cells (http://symatlas.gnf.org/SymAtlas/;data not shown) indicates that a large proportion of these genes may not be significantly transcribed in myeloid cell types.A hypothetical (thus far unknown) targeting mechanism may therefore induce CpG methylation of genes independent of their transcriptional status during cellular differentiation.Although such genes may not have a significant suppressor role in tumor development and/or progression, they may still serve as valuable biomarkers, provided that the targeting mechanism is specific for the disease.Methylation profiling of larger sample groups, using the described (or similar) approaches, may help to clarify whether aberrant methylation of CpG islands in malignancies is random or specific.
Acute leukemia is characterized by a block of differentiation of early progenitors, which leads to the accumulation of immature cells in bone marrow and blood.The frequent mutation or downregulation of a relatively small number of transcription factors in AML patients suggests that the inactivation of transcriptional regulators may be critically involved into the malignant transformation process.Our methylation profiling of leukemia cell lines preferentially identified genes that are involved in transcriptional regulation.Half of the listed genes with an assigned molecular function (46 of 89) are involved in DNA binding and transcriptional regulation, which represents a significant overrepresentation.Aberrant hypermethylation of these transcription factor genes may lead to their epigenetic down-regulation and likely contributes to the observed differentiation arrest in leukemia cells.This observation is in line with a previous study from Rush et al. (39) that investigated the methylation status of a large set of CpG islands in AML patients using RLGS and also found that a large proportion of the known methylated promoters (4 of 11) corresponded to genes involved in transcriptional regulation.
The list of hypermethylation targets contains several transcription factor genes, including MAFB, JUN, and KLF11, which are highly expressed in normal myeloid cells.A good tumor suppressor candidate, for example, is represented by the bZip transcription factor MAFB, which is expressed specifically in the myeloid lineage of the hemopoietic system.Its expression is up-regulated successively during myeloid differentiation from multipotent progenitors to macrophages, suggesting an essential role of MAFB in early myeloid and monocytic differentiation (40).
In summary, the data provided validate the experimental and possibly diagnostic potential of the MCIp technique.The recombinant MBD-Fc protein and its application in DNA methylation analysis may represent an important step toward genome-wide CpG methylation profiling not only in cancer diagnostics.The identification of those hypermethylation targets that are relevant in cancer development and/or progression will be a major future challenge.

Figure 1 .
Figure 1.Detection of CpG-methylated DNA using the recombinant MBD-Fc protein.A, PCR fragment of a CpG island promoter (ICSBP gene) was methylated using Sss I, subjected to agarose gel electrophoresis [ethidium bromide (EthBr ) staining is control], and directly blotted onto a nylon membrane.Membranes were stained using MBD-Fc, HRP-conjugated anti-human Fc, and ECL reagent (Amersham).B to D, fractionated MCIp was used in combination with real-time LightCycler PCR to detect the methylation status of the indicated genes from untreated (gray columns ) and SssI-methylated and MseI-restricted genomic DNA fragments (black columns).Gene fragments recovered from MCIp eluates [NaCl concentrations (in mmol/L)] and an equivalent amount of genomic input DNA (gDNA) were amplified by LightCycler PCR.A standard curve (3 log scales) was used to obtain the concentration of a specific gene fragment in the MCIp eluate relative to its concentration in the genomic input DNA.Values of individual fractions represent the percentage of recovery and are calculated relative to the amount of PCR product generated from the respective genomic input DNA (100%).Columns, mean (n = 4) using at least two different preparations of MBD-Fc (10 log scales); bars, SD.A 3-kb region of the corresponding CpG island is represented above each figure.Vertical lines, CpG dinucleotide.Gray boxes, positions of exons.Arrows, transcription start sites.White boxes, 100 bp; black boxes, positions of the Mse I fragments that are detected.Numbers, number of CpG dinucleotides within the MseI fragment.

Figure 2 .
Figure2.MCIp detection of CpG island methylation in specific CpG island promoters using real-time PCR.A, SNRPN, TLR2, ESR1 , and CDKN2B gene fragments in the high salt (1,000 mmol/L after 600 mmol/L) MCIp fraction of three human myeloid leukemia cell lines (KG-1, U937, and THP-1) as well as normal human blood monocytes (N) were analyzed by real-time PCR as described in Fig.1.B, decreasing amounts of MseI-treated U937 DNA were subjected to MCIp.CDKN2B and TLR2 gene fragments were quantified as above.C, MseItreated DNA of normal human blood monocytes and KG-1 cells was mixed at the indicated ratios, the mixture was subjected to MCIp, and the TLR2 gene fragment was quantified in the 1,000 mmol/L fraction using LightCycler PCR as described above.

Figure 3 .
Figure 3. Validation of CpG island microarray results by single-gene MCIp.A to D, schematic representation of the MCIp enrichment detected by single-gene real-time LightCycler PCR for MAFB, JUN, KLF11 , and ZNF516 Mse I gene fragments in the three human myeloid leukemia cell lines (KG-1, U937, and THP-1) as well as in the normal human blood monocytes.Results are relative to the amount of PCR product generated from the genomic input DNA (100%) of each cell type.Columns, mean (n = 4) using at least two different preparations of MBD-Fc; bars, SD.Corresponding CpG island on top of each figure as described in Fig. 1.

Figure 4 .
Figure 4. Derepression of hypermethylated target genes by decitabine.MAFB, JUN, KLF11, and ZNF516 mRNA expression levels were quantified in U937 cells after decitabine treatment (5 Amol/L) using real-time PCR.The relative units were calculated from a standard curve with four different concentrations of log dilutions to the PCR cycle number at which the measured fluorescence intensity reaches a fixed value.Expression levels are relative to HPRT expression.Columns, mean of two experiments; bars, SE.

Figure 5 .
Figure 5. Methylation profiles of AML patients.Schematic representation of the MCIp enrichment detected by single-gene real-time LightCycler PCR for CDKN2B, MAFB (promoter), KLF11 (promoter), and KLF5 in MseI/Csp 6Irestricted DNA in the 12 AML samples (AML-P no.; age in years) as well as two normal human blood monocytes samples (N1 and N2).Results are relative to the amount of PCR product generated from the genomic input DNA (100%) of each cell type.Columns, mean of at least three LightCycler amplifications using at least two different preparations of MBD-Fc; bars, SE.

Table 1 .
Hypermethylated gene fragments in myeloid leukemia cell lines (Cont'd) June 15, 2006e 1 are boxed in black).Results of expression array analysis are presented as mean log 2 ratios between normal and tumor cell lines if a significant change was detected (negative log 2 ratios indicate lower expression in tumor cell lines and are boxed in black).Genes that were independently analyzed by MCIp and real-time PCR (Fig.4; Supplementary Fig.S2) are indicated in bold lettering.Confirmed targets of aberrant hypermethylation are indicated in bold italics.Genes that have been shown to be hypermethylated in other types of tumors are in italics.Abbreviations: P, present; A, absent; NC, no change; ND, not detected; NA, not available.Cancer ResearchCancer Res 2006; 66:(12).June 15, 2006