Comprehensive validation of genome-wide CpG island methylation profiles

Data from this section have been published in the journal Cancer Research. Comparative MCIp hybridization data were deposited with the GEO Data Library under Series Entry:

GSE17455, GSE17510, GSE17512.

To establish and test the newly adapted and improved MCIp-on-chip technique, the first genome-wide methylation analyses using Agilent 244K CpG island microarrays were performed with MCIp-enriched methylated fragments of the well-established leukemia cell lines U937 and THP-1 in comparison with enriched fragments of blood monocytes of a healthy donor.

Because the signal intensities were biased in correlation to their GC content (higher GC content lowered the average signals), the probe signals were GC normalized. Three independent replicates of each cell line were highly similar (mean r²=0.79 and 0.87 for log10

ratios of THP-1- and U937-monocyte comparisons, respectively). A typical scatter plot (Figure 5-17) of a comparative hybridization of MCIp-enriched material from U937 cells highlights the three types of hybridization behavior: probes that show low signal intensities in both samples (absence of DNA methylation), probes indicating specific enrichment (aberrant DNA methylation) in the leukemia samples, and probes that show high signal intensities but low signal ratios in both samples (methylated in both samples). Altogether, more than one third of all microarray probes showed significant enrichment in the U937 cell line.

Figure 5-17 Comparative DNA methylation analysis of U937 cells and normal human blood monocytes using methyl-CpG immunoprecipitation (MCIp)

Representative scatter plot of a comparison of MCIp-enriched material from 2 µg genomic DNA of a leukemia cell line (U937) and a pool of normal human blood monocytes from healthy donors on human 244K CpG island arrays. Signal intensity ratios are plotted against average signals (MvA plot, log10 scale). More than one third of all microarray probes show significant enrichment in the cell line.

All CpG islands that were validated as hypermethylated in the first study (with 12K CpG island arrays) were again detected as hypermethylated in these experiments (performed with 244K Agilent arrays). In total, approximately 11,300 or 8,700 (out of 23,000) independent regions were significantly enriched or depleted (>2.5-fold different) in U937 or THP-1, respectively. The majority of differentially enriched regions showed signs of hypermethylation (or amplification) in both cell lines. In U937 cells 10,700, in THP-1 cells 6,800 differential methylated regions (DMR) were detected. Hypomethylation or deletion of individual regions were mainly found in THP-1 cells.

Large-scale validation of MCIp microarray data using mass spectrometry analysis was performed to quantify methylation differences at the resolution of single CpGs. Therefore a representative set of differential methylated regions (DMRs) as well as regions without methylation difference between cell lines and monocytes as a negative control were chosen to be analyzed by the MassARRAY system. This method is based on Matrix-Assisted Laser Desorption/Ionization Time-Of-Flight Mass Spectrometry (MALDI-TOF MS) measurement of bisulfite converted DNA using the EPITYPER platform (Sequenom, San Diego, US). Bisulfite treatment generates methylation dependent sequence variations, which can be measured by the MassARRAY system. Moreover, this procedure allows the analysis of multiple CpGs in one amplicon and the comparison of their methylation status between different samples (for

more details see section 4.4.6). The validation panel comprised a set of 140 genes that were selected based on the comparative MCIp methylation profiles of 23,000 CpG islands from the two myeloid leukemia cell lines (U937 and THP-1). In total, 1,150 primer pairs, which cover about 13,500 CpG sites, were designed for the amplification of bisulfite-treated DNA.

Besides THP-1 and U937, MALDI-TOF MS was also performed with DNA derived from monocytes of three different healthy donors and with unmethylated and fully methylated DNAs as controls. The complete MALDI-TOF MS data are provided in the supplementary part of the corresponding publication (Gebhard et al., 2010).

Validation using mass spectrometry analysis of bisulfite-treated DNA (MassARRAY System Sequenom) was highly consistent with the microarray data. Figure 5-18 shows examples of microarray-MassARRAY comparisons for four genes (MLL, SMAD6, HOXB5 and EPAS1) which demonstrate a high degree of correlation between both approaches.

Figure 5-18 Examples for correlation between MCIp and bisulfite data

The panel on top shows microarray results (log10 ratios; blue: U937 versus monocytes; red: THP-1 versus monocytes) for selected CpG island regions (MLL, Chr 8: 9798000-97992000; SMAD6, Chr 15:

64781200-64783000; HOXB5, Chr 17: 44025400-44026600; EPAS1, Chr 2: 46378600-46380800). The middle panel depicts the GC-content of microarray probes. The bottom panel shows quantitative MALDI-TOF MS (EpiTYPER) methylation levels for individual CpG residues within the analyzed region. Each spot represents one CpG unit.

In a next step all results extracted from both data sets were compared. To correlate differential signal intensities on CpG island microarrays with data points for individual CpG dinucleotides, the mean methylation difference between tumor cell lines and monocytes of all measured CpG dinucleotides that are located in a 300 bp radius around a microarray probe was calculated (“EpiTYPER methylation ratio”, for further information see section 4.4.6.10).

Figure 5-19 illustrates the high consistency of both approaches regarding the comparison of the two leukemia cell lines THP-1 and U937.

When plotting the “methylation ratio” against the probes’ signal ratio, a good correlation (r² = 0.51 for U937 and r² = 0.67 for THP-1) was observed between microarray probe intensity ratios on microarrays and mean bisulfite methylation ratios of CpG dinucleotides located around the microarray probe for the U937 as well as the THP-1 cell line. The high reproducibility of both approaches is shown exemplary for U937 leukemia cells and normal monocytes in Figure 5-20A.

Figure 5-19 Correlation of microarray and mass spectrometry data

Differential methylation between the two cell lines THP-1 and U937 can be reliably detected using both methods, MCIp combined to microarray analysis and MALDI-TOF MS. Differential methylation of the microarray data is illustrated by plotting the signal ratios from one cell line against the signal ratios from the other cell line. For the EpiTYPER data methylation ratios for both cell lines are calculated and plotted against each other. In the lower graph, the high consistency between microarray signal ratios and EpiTYPER methylation ratios is illustrated on probe level.

A similar comparison was performed to correlate both data sets based on region level instead of probe level. Here, mean methylation ratios (log10) for 225 regions (covering more than 300 bp each) (see section 5.4.2) based on the above amplicons were calculated and then plotted against mean signal ratios of all microarray probes within each region. As shown in Figure 5-20C and D, a good correlation between microarray and EpiTYPER data was observed (r² = 0.62 for U937 and r² = 0.75 for THP-1).

The correlation between both methods increased on the region level compared to the probe level, mainly because the resolution of the microarray approach dropped at extremely GC-rich microarray probes which tended to cross-hybridize, even under stringent hybridization conditions. Mass spectrometry data was also able to provide higher resolution

of the boundaries between methylated and unmethylated domains. Stretches of unmethylated DNA in close vicinity to metylated domains were often detected as methylated due to the DNA fragmentation range. Nevertheless, the MCIp approach clearly discriminated methylation levels between the two leukemia cell lines U937 and THP-1 (Figure 5-20B and Figure 5-19). Thus, our comprehensive validation demonstrates a good overlap of MCIp and MALDI-TOF MS data and suggests that our technique allows for reproducible and valid detection of comparative CpG methylation levels.

Figure 5-20 Methyl-CpG immunoprecipitation and its validation using MALDI-TOF MS

(A) Microarray probe-based correlation of MALDI-TOF MS (EpiTYPER) and MCIp microarray results of U937/monocyte comparisons. Microarray probe signal log10 ratios were plotted against an EpiTYPER score that consists of a scaled, average methylation level of all CpGs located in a radius of 300 bp around the microarray probe (r² = 0.51 for U937 and r² = 0.67 for THP-1, data not shown). (B) Microarray probe-based correlation of differential CpG methylation ratios measured by MALDI-TOF MS (EpiTYPER) and MCIp microarray. MCIp reliably detects differential methylation between the two cell lines (r² = 0.65). Correlation of MALDI-TOF MS (EpiTYPER) and MCIp microarray results of U937/monocyte (C) and THP-1/monocyte (D) comparisons for regions covered by the EpiTYPER analysis. Mean probe signal log10 ratios are plotted against mean log10 transformed EpiTYPER methylation ratios (r² = 0.62 for U937 and r² = 0.75 for THP-1).

5.3.3 Genome-wide hypermethylation profiling in AML and patients