Establishment of a new microarray platform

In our early studies the MCIp technique in combination with human 12K CpG island microarrays (HCGI12K, Microarray Center, UHN, Toronto, Canada) was used to identify more than one hundred genes with aberrantly methylated CpG islands in three myeloid leukemia cell lines. These results showed that the MCIp technique discriminates DNA fragments according to the methylation degree and allows an unbiased genome-wide detection of hypermethylation.

However, the initial experiments performed with the Human CpG 12K microarrays highlighted several issues. Besides quality problems, many other aspects encouraged us to switch to another microarray platform. The previously used 12K microarray platform contained many genomic Mse I fragments with high variation in fragment length. In addition, the array contained repetitive fragments leading to unwanted cross-hybridization events (non-specific binding), which possibly gave rise to misleading results. Furthermore, the number of the representative genes on the array was relatively small. Thus, for global analysis of patient samples, another array platform provided by the company Agilent seemed to be better suited for this purpose. This array contains 244,000 probes (50-60 mer oligonucleotides) covering about 23,000 CpG islands within coding and non-coding regions of the human genome (Agilent 244K CpG island microarrays).

To adapt the fractionation approach to the new Agilent DNA microarrays, several modifications were required. Instead of the previously used Mse I digestion, genomic DNA was sonicated to a mean fragment size of 350-400 bp. Sonication of genomic DNA leads to a statistical fragmentation which is necessary for an unbiased genome-wide methylation profiling. Moreover, large-scale MCIp (4 µg DNA instead of 300 ng) was used in order to provide sufficient amount of gDNA for subsequent labeling and microarray hybridization.

Therefore PCR bias caused by ligation-mediated amplification (LM-PCR) (see section 4.3.6.2) could be avoided. Empirical evidence showed that the coating of the protein A sepharose beads influences the fractionation behavior. In order to compensate for possible variations concerning the coating (due to varying quality of different protein batches) prior to each set of fractionation of the normal and tumor samples, a test MCIp with DNA derived

from the U937 cell line was performed with a part of the freshly coated beads to define the cut-off for highly methylated DNA. The individual MCIp fractions from U937 were spin-purified (PCR purification kit, Qiagen) and eluted in 100 µl EB buffer. Subsequently the fractionation of U937 DNA was controlled by qPCR using control primers covering the imprinted region of SNRPN as well as a genomic region lacking CpGs (Empty 6.2) and the CpG island region RPIB9 (strongly methylated in U937). While both alleles of the imprinted SNRPN are eluted in different fractions (the unmethylated one with a low salt buffer, the methylated one with a high salt concentration), the bulk of the unmethylated CpG empty region (negative control) is enriched in the low salt fractions due to the complete absence of CpGs (data not shown). In contrast, the RPIB9 fragments which are highly methylated in the U937 cell line were detected in the high salt fractions. According to these results a threshold is defined at a salt concentration which separates the strongly methylated DNA fragments from the intermediate and low methylation fragments. Assuming that the other samples were enriched for methylated DNA in the same manner, MCIp was performed with the actual samples (tumor and normal samples) according to the determined cut-off. The high salt fractions containing the highly methylated CpG island fragments were directly labeled for microarray hybridization. Cancer cell DNAs were labeled with Alexa Fluor 647 and DNA from normal cells was labeled with Alexa Fluor 555 using the Bioprime Plus Array CGH Genomic labeling System (Invitrogen, Carlsbad, CA, USA). Efficiency of the labeling reaction was controlled with UV-spectroscopy and comparative hybridization on CpG island oligonucleotide microarray was performed using the recommended protocol (Agilent). Image data was extracted with Agilent feature extraction software and imported to Microsoft Excel for further analysis.

Because the signal intensities were quite low compared to the background noise and the method was not as robust as expected, the application had to be further improved. Different conditions were tested to achieve optimal results. First, after MCIp, each fraction was purified using MinElute Columns to reduce the volumes (elution in 20 µl EB). Therefore loss of DNA by lyophilization could be circumvented. To achieve a better control after MCIp, not only a pretest with U937 DNA was performed, but all individual samples were controlled by qPCR with control primers to determine the cut-off. Moreover to attain improved labeling, a new Kit from Invitrogen was used (Bioprime total Genomic Labeling System). Consequently, enriched methylated DNA fragments of the high salt MCIp fractions were labeled with Alexa Fluor 5-dCTP (cancer cells) and Alexa Fluor 3-dCTP (normal cells).

Since GC-rich probes have the tendency to cross-hybridize, the stringency of hybridization was increased by a combination of a higher incubation temperature (67°C instead of 65°C) and by addition of formamide (15%) to the hybridization reaction mix. Therefore misleading

results such as false positive and false negative signals should be minimized. This step probably had the major impact on better results. Figure 5-14 demonstrates that using stringent hybridization conditions, many more probes could be detected as hypermethylated in the tumor cell line.

Figure 5-14 Comparison of both hybridization protocols

(A) The two diagrams demonstrate the difference between the standard (65°C, without formamide) and the stringent (67°C, 15% formamide) hybridization protocol. The signal ratio between U937 and monocytes is shown as a function of the average signal intensity. (B) The diagrams demonstrate the difference when the signal ratios of both protocols are directly compared showing that with the stringent protocol much more probes are detected as hypermethylated than with the standard hybridization conditions, especially when the probes have a high GC content (˃60%).

To explore if the signal intensities increased with greater quantities of DNA, MCIp and subsequent microarray analysis were performed with 1 µg, 2 µg and 4 µg DNA. Figure 5-15 and Figure 5-16 compare data using different amounts of DNA and different hybridization protocols. Figure 5-15 illustrates a comparison of genome-wide hypermethylation profiles, whereas Figure 5-16 depicts three selected regions (FOXP3, MARVELD2, IRX3). The studies showed that robust methylation profiles could be obtained with as little as 1 µg of genomic DNA using the stringent (new) protocol. Best results were achieved with 4 µg of genomic DNA while 2 µg DNA were sufficient for good, reproducible results.

Because we were limited in patient material, all following experiments were performed with 2 µg DNA instead of 4 µg DNA.

Figure 5-15 Major modifications of the MCIp-on-chip protocol in global screening for tumor-specific hypermethylation

The old hybridization protocol using standard hybridization conditions (65°C and no formamide) was compared to the newer stringent hybridization protocol (67°C and 15% formamide). The new stringent protocol was performed with three different amounts of input DNA (1 µg, 2 µg, 4 µg subjected to MCIp). The signal ratios between tumor and normal DNA were plotted as a function of the average signal intensity.

Figure 5-16 Examples of microarray results using different hybridization conditions and increasing amounts of DNA

Shown are data points for three CpG island regions of FOXP3, MARVELD2 and IRX3 using different hybridization conditions. Each data point represents one microarray probe. The old standard protocol involved hybridization at a temperature of 65°C, whereas in the new stringent protocol the hybridization temperature was increased to 67°C and 15% formamide was added. The log ratio (tumor/normal) is plotted as a function of the relative position on chromosome 1.

5.3.2 Comprehensive validation of genome-wide CpG island