“About a decade ago, Rafael Irizarry [Ph.D.] and I developed a genomic array-based method to look in an unbiased way at methylation over a region encompassing a large fraction of the methylome,” says Andrew P. Feinberg, M.D., professor of molecular medicine at Johns Hopkins University School of Medicine.
Taking this approach, known as comprehensive high-throughput array-based relative methylation, the researchers revealed that most tissue-specific DNA methylation differences, which historically were thought to be confined at CpG islands, are in fact situated at regions within approximately two kilobases from their boundaries, in the so-called CpG island shores. Whole-genome bisulfide sequencing showed that the sharp demarcation between the high methylation found at CpG islands and the lower methylation found at CpG island shores is lost in cancer, and the shift of the boundaries toward the CpG islands or away from them results in aberrant methylation patterns.
An additional stride toward understanding genome-wide DNA methylation in disease came about when, in a comparison between colorectal cancer and normal colorectal mucosa from the same patients, the researchers identified large chromosomal blocks of DNA, extending over approximately half the genome, that become hypomethylated in cancer. Largely corresponding to structures known as LOCKs (large organized chromatin lysine modification regions), these regions also show an approximately 80% overlap with lamin-associated domains, which were found microscopically and are thought to be associated with the nuclear membrane. These genomic regions, which correspond to heterochromatin and are highly methylated in normal cells, increase in size during differentiation and are lost in cancer.
Collectively, these findings unveil a much more complex set of epigenetic perturbations characterizing the malignant state than previously thought. “A general dysregulation of DNA methylation occurs in cancer,” Dr. Feinberg explains.
These structural insights into genome-wide DNA methylation led to a key observation. “We found an enormous overlap in the regions that are altered in several types of cancer and in those that are changed during stem cell reprogramming, and both of them are very much related to regions where DNA methylation normally changes in a tissue-specific manner,” he says.
This pointed the researchers toward the possibility that, epigenetically, cancer cells may acquire methylation patterns similar to the ones that are found in other tissue types. By using experimental and modeling approaches to integrate these findings, Dr. Feinberg and his colleagues revealed that the defining epigenetic trait in cancer is not so much the nature of the DNA methylation changes, as their stochasticity.
“There is some type of randomization of the methylation pattern, and as a result a tumor might be much better defined by its departure from the normal epigenetic pattern than by a switch to a different pattern,” Dr. Feinberg says. This is reminiscent of the stochasticity that can be observed during normal development, where highly variable DNA methylation patterns are reported even in genetically identical animals. “We think that the same is true in cancer, and the opening of this stochasticity allows cancer cells to be selected and to gain a growth advantage at the expense of the host,” he adds.