Paula Vertino, Ph.D., associate professor of radiation oncology at Emory University School of Medicine and the Winship Cancer Institute, discussed gene silencing at the meeting. She looked at some of the local features that contribute to which genes or CpG islands are subject to epigenetic silencing and new work that involves a novel epigenetic switch that may predispose certain genes to methylation.
“CpG islands tend to remain unmethylated in normal cells,” Dr. Vertino stated, “so the lack of CpG methylation is thought to be a permissive state for active transcription.”
CpG islands that are actively transcribed are associated with strongly positioned nucleosomes. That means, in each cell, the nucleosomes are in the exact same position, in contrast to other parts of the genome where they tend to be more randomly distributed. CpG islands tend to be marked by active histome marks.
“In cancer cells, there tends to be a global loss of DNA methylation from much of the genome,” added Dr. Vertino. “This is accompanied by loss of the more heterochromatic marks, which may contribute to genome instability through chromosomal aberrations and potential reactivation of transposable elements or certain microRNA.
“At the same time, there is a focal gain of methylation at certain CpG islands,” she explained. “This corresponds with the random positioning of nucleosomes around CpG islands, as well as the loss of active histome marks and the gain of repressive marks.”
In examining some 1,700 CpG islands of the 28,000 that exist, she found that between 3% and 8% of CpG islands may undergo methylation. “The vast majority remain completely unmethylated, even when under great duress.”
In distinguishing between CpG islands that are resistant to DNA methylation from those that are prone to DNA methylation, her lab identified certain local features that may put these islands at risk for permanent silencing associated with DNA methylation.
She induced CpG island methylation in human cells by overexpressing one of the DNA methyltransferases, DNMT1. Using a computational approach developed with collaborator Eva Lee associate professor at Georgia Institute of Technology, she applied a machine-learning approach to discriminate between methylation-prone and -resistant CpG islands, looking at DNA sequence patterns in the areas surrounding the regions that are prone to, and resistant to, DNA methylation. The classifier, she explained, is 85% accurate at predicting the epigenetic status of a given sequence.
When applied across the genome to predict hypermethylation, the genes that were predicted to be prone to aberrant methylation were, in general, much more susceptible to aberrant methylation in lung and breast cancer cell lines than those that were predicted to be methylation resistant.
When they developed this classifier, dubbed PatMan, Dr. Vertino and her colleagues noticed overrepresentation of Hox-related genes within the methylation prone class. “Hox genes tend to be targets for the polycomb repressor complex,” she said, which led to further studies.
Genes that were predicted to be methylation prone also showed an enrichment of H3K27 mer3, the mark imposed by polycomb complex at the boundaries of the CpG island in normal cells. “Genes that were predicted to be methylation-resistant lacked that enrichment.”