DNA hypomethylation was identified as an epigenetic feature of cancer back in the early 1980’s and later shown to be commonly associated with repeated DNA elements. However the subsequent discovery of DNA hypermethylation of promoters and the resulting silencing of tumour suppressor genes, soon drew attention away from hypomethylation. It is now recognised that hypomethylation more often than not accompanies hypermethylation of the cancer genome. Taking a closer look at methylation states in solid tumours, Rafael Irizarry from Harvard University, USA, Andrew Feinberg from Johns Hopkins University, USA, and colleagues identify large hypomethylated blocks as a universal feature and uncover how hypermethylation fits into this landscape, in a recent Genome Medicine study. Here Irizarry and Feinberg talk us through how their results add to current understanding, with Irizarry tackling the novel statistical approach employed to identify these hypomethylated blocks, and Feinberg addressing the biology behind their findings.
Can you briefly explain what was already known, prior to your current study, about the state of hypomethylation in cancer in solid tumours?
Previously our team of investigators had found that the widespread hypomethylation in cancer that’s been known about for decades involves large but discrete regions we call ‘blocks’, which correspond fairly well in location to another genomic structure we reported in 2009 termed ‘LOCKs’ (an acronym for large regions of heterochromatin characterised by methylated lysines typically at H3K9Me2 or H3K9Me3). The implication is that cancer is a genome-scale epigenetic, i.e. epigenomic, disease involving in large part some degree of relaxation of heterochromatin regions. The consequence of this, we argued, was an increase in stochasticity of methylation and gene expression that could give cancers a selective advantage in a changing environment within the patient.
How does your study add to the story of hypomethylation in cancer, and what were your key findings?
This is the first whole genome study of multiple tumour types, focusing on key causes of cancer mortality including breast, colon, lung, pancreas and thyroid, and also includes early stage neoplasia for most of these tissues. The key findings are: (1) much of the hypermethylation seen in cancer is within hypomethylated blocks, suggesting that instability of DNA methylation through these regions also leads to stochastic increases in DNA methylation in regions (i.e. islands) that can get less methylated than they already are. (2) These hypomethylated blocks and dysregulated methylation occur at the earliest stages of cancer and progress over time. (3) It appears that hypomethylated blocks are a universal defining epigenetic alteration in common solid tumours.
In your study you detected an enrichment of hypermethylated islands in hypomethylated blocks. What does this tell us about the relative roles of hypermethylation and hypomethylation in cancer?
Current conventional wisdom is that the way DNA methylation affects cancer is via local effects on promoters and shores with both hypermethylation and hypomethylation. Our findings show that these local changes are largely part of a bigger picture involving larger regions of methylation instability that correspond to the heterochromatin LOCKs described earlier.
You used a novel statistical approach to identify large blocks of hypomethylation. Can you explain what this entailed, and what advantages it brings over previous methods?
We discovered the hypomethylated blocks using a technology called whole genome bisulfite sequencing on eight samples. The cost of this technology made it unfeasible for an individual lab to use dozens of samples as we needed to do if we wanted to study other cancers. While working on other projects with a much cheaper technology, the Illumina 450K array, we realised that we could detect these blocks if we changed the resolution of the analysis. Specifically, these arrays contain small regions in which one can look for regional changes of about one kilobase in size. We had a very powerful approach, referred to as ‘bump hunting’, for detecting these regions on these arrays. What we realised is that one can group probes into regions, summarise these with one value, and then perform the same analysis at a larger scale (~100 kilobases) by considering the cluster summaries and the measurement units. This gave us an affordable way to search for blocks in many more samples, as we did in this study.
Do you think hypomethylated blocks may also be a universal feature in more rare solid tumours as well as liquid tumours?
We know the first answer is yes, at least for some rare solid tumours, as the primitive neuroectodermal tumours we examined are exactly that. I wouldn’t want to guess about liquid tumours.
What are the potential implications of your findings for diagnosing cancer and tracking cancer progression?
Many biopsies of suspicious lesions are non-determinative. Measurements based on hypomethylated blocks might help to clarify their status, but that would require a clinical trial to determine, of course.
What’s next for your research?
What’s driving the blocks/LOCKs seems important in cancer, and the answer may perhaps lie closer to an epigenetic mechanism than the individual genes, islands, or shores within them. We are also interested in determining which cancer mutations and epigenetic changes are driving these epigenomic structures, and which drugs are actually targeting them.
Genome Medicine 2014, 6:61
Go to article >>