Dr. Andrew Feinberg discusses how epigenetics might be related to missing heritability of disease via variable methylation regions. This interview was shot on campus at Johns Hopkins University.
Variable Methylation Regions and Missing Heritability
So a conundrum in modern human genetics is how epigenetics might be related to missing heritability of disease, what’s commonly called missing heritability of disease. The problem with this area is that epigenetic changes are by definition reversible, and they’re also powerfully reprogrammed during germ line transmission, during the development of germ cells and then reproduction. So it’s hard to understand how you might get an epigenetic change that might be not only stably maintained and transmitted from generation to generation, but also undergo Darwinian selection. There has to be a selective advantage for any modification to be maintained and then to be represented at significant frequency in a population and that’s a precondition really for it to be involved in some phenotyping, including a disease phenotype. So how does one reconcile all that?
Stochastic Variation in Sequences
So a couple of years ago I actually got this idea that maybe there are genetic sequences that themselves underlie stochastic variability that might be important in normal development, but also might be important in evolutionary bet hedging, so that you might have a phenotype that’s defined by genetic sequence, but then you also might have sequences that themselves increase the variance for that particular trait within a population and just to sort of be practical about that, something like the CpG island shores might do something like that where you have a sequence of intermediate CpG density that might, depending upon how CpG dense it is, might be more or less prone to getting methylated and thereby showing stochastic variation in sequence.
“…tumors seemed to be distinguished by an anti-profile, like a departure from a normal pattern you would expect in that tissue, more so than a specific shift…”
Variably Methylated Regions
But it also might have something to do with this missing heritability idea because you could get perfectly normal Darwinian selection of genetic sequences that, based upon that sequence, confers for that particular region say a great propensity towards epigenetic variation and consequently a phenotypic variation with a given population. So my colleague in this study and most of what we’ve been doing recently, Rafaela Rosaria, a brilliant biostatistician and I, interrogated the data set we already had to see if there are sequences across the genome that even in inbred mice, that have an essentially identical genome that were raised in the same cage and even came from the same litter, whether there are regions that show a high degree of variation in DNA methylation and we found hundreds of these. We refer to them as variably methylated regions, but we mean by that they are not variable across tissues, but they are variable within the tissue across individuals even with the same genotype. And those sequences, those genes that were involved were incredibly important. If you do a GO annotation analysis, you find that they’re involved in development of most normal organ systems and embryonic pattern formation. And so it seems that those sequences exist and they could serve a role in normal variability within a given population.
Variably Methylated Regions in Cancers
That idea of variability that might be normal developmentally also had a lot to do with why we did those experiments that we reported last summer in Nature Genetics in the first place. So we wondered whether or not the sequences that seemed to show a marked difference in DNA methylation between cancers and normals also are highly variable when you compare say a group of colon cancers to their matched normal tissue. And what we found is that the variability in DNA methylation across a series of cancers for each of these sites, for most of them, was much greater than the variability across the matching normal mucosa for those same individuals and that this variability index was in fact more strikingly different in cancer than more mean differences. Another way of saying this is that the tumors seemed to be distinguished by an anti-profile, like a departure from a normal pattern you would expect in that tissue, more so than a specific shift or a particular cancer-specific profile and not only that; the sites of greatest variability in colon cancer when you compare across a group of colon cancers was also, those very same sites, were also the sites of greatest variability in ovarian cancer and breast cancer and lung cancer.
And so we think that even though cancer involves clearly mutations that are specific for individual malignancies, it seems like, at least for solid tumors, a common feature across most of the common ones is that there are sites that are highly variable and depart from a normal mean pattern of expression, but in a random stochastic way that seem to distinguish this whole class of tumors from each other. And if you take the most hyper variable sites in cancers, even if you pick something like colon cancer, those sites are completely able to distinguish the normal colon from the normal breast from the normal kidney and so forth. So it’s like there are sites that are very important in their variability of methylation per say that are critical in the normal development of different tissue types and that highly variable process becomes reactivated in malignancies and maybe that’s largely responsible for why cancers start to accrue properties that you would not expect in the tissue they’re supposed to be, but more resemble an amalgam of other tissue types. That might help to explain also why they invade and why they settle in other tissues and so forth.