A new timepiece: an epigenetic mitotic clock

In this work, Yang and colleagues [4] select specific Polycomb target loci that are both unmethylated in multiple fetal tissues and show age-associated hypermethylation and hypothesize that methylation at these sites reflects relative mitotic activity. They then construct a model that shows that cancer and pre-cancer tissues have increased DNA methylation relative to relevant normal tissues. This, they posit, reflects enhanced stem cell activity and increased cancer risk.

Yang and colleagues [4] are to be congratulated for combining cutting-edge biological knowledge with state-of-the-art bioinformatics in building a cancer prediction model. Scrutiny of this provocative model is certain to result in modifications and refinements to it as the underlying assumptions (of both the model and past experiments) are challenged and the understanding of the underlying biology improves. At the outset, we note that there are a few important assumptions and limitations in this work.

First, the stem cell division rates applied in this work are derived from those presented in Tomasetti and Vogelstein [6]. While this is reasonable, as Tomasetti and Vogelstein indicate in their work [6] there is room for improvement in the estimates they present. In addition, the current model is tested in cancer tissues and shows universal increases, but the authors have not yet shown evidence of prediction of risk in a prospective setting, where availability of data is still very limited. The current model does not include or apply estimates of the contribution that somatic alterations in non-stem cells (in any tissue or tissue environment) may make to tick rate. Altered somatic cells, particularly in tissues with higher levels of carcinogen exposure, may have non-stem cells that propagate alterations and increase the estimated tick rate of the mitotic clock. We also note here that the potential contribution of immunity and inflammation, which are particularly important in many solid tumors, is not yet specifically included in the model. The current model is also built, appropriately, using data from just one tissue source, and additional methylation data from normal tissues in healthy subjects are needed to expand and further examine the predictions of the model. Finally, reference-free and reference-based approaches would have to be properly applied to adjust for cellular heterogeneity in the setting of various other normal tissue types.

We also highlight that, as the authors note, this model necessarily assumes that methylation at the informative loci occurs only in stem cells. This is novel biology for which there is little to no experimental evidence. If true, it would imply that locus specificity in methylation is differentially determined in numerous distinct cellular and tissue-specific compartments.

An important aspect of the work presented by Yang and colleagues [4] is that it highlights the differences between genome-wide association studies (GWAS) and epigenome-wide association studies (EWAS). In their work on this issue, the authors define additional building blocks of the DNA methylome, adding loci that putatively act in coordination and display a novel “mitotic clock” phenotype. These loci, as well as others that have been previously defined (e.g., the Horvath “aging” loci), represent the beginning of our ability to organize the methylome into distinct loci-driven phenotypic units [7]. Importantly, future discovery-based interrogation seeking to define the nature of differences in tissues or of tissues within population groups can now begin to group loci for testing rather than treating them independently. These groups can be compared for differences in the locus-associated phenotype, offering potential for better interpretation of some of the results of these studies. This is not unlike applying the now-standard techniques for delineating cell types within tissues [8–10], which represented the initial attempts to arrange epigenetic data into organizational, phenotypically defined units.

Adding these new building blocks allows us to imagine new approaches to future studies, including examination of the dynamic changes to the epigenome over the life course. For example, does epiTOC provide a window into variation in the extent to which chronological age contributes to cancer risk? The ability to estimate mitotic tick-rate acceleration has value for researchers collecting DNA methylation data, as it provides a directed approach to investigating age-correlated cancer risk and may inform on the biology of cancer risk factors studied across the life course.