You have a molecular clock ticking inside of you and, if you read it properly, it can predict how much longer you will live. Want to know how to read it? Well, grab your DNA methylation profiler of choice, measure the percent of your genome that is methylated at these 353 CpG sites, and then just go use this online calculator*. Go ahead. Calculate it. I’ll wait…
Then you’ll need to do what Riccardo Marioni and colleagues did in their recently published paper. They measured genome-wide** DNA methlyation in over 4,500 people, plugged it into two different “DNA methylation age” equations, and then calculated Δage – the difference between an individual’s DNA methylation age and his/her chronological age (i.e., how old he/she was).
the difference between methylation-predicted age and chronological age (i.e., ∆age) was put forth as an index of disproportionate ‘biological’ ageing and was hypothesized to be associated with risk for age-related diseases and mortality
So the authors decided to model how this variable*, Δage, predicted all-cause mortality. They found that a 5-year increase in Δage came with a 21% higher mortality risk. When they controlled for a bunch of other variables that influence mortality – smoking, hypertension, diabetes, to name a few – they found that a 5-year increase in Δage conferred a 16% higher mortality risk.
So the very dark takeaway is this: the older your methylome is relatively to your chronological age, the more likely you are to die. Even worse is the fact that they say there is a strong genetic component to Δage as it is ~40% heritable.
Here is one other thing to chew on from the study:
Control for cell-type composition. DNA methylation patterns (and pretty much any other gene regulatory measure) are known to have strong cell-type specific signatures. And since the authors used data collected from blood, a heterogenous cell population, this means that it is possible that the authors measure of DNA methylation age is just a proxy for changes in the proportions of white blood cells, which are already known to change with age. In other words, their complicated DNA methylation age could just as easily be measured by counting the proportion of different blood cells – no need for the Illumina chips. They attempt to address this issue by adjusting for the proportions of some of the blood cells (basophils, eosinophils, monocytes, lymphocytes, and neutrophils), which they say resulted in “mostly minor difference in the results”.
They even went a bit further and specifically looked the correlation between the residuals of a model of DNA methylation age on chronological age (instead of Δage) and the proportion of naive T cells – which are known to decline with age. They found a weak relationship between residuals and the abundance of naive T cells, but even after adjusting for this (who knows how they adjusted for this) Δage “was still significantly associated with mortality”. The real kicker, however, is buried deep in the methods: They estimated the the abundance of naive T cells using a set of CpGs that were published in another paper as predictors in a regression (and we all know regression outputs come with their fair share of error). So, they used methylation patterns to predict naive T cell abundance, which they then say didn’t influence their specific measures of DNA methylation age – it seems a bit circular. It would be interesting to see if some of the markers that they used to estimate naive T cell abundance also overlap with the 353 and 71 CpG sites that they used to estimate DNA methylation age.
Marioni RE, Shah S, McRae AF, Chen BH, Colicino E, Harris SE, Gibson J, Henders AK, Redmond P, Cox SR, Pattie A, Corley J, Murphy L, Martin NG, Montgomery GW, Feinberg AP, Fallin M, Multhaup ML, Jaffe AE, Joehanes R, Schwartz J, Just AC, Lunetta KL, Murabito JM, Starr JM, Horvath S, Baccarelli AA, Levy D, Visscher PM, Wray NR & Deary IJ (2015) DNA methylation age of blood predicts all-cause mortality in later life. Genome Biol. 16, 25. doi:10.1186/s13059-015-0584-6
*This measure, DNA methylation age, is pretty good at estimating your age if you are older, but pretty poor when you’re younger. Interestingly, you can take a look at figure 1 from the paper and see that one of the authors’ methods did a pretty poor job of predicting chronological age in the two study cohorts that had very little age variation (LBC1921 and LBC 1936).
**as genome-wide as the Illumina 450K chip gets. It covers a 450K sites, but that stills leave many CpG sites unmeasured.