A holy grail of aging research is the ability to measure a biomarker in an individual animal, and then use the result of that measurement to calculate how old that animal is—without knowing its chronological age. Such a biomarker of aging could be used to speed up the testing of potentially lifespan-extending intervations without having to wait an entire lifespan to obtain a result.
Many such biomarkers have been proposed, including decreased telomere length, mutation accumulation, and changes in gene expression. (Recently, we reported an effort to predict age by data mining electronic medical records).
To date, the most accurate ‘aging clocks’ are based on alterations in DNA methylation, an epigenetic mark on cytosine (‘C’) nucleotide bases whose levels change in systematic and site-specific ways over the course of the lifespan. Specifically, the genome-wide level of DNA methylation, which is generally associated with repression of gene transcription, tend to decreases as we age. In modern labs, methylation levels at specific sites throughout the genome are detected using a method called bisulfite sequencing, a technique involving a chemical treatment that enables researchers to distinguish methylated from unmethylated Cs.
Several ‘aging clocks’ based on DNA methylation have been reported in the literature. However, the accuracy of these methods is relatively low, especially for clocks designed for mice. This is problematic because mice, for better or worse, are the workhorse model organism for aging research, and it is of critical importance that the field ultimately come up with a way to measure the rate of mouse aging in real time rather than trying to extract useful information about the rate of aging from the distribution of times that individual mice within experimental populations decide to keel over. (I’ve said it before and will say it again: even when one is studying lifespan, lifespan itself is a lousy endpoint, ) Worse, because most of the existing clocks were calibrated on rather young mice, their accuracy tends to get worse as the animals age, making them less useful for monitoring differences in the rate of aging later in life.
The new clock uses DNA methylation patterns to predict age, throughout the lifespan, significantly more accurately than any previously published model.
Moreover, previous clocks tended to be trained on single tissues, limiting their applicability to studies of other parts of the body and (when the tissue in question is something other than “blood”) locking researchers into invasive sample extraction procedures that are difficult or impossible to perform multiple times on the same animal, posing an obstacle to longitudinal studies.
To overcome these challenges, a team at the Division of Genetics at Brigham & Women’s Hospital has designed a new methylation clock for mice that purports to solve many of these problems. The approach was fairly simple: Because the publicly available data used to train previous clocks were biased toward young animals, the authors supplemented those data with their own bisulfite sequencing data from mice of many ages. Similarly, to overcome the tendency of previous efforts to focus on single tissues, the authors collected their data from several organs. They then identified C nucleotides in the genome whose methylation patterns varied systematically with age, and built a model that took as input the methylation frequencies at each of those positions, and gave as output a prediction about the mouse’s age.
The resultant clock (which the authors called WLMT for “whole lifespan, multi-tissue”) could predict age from methylation patterns throughout the lifespan, in both old and young mice, significantly more accurately than any of three previously published models.
Importantly, WLMT could also detect the anti-aging effects of lifespan-extending interventions, both genetic and lifestyle-related. For example, WLMT predicted that growth hormone receptor knockout (GHR-KO) mice, which are smaller than their wild-type counterparts but have significantly longer lifespans, to be younger than their chronological ages. This is consistent with the idea that these mutants live longer because they age more slowly. Likewise, calorie-restricted mice of the commonly used C57BL/7 strain seemed younger than free-feeding animals. Interestingly, the largest effect was seen in induced pluripotent stem cells derived from fibroblasts, in which predicted age was dramatically lower than the source cells—implying that the return to pluripotency “resets” the aging clock, perhaps explaining why the germ line does not age.
The value of this feature is clear: A biomarker-based model that is both good at predicting chronological age in wild-type or untreated mice, but systematically makes lower predictions in mutant or experimentally manipulated animals with longer lifespans, is one that is capable of detecting interventions that slow the rate of aging. This ability would dramatically accelerate research aimed at identifying new lifespan enhancers, reducing the time required to obtain data from the entire lifespan of a mouse (years) to the temporal resolution of the measurement (i.e., the minimum amount of time that must pass for the biomarker to change significantly).