Entropy is one of those slippery concepts wherein the same word has been adopted by different scientific disciplines to mean subtly different things. I'd recommend a recent article that attempts to explain for the layperson how these this different meanings arose, and that they overlap at the concept of measuring our ignorance of the state of a system, our inability to predict the state of that system. Here we'll talk about entropy as a measure of the randomness of a distribution; the more random the distribution, the less our ability to predict its specifics. The distribution of interest for today is the methylation state (methylated or not methylated) at one or more CpG sites on the genome, across many genome copies in many cells.
DNA structure determines whether or not a given gene sequence is exposed to transcription machinery and RNA is produced. One of the mechanisms determining the shape of DNA is whether or not methyl groups are added at specific locations called CpG sites, named because a a cytosine nucleotide © is followed by a guanine nucleotide (G) with the two linked by a phosphate group (p). This DNA methylation is the basis for epigenetic clocks that assess chronological and biological age, because the methylation status of some CpG sites is characteristic of the damage and dysfunction of aging. While whether or not a CpG site is methylated is a binary outcome, this data is measured across the many, many cells and genomes in a given blood or tissue sample. Current epigenetic clocks take the average of all of those 1s and 0s as the input of that specific CpG site to the clock algorithm.
In today's open access paper, researchers start instead by considering the entropy of the distribution of methylation status at a CpG site across the many measured genomes. For this purpose, entropy is a measure of how noisy or random the data is. The researchers then show that one can construct an epigenetic clock from the entropy values per CpG site that performs as well as clocks built using average values of methylation state. This suggests that aging is not just resulting in a move of some CpG sites towards one status, but also an increase in noise in DNA methylation, a move in both directions, an increase in randomness. Age-related noise in gene transcription is already a topic for discussion in the field, so why not age-related epigenetic noise as well?
DNA methylation entropy is a biomarker for aging
To measure age associated changes in DNA methylation, we collected buccal swabs from 100 individuals ranging from 7.2 to 84 years old. The DNA methylation profiles were generated using targeted bisulfite sequencing. Our target panel contained approximately 3000 regions that were selected to cover age associated CpG sites that were identified in multiple epigenetic clocks. Each probe is 120 base pairs, and therefore captures a region of DNA that is slightly larger than the probe length. We obtained an average coverage of 293 reads per sample across these regions.
We first calculated the mean methylation of each CpG site in each of the 3000 loci across the 100 samples, and then averaged these levels over a region. We also computed the Cellular Heterogeneity-Adjusted cLonal Methylation (CHALM). This approach computes the read level methylation of a region after reads are dichotomized into methylated or unmethylated based on the presence of one or more methylcytosines. We also computed the methylation entropy for each locus using four CpG sites within each region, using the Shannon entropy formula. With four CpG sites, there are 16 possible methylation states, and we computed the probability of each state as well as the entropy of the four CpG sites.
We next generated scatter plots that compared the values of the three metrics across loci. Age-related changes in mean methylation and CHALM were strongly correlated. By contrast, the scatter plots of entropy versus mean methylation or CHALM resulted in more complex patterns with both positive and negative trends. This demonstrates that methylation entropy is measuring different properties of a locus compared to mean methylation and CHALM, and that loci can become both more or less disordered with age, independently of whether the methylation is increasing or decreasing with age.
We next asked whether we could compare the use of these three metrics to construct epigenetic clocks that predict the age of each individual. Selecting only four CpG sites per region to calculate entropy was sufficient to achieve chronological age estimates that were correlated with the actual age. The mean average error was 5.199 years, which was lower than the other mean-based methods that incorporated many more CpG sites. This suggests that the entropy of a locus is potentially a more useful biomarker of aging than the methylation level of individual sites. Though the 3000 loci analyzed may or may not be representative of the whole genome, this suggests that the entropy of an organism's methylation profile is informative of its epigenetic age.
View the full article at FightAging
