Nonlinear Analysis for a Better Epigenetic Clock

A team publishing in Human Genomics has developed a new model for analyzing epigenetic changes that uses nonlinear analysis.

Current clocks are linear

Epigenetic Clocks - What are they?

An epigenetic clock is a biochemical test that uses DNA methylation levels and accumulation of methyl groups on DNA to determine biological age. There are many different kinds of clocks, and these clocks are often considered to be the gold standard in aging biomarkers. Researchers are currently examining them as potential tools for uncovering the underlying mechanisms of aging, as biomarkers for anti-aging interventions, and as predictors of patient risk for morbidity and mortality.
Read More

Current epigenetic clocks, including the original Horvath clock and the death-predicting GrimAge, use linear analyses of methylation sites to estimate biological age. However, actual epigenetic aging, much like the rest of aging, does not necessarily occur linearly; some research has found that a power law is more appropriate for some methylation sites [1].

However, the actual relationship between nonlinear changes in epigenetics and aging has not been completely discovered, particularly since each epigenetic site might have a different relationship. To accomplish this, these researchers turned to functional data analysis, a statistical technique that attempts to determine where actual data can best be placed in the limitless possibilities of mathematical functions [2]. This new method is called Data-driven Identification and Classification of Nonlinear Aging Patterns (DICNAP).

New step-by-step work on established data

This analysis was performed on four datasets that originated from the National Center for Biotechnology Information (NCBI), each of which contained between 300 and 400 people. Half of these datasets contained only men, and half contained only women, as epigenetic aging have been shown to be sex-specific [3]. A female dataset with 388 people was the largest, so it was used as the main one; the other three were used to confirm the stability of this work.

The first step in this process was to find which methylation sites had any correlation at all using a technique built around the maximal information coefficient, which allows for analysis of both linear and nonlinear relationships [4]. Only the strongest correlations were included in further analysis.

These correlations were further broken down into linear and nonlinear relationships, which were then clustered into groups based on principal components (PCs): the methylation areas that are the most predictive. PC-based versions of current clocks are already in use [5].

The researchers were able to fit linear correlations to many, but not all, of the epigenetic sites they examined. Genes related to neurology, development, and membrane transport seemed to have the strongest nonlinear correlations. However, the effects of each of these individual genes were too weak to be singled out and linked to specific biological processes.

Potential, and contradictory, reasons for nonlinear changes in methylation are mentioned. The rate of methylation changes might accelerate with aging as mechanisms to prevent it deteriorate. However, the rate of cellular division declines in advanced age [6], and division is when many methylation changes occur.

Conclusion

As the researchers acknowledge, this proof-of-concept study used cross-sectional instead of longitudinal data, so it couldn’t identify any single person’s epigenetic changes over time, and individuals vary measurably in their epigenetic aging. To account for individual, population, and regional variation, much larger datasets might be required to get a clearer and more complete picture.

With those hurdles in mind, however, it is clear that a more detailed clock that takes nonlinear variations into account may be substantially stronger than current age-predicting and pace-of-aging clocks. Further work will be required to put DICNAP-based clocks into practical research.

Literature

[1] Vershinina, O., Bacalini, M. G., Zaikin, A., Franceschi, C., & Ivanchenko, M. (2021). Disentangling age-dependent DNA methylation: deterministic, stochastic, and nonlinear. Scientific reports, 11(1), 9201.

[2] Sørensen, H., Goldsmith, J., & Sangalli, L. M. (2013). An introduction with medical applications to functional data analysis. Statistics in medicine, 32(30), 5222-5240.

[3] Yusipov, I., Bacalini, M. G., Kalyakulina, A., Krivonosov, M., Pirazzini, C., Gensous, N., … & Franceschi, C. (2020). Age-related DNA methylation changes are sex-specific: a comprehensive assessment. Aging (Albany NY), 12(23), 24057.

[4] Reshef, D. N., Reshef, Y. A., Finucane, H. K., Grossman, S. R., McVean, G., Turnbaugh, P. J., … & Sabeti, P. C. (2011). Detecting novel associations in large data sets. science, 334(6062), 1518-1524.

[5] Reed, R. G., Carroll, J. E., Marsland, A. L., & Manuck, S. B. (2022). DNA methylation-based measures of biological aging and cognitive decline over 16-years: preliminary longitudinal findings in midlife. Aging (Albany NY), 14(23), 9423.

[6] Tomasetti, C., Poling, J., Roberts, N. J., London Jr, N. R., Pittman, M. E., Haffner, M. C., … & Vogelstein, B. (2019). Cell division rates decrease with age, providing a potential explanation for the age-dependent deceleration in cancer incidence. Proceedings of the National Academy of Sciences, 116(41), 20482-20488.