Scientists have devised a transcriptome-based aging clock that allows for precise measurement of biological age in C. elegans worms and shows promising results in humans as well [1].
The purpose of aging clocks
One of the central questions of longevity research is how to measure biological age. If a pathology, genetic difference, lifestyle choice, or treatment either shortens or prolongs life, it affects your biological age. It’s crucial to be able to accurately measure this because human lifespan is so long: if we couldn’t measure alterations in lifespan without waiting for people to die, human trials of anti-aging interventions would take decades.
Theoretically, an aging clock could use any set of parameters that correlate with aging; it just has to correctly predict the effect of lifespan-altering events. Recently, methylation clocks have been steadily gaining in popularity, as DNA methylation controls gene expression by changing the shape of our chromatin: a methylated site (one that has a methyl group added to it) coils and stops being transcribed. Although the expression of certain genes can go up or down as we age, methylation levels generally decline with time [2], which means that more transcription is going on, including harmful transcription of sites that are normally silenced, such as retrotransposons [3].
Introducing binarization
Methylation is not the only factor that affects transcription; if transcription is what matters, an aging clock based on transcription levels rather than simple methylation is theoretically superior. Also, not all model organisms in longevity research modulate their transcription the way we do. C. elegans nematode worms are an extremely popular subject of aging research, for various reasons, but they mostly lack DNA methylation [4].
In this new paper, the researchers describe building an aging clock based on a set of 576 genes whose expression correlates with aging in C. elegans. After determining the set of genes, an AI model was trained on 900 worm transcriptomes obtained during various other studies, and then they tested it on another 100.
This was not the first attempt to create a transcriptomic clock for C. elegans, but this time, the researchers introduced an alteration that proved crucial: they binarized the levels of transcription. In this process, they assigned only one of two possible values to the transcription level of each gene, depending on whether it is above or below average. By doing so, the researchers were able to discard a lot of noise, and they claim that the resulting binary transcriptomic (BiT) model is more robust and approaches the theoretical maximum accuracy for C. elegans lifespan.
High prediction power
The researchers were able to correctly predict the lifespan-altering effects of certain genetic variations, environmental factors, and interventions, including caloric restriction, metformin supplementation, extreme heat, and harmful bacteria. The aging clock was stress-tested in various scenarios, such as the combined effect of more than one drug (some drugs cancel the life-prolonging effect of rapamycin, while others amplify it, and the model successfully predicted effects of such cocktails). The model accounted for such subtle effects as the duration of treatment and the genetic variance of the worms’ E. coli food supply. At times, the model was able to predict the future effect of a factor that had occurred on the first day of the worms’ lives.
The model was also successfully tested on different unrelated datasets with other lifespan-affecting stressors to show that there was no overfitting. Overfitting, one of the fundamental problems in machine learning, is when the model learns the quirks of its training dataset and makes spectacular predictions on similar datasets but fails on all others.
Human results
The researchers then applied the same method to human transcriptomes, using a different set of genes but retaining the principle of binarization. The resulting model shows superior predictive power over similar, non-binarized models – for instance, it successfully predicted the greatly increased biological age of children suffering from progeria, a genetic disease that accelerates aging.
Conclusion
If the results of this study can be successfully reproduced, it may show that the researchers have struck gold with their binarization approach that successfully removes noise from transcriptomic data. Longevity research is in dire need of reliable aging clocks, and this binary aging clock may be an improvement over existing methylation-based clocks.
Literature
[1] Meyer, D. H., & Schumacher, B. (2021). BiT age: A transcriptome‐based aging clock near the theoretical limit of accuracy. Aging Cell, e13320.
[2] Johnson, A. A., Akman, K., Calimport, S. R., Wuttke, D., Stolzing, A., & De Magalhaes, J. P. (2012). The role of DNA methylation in aging, rejuvenation, and age-related disease. Rejuvenation research, 15(5), 483-494.
[3] Mahmood, W., Erichsen, L., Ott, P., Schulz, W. A., Fischer, J. C., Arauzo-Bravo, M. J., … & Santourlidis, S. (2020). Aging-associated distinctive DNA methylation changes of LINE-1 retrotransposons in pure cell-free DNA from human blood. Scientific reports, 10(1), 1-12.
[4] Weinhouse, C., Truong, L., Meyer, J. N., & Allard, P. (2018). Caenorhabditis elegans as an emerging model system in environmental epigenetics. Environmental and molecular mutagenesis, 59(7), 560-575.
