Thread by @AlexandreTrapp

Last week, on the cover of @NatureAging, we published the first computational framework for epigenetic age estimation at single-cell resolution.

A fantastic collaboration with @CsabaKerepesi and @VadimGladyshev in the @gladyshev_lab @harvardmed!

nature.com/articles/s4358…

1/🧵

First off, I want to express my sincere gratitude to the editors @NatureAging for choosing to feature our study on the cover.

This cover art was expertly designed by my dear friend Tiamat Fox! Please DM me for her contact info if you are interested in her services.

2/🧵

The code, data, tutorials, and documentation for the platform are all publicly available under an open-source license on my GitHub page:

github.com/alex-trapp/scA…

3/🧵

Single-cell methylation data is extremely sparse and mostly binary.

This poses profound limitations for the application of conventional bulk epigenetic clock methods to partial single-cell methylomes.

4/🧵

To address these fundamental constraints, I developed a multi-step framework, called scAge, that enables profiling the epigenetic age of individual cell lineages.

5/🧵

To begin, we trained CpG-specific linear regression models using bulk data that map how methylation in tissues changes as a function of age.

This enabled us to generate reference sets summarizing tissue-specific methylation dynamics with chronological age.

6/🧵

Next, the framework intersects a sparse single-cell profile with the reference, and selects only CpGs that are common between this reference and a given methylome.

Subsequently, these CpGs are ranked and filtered based on their linear association with chronological age.

7/🧵

The crux of the algorithm rests on the following fundamental intuition: as methylation at a CpG locus in a tissue increases with age, more individual cells in that tissue will be methylated at that locus in older samples (and vice versa if methylation decreases with age).

8/🧵

Practically, we assess the distance between the observed binary methylation state of a CpG in a cell and the modeled estimate of bulk methylation at a particular age.

In a way, this distance conveys a probability of finding this particular cell in a tissue of a given age.

9/🧵

Due to the high dimensionality of methylation data, several thousand CpGs can be leveraged in this manner.

By taking the product of each CpG-specific probability at each age, we obtain a likelihood profile that designates the most probable epigenetic age for this cell.

10/🧵

Presently, this algorithm relies on one major assumption: independence of methylation states between CpGs. Biologically, there is likely much more at play here, so it is of particular interest to refine more advanced models that integrate these inter-CpG relationships.

11/🧵

To test our framework, we used a fantastic published dataset from Jan Vijg’s group @EinsteinMed. Here, they isolated individual hepatocytes from young (4-months-old) and aged (26-months-old) mice and obtained partial methylomes by single-cell WGBS.

genomebiology.biomedcentral.com/articles/10.11…

12/🧵

When we trained a reference model on liver tissue and applied it to single hepatocytes, we obtained remarkably strong predictions, with a median error of just around 2 months.

Interestingly, we observed higher epigenetic age variance in old cells compared to young cells.

13/🧵

We also applied our tool to muscle stem cell data from @ReikLab.

Corroborating their earlier results, we observed attenuated epigenetic aging in muscle stem cells. This may suggest that compositional dynamics are one key driving force in muscle tissue epigenetic aging.

14/🧵

While I was writing this manuscript, my colleague and friend @CsabaKerepesi led a study analyzing bulk epigenetic age dynamics during embryogenesis.

Amazingly, we uncovered a potent, natural rejuvenation event, now published in @ScienceAdvances.

science.org/doi/10.1126/sc…

15/🧵

Within this context, we were curious whether this rejuvenation event could be deciphered at higher resolution.

To investigate, we used a great single-cell multi-omics dataset from @ReikLab profiling early embryogenesis and gastrulation in mice.

nature.com/articles/s4158…

16/🧵

With our scAge approach, we observed rapid rejuvenation at the single-cell level during mouse gastrulation.

Interestingly, this was linked to de novo hypermethylation during the same period. We hypothesize that both these phenomena may be intricately connected.

17/🧵

Using lineage mapping techniques with single-cell gene expression data and an existing mouse atlas, it was possible to further increase resolution into individual cell types.

With this, we observed that epiblast cells account for the majority of the rejuvenation signal.

18/🧵

As expected, the newly formed germ layers (endoderm, mesoderm, ectoderm) display a rejuvenated epigenetic signature.

However, supportive extra-embryonic cell types do not appear to undergo rejuvenation, suggesting intricate and calculated stratification in this process.

19/🧵

Overall, we report here scAge: a flexible and scalable statistical framework designed to enable dissection of epigenetic aging trajectories at single-cell resolution.

We hope this tool will be useful to assess the great heterogeneity of aging and rejuvenation!

20/🧵

I want to recognize and thank our great reviewers @biogerontology, Danica Chen, and Lenhard Rudolph, who each provided extremely valuable and constructive feedback during the review process.

21/🧵

This thread was just a small glimpse into our paper. For those interested in a full PDF of the article, it is also available on my GitHub page. Feel free to DM me with any questions!

github.com/alex-trapp/scA…

22/🧵