Thread by @jsschiffman

I'm excited to share a paper w co-1st @ar_davino, @TamaraPrietoF + @landau_lab. We introduce tools to study the heritability + plasticity of cell state diversity in somatic evolution, now possible w single-cell tech. It all began with a simple question... biorxiv.org/content/10.110…

#scRNAseq has revealed that across biology, tissues contain lots of cellular diversity, with different groups of cells expressing different sets of genes. But how are these cells ancestrally related? Are cells that look similar also closely related? What would this mean?

To answer these questions, we first need two single-cell measurements, a measure of cell state (eg from scRNAseq) and a measure of ancestry (from lineage tracing). By putting these together, we can see if cells group by both gene expression and ancestry (on a phylogenetic tree).

To approach this quantitatively we introduced "phylogenetic correlations" to assess how clustered cell states are on phylogenies. This approach builds on Moran's I, previously used to measure "phylogenetic signal" in evolutionary biology, and before that, spatial correlations.

We can use phylogenetic corrs to quantify how clustered a cell state is on a tree, and also how clustered pairs of cell states are on a tree. Applying this, we saw that some cell states, eg brain cancer stem cells, do form clusters, whereas others, eg cell-cycle phase, do not.

But what does this mean? We reasoned that if a cell state is phylogenetically clustered that it is likely heritable and maintained through cell divisions. If not clustered, perhaps the state is more plastic. If 2 states co-cluster, maybe cells can transform from one to the other.

Evolutionary change depends in part on the selection of heritable variation, and so identifying heritable cell state variation in somatic tissues could help us better understand somatic evolution. Further, cell state transitions are thought to play a role in cancer progression.

To work this out we connected phylogenetic corrs with a model of cell state transition dynamics. This showed that if cell state is inherited, and transitions to other states are rare, the state will be clustered. We could also use this connection to infer transitions from trees.

We can infer transitions between cell states b/c phylogenetic corrs measure the "excess" probability of observing closely related cells in particular states (vs cells drawn at random), and we realized that this "excess" probability is related to the state transition probability.

Together, @ar_davino, @TamaraPrietoF and I applied this method, first to simulations, and next to published single-cell phylogenies to reveal cell state heritability and plasticity (or commitment strength) in both developmental and cancer contexts.

As an example of what an approach like this can reveal, in brain cancer (glioblastoma), we saw that astrocyte-like cells might be a bridge between stem-like and mesenchymal-like cells, and collaborated w @MarioSuva + team to validate w CRISPR scarred gliomasphere phylogenies.

This framework, which we call PATH, can be applied to measure phylogenetic correlations and infer state transitions for any phenotypically annotated single-cell tree, and "phenotypes" do not have to be limited to only transcriptional states. See code here: github.com/landau-lab/PATH

We're really excited about this project and think that as single-cell lineage tracing technologies continue to improve, the application of PATH to these data will help reveal key features of somatic evolution.

For more, see this great overview by @landau_lab!

https://twitter.com/landau_lab/status/1608979794192109569

This project wouldn't have happened without the support, vision and deep insights from my PI @landau_lab and co-authors, @ar_davino, @TamaraPrietoF, Catherine Potenski, Yilin Fan, @hara_toshiro, and @MarioSuva, and the environment at @nygenome + @WeillCornell. Thanks for reading!