(736d) Reconstruction of Waddington's Epigenetic Landscape from Single-Cell Transcriptomics of Stem Cell Differentiation | AIChE

(736d) Reconstruction of Waddington's Epigenetic Landscape from Single-Cell Transcriptomics of Stem Cell Differentiation


Gunawan, R. - Presenter, SUNY Buffalo
Papili Gao, N., ETH Zurich
Cell differentiation is the process through which stem cells form more specialized cell type(s). Because of its importance in organismal development, cellular repair, and homestasis, the molecular mechanisms of cell differentiation has been the subject of intense scrutiny. Roughly 50 years ago, along with the promulgation of the central dogma by Francis Crick and the characterization of the lactose operon by François Jacob and Jacques Monod, the existence of a “genetic program” becomes a prevailing explanation for cell differentiation. Although the mechanistic details were never formally defined, such a genetic program purports a set of “master genes” (i.e., transcription factors) that orchestrate in a precise and spatiotemporal fashion the transcription of downstream target genes, resulting in long-lasting alterations in the gene expression profiles. A notable experiment substantiating this view is the overexpression of myoD inducing a myogenic phenotype in seemingly naive cells (Davis et al., 1987). Over the past decades, the repertoire of such master genes across numerous stem cell systems begin to coalesce.

In 1957, Conrad Waddington proposed the now-famous epigenetic landscape that likens the cell differentiation process to a ball rolling on a downward sloping surface. The landscape itself is shaped by the gene network – depicted in the Waddington’s original figure as ropes that are tied to the surface – creating valleys and hills. Although this epigenetic landscape was intended only as a metaphor of how gene regulation governs cell differentiation, such a view has been revisited within the framework of dynamical systems theory. In this framework, Waddington's valleys are equated to stable steady states of a dynamical system, i.e. attractors.

While attractive, the view of a genetic program driving the cell differentiation process raises several fundamental questions. The recent advances in single-cell technologies has revealed aspects of the cell differentiation that are incompatible with the idea of ordered and programmed (i.e., deterministic) gene expression. The emerging single-cell transcriptomic data paint a stochastic differentiation process that generates a transient increase in the variability of gene expression among individual cells. Such an observation has been made in a wide variety of cell differentiation systems, including chicken erythroid progenitors, erythroid myeloid lymphoid (EML) cells, mouse embryonic stem cells (mESCs), and human CD34+ cells. Based on these observations, a different view of cell differentiation begins to materialize. Instead of stem cells executing an identical genetic program, the cell differentiation is more akin to a stochastic exploratory process. A number of recent works have proposed ways to reconstruct the epigenetic landscape from single-cell transcriptomic data using various approaches including Hopfield neural networks, a cell-density based strategy, and network entropy measurements. None of the previous studies however take into account the stochastic dynamics of gene transcriptional bursts.

In the present work, we adopted a likelihood-based approach using our algorithm CALISTA (Gao et al., 2019) to reconstruct empirically the Waddington's epigenetic landscape. Our landscape reflects the cell's gene expression uncertainty, where a high uncertainty is indicative of gene expression distributions with a high entropy. CALISTA makes use of the two-state model of stochastic gene transcription bursts to account for intrinsic cell-to-cell variability in single-cell transcriptomic data. We reconstructed the uncertainty landscape for ten publicly available transcriptomic datasets from various cell differentiation systems, all of which showed that cells undergo transition states of increased uncertainty during cell lineage specification. We further showed that this increase in uncertainty is associated with an increase in burst size of the stochastic gene transcription, an insight that would have been possible without the use of the mechanistic model in CALISTA. Furthermore, we demonstrated that the increase in the cell uncertainty is preceded by an increase in RNA velocity (Manno et al., 2018). Based on these observations, we proposed an alternative view of cell differentiation landscape centered around stochastic gene transcriptional process.


Davis, R. L., Weintraub, H. & Lassar, A. B. Expression of a single transfected cDNA converts fibroblasts to myoblasts. Cell 51, 987–1000 (1987).

Gao, N. P., et al. CALISTA: Clustering and lineage inference in single-cell transcriptional analysis. bioRxiv, 257550 (2019).

La Manno, G., et al. RNA velocity of single cells, Nature 560, 494-498 (2018).