Minimal Toolkit for Stem Cell Self-Renewal

“We wanted to understand how stem cell transcription factors were interconnected and explain how these cells behave in different culture conditions and still maintain the stem cell state,” said Smith. “This modeling approach allowed us to map the transcription factor connections that is consistent with experimental data.”

”The conclusion here is that the gene network of self-renewal is simple, which should be the case,” said Sui Huang, a professor at the Institute for Systems Biology in Seattle. “We have known for a long time that the best networks for informational processing should have just about two connections per gene, on average. That the results here come close to this universal principle is striking.”

Jordi Garcia-Ojalvo, a professor of systems biology at the Universitat Pompeu Fabra in Barcelona, Spain, agreed. “This study confirms the hypothesis that a relatively small number of components and interactions can define a cell’s behavior, he told The Scientist in an e-mail. “This [concept] has been previously described in other biological processes, such as circadian rhythmicity, the cell cycle, and even stem cell pluripotency, but is shown here through a systematic analysis of a data-driven computational model.”

According to Ihor Lemischka, director of the Black Family Stem Cell Institute at Mount Sinai in New York City, this is one of the first computational models of stem cell fate decisions. “With this model, we now have testable predictions,” he said. “In the future, this will move us towards a way to understand how to generate a cardiac muscle or other specific cell type without having to try an unwieldy number of cell culture conditions.”

Smith and his colleagues began with stable, homogenous populations of cultured mouse ESCs derived from the blastocyst. The software the Microsoft team then built generated billions of possible models that could explain stem cell self-renewal. Most of these were invalidated when challenged with experimental data generated in Smith’s laboratory, which measured gene expression of ESCs across 23 different cell culture conditions, all of which maintained pluripotency.

The remaining models all contained a core network of 12 transcription factors—including Sox2 and Oct4—that captured how self-renewal is regulated through just 16 interactions.

The team then asked this minimal model set to predict whether knockdown of a specific transcription factor would maintain the stem cell state or prompt differentiation. Of the 37 model-generated predictions the researchers tested in the lab, 26 were correct.

Garcia-Ojalvo expressed surprise that a network based only on gene expression data could predict, with relative accuracy, the effect of multiple genetic interactions. He noted that the accuracy could be refined further with additional experiments, or the inclusion of protein interaction data, for example.

Meanwhile, in a separate study published today (June 5) in Molecular Cell, structural and protein biologist Shohei Koide from the University of Chicago, and his colleagues pinpoint a crucial protein-protein interaction that allows mouse ESCs to differentiate. It could be useful, said Garcia-Ojalvo, “to analyze protein-protein interactions in pluripotency using the modeling approach in the Science paper.”

The precise mechanisms that cause mouse and human ESCs to differentiate are not yet known, said Smith. But he suggested his team’s tool could help: “We think that the starting point we now have is good enough to begin to address this.”

Toolkit in hand, Smith’s team is now probing whether these minimal conditions can be used to reprogram human ESCs to a more-naive state.

S.J. Dunn et al., “Defining an essential transcription factor program for naive pluripotency,” Science, doi:10.1126/science.1248882, 2014.

N. Yasui et al., “Directed network wiring identifies a key protein interaction in embryonic stem cell differentiation,” Molecular Cell, doi:10.1016/j.molcel.2014.05.002, 2014.