When cell biologist Sui Huang decided to study what many researchers find to be their biggest nuisance, he found that experimental noise had something surprising to say, something that might help stem cell scientists both differentiate and reprogram cells more efficiently.

Huang wanted to know what he was looking at when he saw a bell-shaped curve that showed, say, varying levels of protein expression even in cell populations that had been grown from the same original cell. Did the curve represent wide, rapid fluctuations within individual cells or a diversity of relatively stable states within the population of cells? If the latter, could that make a biological difference?

Publishing this week in Nature, Huang shows that cells exhibit persistent 'nongenetic individuality' that predicts their behavior1.

He and his colleagues separated a population of haematopoietic progenitor cells into groups according to the relative levels of a marker called Sca-1, and looked to see whether each group of cells could reconstitute the range of levels of Sca-1 observed in the original population.

The cells did so, but surprisingly slowly. Cells that were in the top or bottom 2% in terms of Sca-1 expression required more than 17 days (or 12 cell doublings) to recover the heterogeneity of the original population. But these outlier cells also showed stunning differences in their propensities to differentiate down blood lineages. Even a week after the population had been split according to high, medium and low levels of Sca-1, cells in the lowest fraction were seven times more likely than in the highest level to differentiate toward red blood cells, and the difference in their proclivity to differentiate disappeared over time.

Crucially, says Huang, these outlier cells had not differentiated; rather they were poised to do so. “They are still fully multipotent cells; if left alone they would continue to replace the distribution of the original population.”

Other markers Huang is looking at in embryonic stem cell lines, including pluripotency factor Oct4, show similar patterns of variation. The timing varies, he says, but it's always over days or weeks rather than minutes or hours. The variation isn't confined to a single cell-surface marker; instead, it reflects different states across the entire transcriptome. “One marker is just a projection of an entire network of thousands of genes,” he says.

In fact, he says, the quality of stemness itself may depend, in part, on noise. “The random heterogeneity of gene expression that we observed may be the central driving force behind 'multipotency'.”

“The random heterogeneity of gene expression that we observed may be the central driving force behind 'multipotency'.” Sui Huang

Practically, the observations mean that researchers should be able to sort cells in order to pluck out those within a population that are inclined to a particular behaviour, thus boosting rates of reprogramming or differentiation, processes that are notoriously inefficient.

Theoretically, the results provide strong evidence for a theory of how gene networks create cell types. Huang visualizes different cell types as a series of funnel-shaped wells in a landscape: individual cells are like marbles rolling around those wells, and marbles at lower points of a well would be harder to knock into other wells than marbles higher up.

“Gene expression noise allows these jumps between potential wells,” says Huang, “but we don't know the details.” The current work indicates that the wells are pitted, with locally stable states within a single well located close to its borders. Marbles in these positions represent cells that are poised to respond to, say, differentiation signals or drugs. This theory can explain why cells within a single population show different behaviours. “If you treat cells with a lower and lower dose, instead of seeing every cell have a weak response, some cells have a full response,” he says, “and some do nothing.”

Huang says these explanations are more intuitive for physicists than biologists. Biologists tend to think in terms of pathways, such that pushing a particular gene within a particular type of cell will have a particular effect. In fact, when Huang gained the first empirical evidence for this theory of gene-regulatory networks within cells2, originally posited by Stuart Kauffman over 40 years ago3, he ended up publishing it in a physics journal.

Still, the work got noticed. A few months ago, Kauffman, a professor of biological sciences, physics and astronomy at the University of Calgary, Alberta, Canada, recruited Huang from his post at Harvard University, in Cambridge, Massachusetts, to Calgary's new Institute for Biocomplexity and Informatics.

Huang, says, the study of what was once considered a relic of inaccurate instruments is becoming fashionable. “We assume that this is experimental noise, but actually it's not,” he says. Man-made homogenous beads that are run through cell-sorting machines show sharp peaks, whereas populations of real, live cells show clouds. “Much of the variability we see is biological,” he says, and nature depends on this variability to create the order necessary to maintain multicellular organisms.

“The noise gets amplified into something meaningful,” Huang says. “To put it in romantic terms, randomness is the ultimate creative force.”