stuart_a_kauffman's picture
Professor of Biological Sciences, Physics, Astronomy, University of Calgary; Author, Reinventing the Sacred

Demonstration That Cell Types Are Dynamical Attractors

The human body has, by histological criteria, some 285 cell types. Familiar examples are skin cells, liver cells, nerve cells, muscle cells. But the human embryo starts as a single fertilized egg, the zygote, and in the process call ontogeny the zygote divides about 50 times to create the new born baby, not only with 285 cell types, but the intricate morphology of the human. Developmental biology is the study of these processes. 

The process by which the zygote gives rise to different cell types is called cell differentiation. 

By the 1950s, using protein separation techniques such as gel electrophoresis, it became clear that different cell types manufactured their own specific set of proteins. For example, only red blood cells make the protein hemoglobin. By then genetic work had convinced all biologists that genes were like beads lined up single file on chromosomes. The emerging hypothesis of the day was "one gene makes one protein". If so, different cells had different active subsets of the total complement of human genes on the diverse chromosomes. The active genes would be making their specific proteins.

By 1953 Watson and Crick established the structure of DNA, the race was on to discover the genetic code, soon worked out, by which the sequence of nucleotide bases in a gene encodes a corresponding messenger RNA, which was then translated, according to the soon known genetic code, into proteins by a host of protein enzymes and the RNA ribosome.

These brilliant results, the core of molecular biology, left hanging the deep question: How do different cell types manage to have different subsets of the full set of human genes active in the corresponding cell type?

Two superb French microbiologists, F. Jacob and J. Monod in 1961 made the first experimental breakthrough using the bacterium E. coli. They showed that, adjacent to the gene encoding a protein called beta galactosidase, a small "operator" DNA sequence, "O", bound a "repressor protein", "R". When R was bound to O, the adjacent gene for beta galactosidase could not be copied into its messenger RNA. In short, genes and their products could turn one another "on" and "off". 

By 1963 these two authors cracked, it seemed, the central problem of how 285 cell types with the same genes could possibly have different patterns of gene activity in the 285 cell types. Jacob and Monod proposed that two genes, A and B, which turned one another "off", ie "repressed" one another, could have two "steady states of gene activity: 1) A "on" B "off"; 2) A "off" B "on". 

In brief, by imagining a very simple "genetic circuit" where A and B repress one another, the genetic circuit, like an electronic circuit, could have two different steady state patterns of gene activity, each corresponding to one of two cell types, the first making A protein, the second making B protein.

In principle, Jacob and Monod had cracked the problem of cell differentiation: how cells with the same genes could exhibit diverse and unique patterns of gene activities.

I was entering biology at that time, with a background studying neural circuits and logic. It was clear to most biologists that some huge genetic network among the then thought 100,000 human genes somehow controlled cell differentiation in ontogeny. My question was an odd one: Did evolution have to struggle hard to evolve very specific genetic networks to support ontogeny, or, I hoped, were there huge classes, or "ensembles" of networks that as a total class, behaved with sufficient self-organized order to account for major features of ontogeny? 

To explore this, I invented "random Boolean networks", a set of N "light bulbs", each receiving regulatory inputs from K light bulbs, and turing on and off according to some logical, or Boolean rule. Such networks have the property of having a generalization of the Jacob and Monod "A "on" B "off" versus A "off B "on", property. These two patterns are called "attractors", for reasons I make clear in a moment.

I studied networks with up to 10,000 light bulb model genes, each with K = 2 inputs, and sampled the class or ensemble of N = 10,000 K = 2 possible networks by assigning the 2 inputs to each model gene at random and the logical, or Boolean, rule to that gene at random from the 16 logical functions of two inputs. It turned out that such networks followed a sequence of states, like a stream, and settled down to a cycle of states, like a stream reaching a lake. Many streams, or trajectories, typically flowed into each "state cycle" attractor, called an "attractor" because the state cycle lake attracts a set of trajectories to flow into it. More each network had many such state cycle lake attractors.

The obvious hypothesis was that each attractor corresponded to a cell type.

On this hypothesis, differentiation was a process in which a signal, or chemical noise, induced a cell to leave one attractor, and reach a state that flowed along a trajectory that flowed to another attractor cell type.

This remained a mere hypothesis until a decade ago, a brilliant biologist, Dr. Sui Huang, used a mild leukemic cell line, HL60. He used two different chemical perturbations, All Trans Retinoic Acid and DMSO, to treat two sets of HL60 cells. At regular intervals, Huang used the technique of gene arrays to sample the activities of 12,000 genes. He showed, wonderfully, that under the two chemical treatments, HL60 followed two divergent then convergent "stream-trajectory" pathways in the patterns of activity of 12,000 genes, both of which ended up on the same pattern, corresponding to a normal polymorphoneuterophil blood cell.

Huang's powerful result remains the best evidence that cell types are indeed attractors.. If this is true, it should be possible to perturb the gene activities of cell types to control their differentiation into desired cell types.