A recent paper in Science (Synchronous and Stochastic Patterns of Gene Activation in the Drosophila Embryo, Boettiger and Levine, July 24, 2009, Vol. 325. no. 5939, pp. 471 - 473) describes a mechanism that may explain some of that stability. Development is a time of rapid and contingent gene expression, demanding that at least a critical mass of cells in a developing tissue respond to signals in the same way, so that the next stage of growth can proceed. But not all cells that receive the same signal respond in the same way, such as by expressing a given gene at a specified time.
A note in the September Nature Reviews Genetics (Polymerase stalling gets genes in sync, p. 590) asks:
How is this variability dealt with in situations in which precise patterns of gene activation are important? A recent study [Boettiger and Levine] suggests a mechanism that can reduce variability in the onset of transcriptional activation in the Drosophila melanogaster embryo and may contribute to the precision of the developmental programme.One of the initial steps in gene transcription is the recruitment and assembly of the RNA polymerase II complex that then starts the synthesis of new protein. If that complex isn't ready and waiting when a cell receives a signal to turn on a gene, the cell may not respond in a timely way, and the gene won't be turned on when needed.
Boettiger and Levine describe a series of elegant experiments looking at the timing of expression of a number of important 'control genes' in hundreds of fruit fly embryos.
These studies revealed two distinct patterns of gene activation: synchronous and stochastic [meaning random]. Synchronous genes display essentially uniform expression of nascent transcripts in all cells of an embryonic tissue, whereas stochastic genes display erratic patterns of de novo activation. RNA polymerase II is "pre-loaded" (stalled) in the promoter regions of synchronous genes, but not stochastic genes. Transcriptional synchrony might ensure the orderly deployment of the complex gene regulatory networks that control embryogenesis.
The timing differences are significant; synchronous expression of genes in different cells happens within 2 minutes of each other, while stochastic expression varies by as much as 20 minutes. Boettiger and Levine suggest that this may indicate two classes of genes, those for which timing of expression is crucial, and those for which it's less crucial. What controls the pre-loading of the RNA polymerase is not clear, nor how much play there still is in the process--previous experiments have shown that there is considerable variability in expression of the same gene in different cells, including non-expression, so the Boettiger/Levine classification scheme is clearly not exhaustive.
In many ways, randomness is crucial to evolution, but too much randomness during development can be lethal. As the Boettiger and Levine experiments show, evolution has produced ways to rein it in.