A study in Nature (11/10/05, full text by subscription) by Scott Rifkin and colleagues performs an interesting experiment on gene expression and mutation.
Beginning with 12 identical lines of Drosophila melanogaster, they allowed each to reproduce for 200 generations. During this time, they expected the lines to accumulate some mutations that would affect gene expression. But how many?
On the basis of studies in D. melanogaster, we estimate that each of our 12 mutation accumulation lines contains around 360 mutations. We measured gene expression levels during the third larval instar (before puparium formation; BPF) and at puparium formation (PF), before and after the peak of a large pulse of the hormone 20-hydroxyecdysone that triggers the start of metamorphosis (see Methods and Supplementary Fig. 1). This stage is one of substantial transcriptional activity and turnover, with broad intra- and interspecific variation in gene expression. Of 11,798 genes measured, we detected significant Vm for 3,816 genes at the BPF stage, for 3,475 genes at the PF stage and for 4,658 genes overall, using a false discovery rate (FDR) of 0.05. The expression of 5,729 genes significantly differed between the two stages, although only 2,509 of these genes showed significant Vm (FDR = 0.05) (Rifkin et al. 2005, citations omitted).
The study is about differences in mRNA transcription in these lines, so it is not necessarily generalizable to everyday traits. But their preferred model to explain the level of mutational variance is interesting:
Third, network output, namely the production of a particular product at a specific place and time, may be the target of selection rather than gene expression itself. As in enzyme flux models, the selective effect of any particular change in gene expression may be negligible over a range of values but become substantial when the abundance of mRNA becomes rate limiting or when the variation becomes otherwise functionally relevant. Stabilizing selection, by canalizing network output against perturbations, may facilitate neutrality among members of the underlying network. Gene expression would be able to tolerate a moderate number of mild mutations but would trigger strong selection if the network output were substantially affected. Such a model could also account for moderate correlations between mutational and interspecific variation even when the total level of between-species divergence is far less than expected under neutrality (ibid.).
It's an argument of developmental robusticity via canalization. Individual genes are free to vary somewhat, as long as they don't surpass some threshold. But if the network output exceeds its tolerances, selection kicks in. One possible adaptive response is for certain genes to reduce the phenotypic variability, either by reducing the effects of environmental variability (canalization) or by reducing the disruptive effects of mutations internal to the network (developmental robusticity).
Rifkin SA, Houle D, Kim J, White KP. 2005. A mutation accumulation assay reveals a broad capacity for rapid evolution of gene expression. Nature 438:220-223. Full text (subscription)