Natural variation in genome-wide gene expression is important for adaptation to changing ecological circumstances if (i) there is population variation in genome-wide gene expression, (ii) that variation is heritable, and (iii) variants have differential reproductive success. In this issue, Brown et al. (2008) identify the molecular basis of an amino-acid biosynthesis and transport polymorphism that was discovered in the first study to examine natural gene expression variation on a genomic scale. In that study, Cavalieri et al. (2000) profiled gene expression in a Tuscan vineyard isolate of Saccharomyces cerevisiae (Fig. 1) and four of its progeny. They showed that the progeny segregated 2 : 2 for manifold differences in gene expression, including a strongly segregating phenotype of amino acid biosynthesis versus import from exogenous sources. Tracking down the genetic basis of such a phenotype in a background of natural variation may seem a daunting task. Yet, Brown et al. have mapped the source of the phenotypic polymorphism to a single nucleotide insertion, using a few brilliant leaps of intuition, a detailed knowledge of gene function for budding yeast, and some of the classic molecular tools available for yeast.
How did Brown et al. do it? The first step was a key intuitive leap, which required some knowledge of yeast molecular biology. There is a toxic amino acid analogue of leucine, trifluoroleucine (TFL), that facilitates analyses of amino acid biosynthesis and transport. TFL may be used as a selective agent for transformations of prototrophic strains that otherwise must be mutated to create auxotrophies, because a dominant TFL resistance marker (LEU4-1) is available and may be cotransformed with the desired plasmid or DNA (Bendoni et al. 1999). As it turns out, trifluoroleucine resistance also cosegregates in the yeast progeny with the phenotype of up-regulation of amino acid biosynthesis. Early experimental work with the strains also indicated that the phenotypic differences between these strains diminished when amino acids were excluded from the growth medium. Together, these data led Brown et al. to speculate that a strain deficient in its amino acid sensing machinery would produce these results. Such a deficiency would lead inevitably to the de novo construction of amino acids instead of their import, excluding toxic TFL from the cell. Moreover, cells without the ability to sense amino acids and activate genes for amino acid import would be predicted to show no phenotypic differences when grown in amino-acid-rich or amino-acid-poor media.
An extensive knowledge of gene function in yeast provided Brown et al. with an immediate candidate pathway for amino acid sensing that is comprised of three genes: SSY1, PTR3, and SSY5. Brown et al. used the naturally segregating TFL sensitivity to facilitate transformation of the prototrophic natural strains with ‘wildtype’ plasmids containing the S288c laboratory strain alleles of these three genes, and were rewarded with an immediate success, discovering that of the three genes, the s288c allele of SSY1 alone complemented the natural allele, conferring sensitivity to TFL. Sequencing of the SSY1 alleles from two genetically diverged natural strains showing the phenotypic segregation clinched the case, implicating an insertion segregating in both strains of an extra thymine in a poly-T tract of the coding sequence of SSY1.
To demonstrate that this small genetic change can produce a dramatic effect on gene expression, Brown et al. introduced this ssy1t9 allele into a laboratory strain of yeast. They show that the effect of this allele on gene expression recapitulates nearly half of the differences observed, segregating in the natural offspring, including nearly all of the differences in amino acid biosynthesis and transport. Brown et al. wrap up the details of their research neatly by also showing similar effects on gene expression caused by a knockout of SSY1, confirming the effective loss of function of the naturally segregating nonsense frameshift mutation.
Two major lessons from this study deserve further emphasis. First, the fact that the genetic basis of this dramatic expression phenotype was tracked down to a single nucleotide insertion provides evidence that a small genetic difference can lead to dramatic differences in gene expression. Many recent studies have demonstrated a lack of correlation between total genetic divergence and total gene expression divergence (Tirosh & Barkai 2008), a finding that is easily explained if individual mutations or substitutions have a high variance in effect. Discovering that 45% of the natural variation segregating in the progeny of this isolate of S. cerevisiae is attributable to a single nucleotide change bespeaks such a high variance. Brown et al. justify this percentage by analyses of the divergent statistical power of their experiments. The microarray literature is rife with incautious assertions and comparisons that fail to incorporate considerations of the statistical power that was applied towards each treatment or experiment, so the justification is all too rare, as well as being fundamental to the conclusions they obtained.
Second, the mapping of this extensive expression polymorphism to a mutation of an environmental sensor argues that a broadening of the suite of elements included in models of gene expression variation may be called for. Many models have perhaps inappropriately focused only on cis-regulatory sequences and transcription factors, as though gene expression were purely developmental information rather than an actively modulated response to the environment. In fact, it may not just be reasonable but may be generally desirable to think of gene expression as a norm of reaction or response curve to the environment, rather than as a traditional fixed quantitative trait. Gene expression as a phenotype may share more in common with ethology than with morphology.
One logical extension of this work would be to attempt to discover the genetic basis for the other expression differences and the colony morphology observed by Cavalieri et al. (2000). It seems likely based on the cosegregation of differences within a single tetrad observed in that study that they may be attributable to one or several other segregating genetic polymorphisms, perhaps with small but epistatic effects. However, Brown et al. provide an intriguing lead-in towards a much broader question. They assayed the fitness of the naturally segregating allele in the heterozygous parent and both homozygous offspring, showing that in laboratory conditions the ssy1t9 allele only mildly impairs growth in heterozygotes and more strikingly impairs growth in homozygotes. They further calculate, based on these fitness values and the high insertion mutation rate in homonucleotide runs, that the frequency of the deleterious recessive allele may be as high as 1 in 50 given these parameters. The unexplored implication is that a substantial portion of gene expression variation in populations may be attributable to moderately rare deleterious recessive alleles. The broad question to be answered is whether such genetic variation might underlie the fairly constrained distribution of expression differences observed within populations across diverse taxa (Clark & Townsend 2007), or whether other evolutionary dynamics are at work. To answer this question will require further characterization of the fitness effects of natural variants associated with expression variation as well as modelling of the population genetics of the labile phenotype of gene expression.