Population Genetic Variation in Genome-Wide Gene Expression (original) (raw)

Abstract

Evolutionary biologists seek to understand which traits display variation, are heritable, and influence differential reproduction, because such traits respond to natural selection and underlie organic evolution. Selection acts upon individual differences within a population. Whether individual differences within a natural population include variation in gene expression levels has not yet been addressed on a genome-wide scale. Here we use DNA microarray technology for measuring comparative gene expression and a refined statistical analysis for the purpose of comparing gene expression levels in natural isolates of the wine yeast Saccharomyces cerevisiae. A method for the Bayesian analysis of gene expression levels is used to compare four natural isolates of S. cerevisiae from Montalcino, Italy. Widespread variation in amino acid metabolism, sulfur assimilation and processing, and protein degradation—primarily consisting of differences in expression level smaller than a factor of 2—is demonstrated. Genetic variation in gene expression among isolates from a natural population is present on a genomic scale. It remains to be determined what role differential gene expression may play in adaptation to new or changing environments.

Introduction

The attention of evolutionary biologists has long been focused on how natural selection brings about adaptation. The ability to sequence DNA and infer primary protein structure has focused most of this attention on changes in amino acid sequences that alter the kinetics or function of a protein (Kreitman and Comeron 1999). The question of whether other types of mutations are evolutionarily relevant has not been satisfactorily answered (Stern 2000). Many authors have suggested that regulatory mutations may play a much greater role in evolutionary change than do structural mutations (Zuckerkandl 1963; Macintyre 1982; Wilson 1985).

Recently, the debate over the relative importance to adaptive evolution of regulatory compared with structural genetic variation has diminished, despite the discovery of examples of interspecies variation in gene expression level important to evolution (Wilson, Carlson, and White 1977; Dickinson 1987; Hammer and Wilson 1987; Dickinson 1991; Wang, Marsh, and Ayala 1996). New tools, now including those emerging from genomics, can identify regulatory polymorphism on a genomic scale, rather than gene by gene, allowing a comprehensive assessment of whether regulatory variation contributes to functional evolutionary innovations (Paigen 1989). For regulatory variation to be evolutionarily relevant, it must be present in natural populations, it must be heritable, and it must lead to differential lifetime reproductive success.

The first of these requirements has been satisfied only with regard to a number of single genes or protein products (Paigen 1979, 1986; Laurie-Ahlberg et al. 1982; Laurie-Ahlberg 1985). Many of these examples have been explored in Drosophila or Mus. Although the potential for complex tissue–level regulatory changes to be important in adaptive evolution in these multicellular organisms is clearly very great, discerning the nature of such differences on a genome-wide scale will be difficult because few tissues are easily isolated and cellularly homogeneous. Here we compare four natural isolates of unicellular wine yeast from Tuscan vineyards, assessing the degree of variation in transcription on a genomic scale. Our microarrays contain all identified open reading frames from the sequenced genome of S. cerevisiae, eliminating any bias from choosing a single gene or pathway for study. The demonstration of variation in gene expression among individuals from a natural population is a necessary prerequisite to evaluating the larger claim that regulatory changes play an important role in organismic evolution.

Methods

Methods were modified from those of Eisen and Brown (1999).

DNA Microarray Construction

A set of clones containing 6,188 verified open reading frames (ORFs) of the yeast genome were obtained from Research Genetics (Huntsville, Ala.) and amplified to levels required for preparation of DNA microarrays by means of the polymerase chain reaction as in Hardwick et al. (1999). Some of the longer ORFs were amplified with the Gibco BRL Elongase Amplification Kit (Life Technologies, Rockville, Md.). Each amplified product was confirmed by agarose gel electrophoresis. Ninety-eight percent of the ORF amplifications yielded bands of appropriate length. The amplified DNA was precipitated with isopropanol, washed with 70% ETOH, and resuspended in Micro Spotting Solution (Telechem, Sunnyvale, Calif.). The DNA was spotted on CMT-GAPS γ-aminopolysilane–coated glass slides (Corning, Corning, N.Y.) or on polylysine slides (Eisen and Brown 1999), using a microarraying robot with a 16-pin head constructed from a design by Patrick O. Brown (http://cmgm.stanford.edu/pbrown/).

Extraction of mRNA

RNA was extracted from flash frozen pellets of yeast cultures grown aerobically in 100-mL culture at 30°C in a shaker at 225 rpm to an optical density of 0.8 in YPD medium (1% yeast extract, 2% peptone, and 2% dextrose). The flash-frozen yeast pellet was resuspended and mRNA extracted with a hot acidic phenol/chloroform extraction. Nucleic acids were ethanol precipitated, washed, dried, and redissolved in TE buffer. Yield ranged from 10 to 15 mg, with a spectrophotometric ratio of absorption (260 nm/280 nm) of approximately 2.0. Pellets were stored frozen at −20°C. The mRNA was purified using the Quiagen (Valencia, Calif.) Extraction Kit. The poly-A RNA was stored at −20°C.

Reverse Transcription and Hydrolysis

Reverse transcription was performed with oligo-dT (a mixture of dT 16-mer, 17-mer, 18-mer, 19-mer, 20-mer, 21-mer, and 22-mer) and poly-dN primer (Poly-dN6). Amino-allyl-dUTP (Sigma) was incorporated into cDNA along with dNTPs using the reverse transcriptase Superscript II. After at least 2 h, 1 M NaOH and 0.5 M EDTA were added, and the mix was incubated at 65°C for 15 min. Then 1 M HEPES pH 7.5 was added.

Buffer Cleanup and Cyanine Dye Coupling

The reverse transcription reaction product was diluted 10-fold, then concentrated 20-fold with Microcon-30 microconcentrators. Dilution 20-fold and concentration 20-fold was performed another two times. To this purified concentrate, 1 M NaHCO3 pH 9 was added, with an appropriate NHS-cyanine dye aliquot. This coupling reaction was incubated in the dark at 25°C for 75 min and then stored in the dark at 4°C for less than 24 h.

Probe Purification

The labeled probe was purified with a Quiaquick column. This elution of purified cyanine-labeled cDNA was stored at 4°C for less than 24 h.

Hybridization

The labeled cDNA was concentrated in Microcon-30 microconcentrators, combining appropriate cyanine-3–labeled and cyanine-5–labeled paired samples. Poly-dA 12-mer to 18-mer, SSC, and HEPES pH 7.0 were added. The mix was filtered with a Millipore 0.45 μm filter. 10% SDS was added, and the mix boiled for 2 min. It was then cooled at 27°C for 10 min. A microarray slide was set in a hybridization chamber, using drops of 3 × SSC on the underside adsorbing to the slide corners and the chamber bottom. Then 3 × SSC was added to the hybridization chamber wells. A Lifterslip coverslip was cleaned with ethanol, then placed over the printed microarray. The labeled cDNA mix was injected at the corners of the Lifterslip, and the chamber was sealed and then placed level in a 60°C water bath to be incubated at 60°C to 63°C for 12 to 15 h.

Array Wash

Hybridized microarray slides were washed in a solution of 387 mL purified water, 12 mL 20 × SSC, and 1 mL 10% SDS and rinsed in a solution of 399 mL purified water and 1 mL 20 × SSC. The array was stored, if needed, in the dark, for less than 2 h and then scanned.

Data Acquisition and Analysis

Fluorescent DNA bound to the microarray was detected with a GenePix 4000 microarray scanner (Axon Instruments, Foster City, Calif.), using the GenePix 4000 software package to locate spots in the microarray. Fluorescence intensity values were adjusted by subtracting background from foreground. To eliminate signals that are most prone to estimation error, any spot was excluded from analysis if both the cyanine-3 and cyanine-5 fluorescence signals were within three standard deviations of the distribution of intensities of the background pixels for that spot. This procedure avoids artificially inflated measurements of expression due to near-zero values in the denominator. Expression values were normalized by linear scaling of the cyanine-5 values so that the mean cyanine-5 and cyanine-3 background-corrected intensity values of nonexcluded spots were equal. Because the hybridizations were of uniformly high quality, this straightforward method yielded linear log-log cyanine-3–cyanine-5 intensity scatterplots for all hybridizations and no further manipulation of the data was necessary.

We chose to study four homothallic, diploid natural isolates from Montalcino, Italy. These isolates were previously characterized using RAPD, genetic analysis, and a subtelomeric probe (Cavalieri et al. 1998). Results here are derived from comparisons of global gene expression of those four isolates in an eight-microarray, dye-swap circle (fig. 1), so that there would be equivalent information on each isolate. Both cross-comparisons were performed once to provide a direct comparison between nonneighboring strains in the experimental design.

These data were analyzed using a Bayesian analysis of gene expression level (Townsend and Hartl 2002). To arrive at estimates of gene expression level across samples, BAGEL uses Markov chain Monte Carlo integration of the likelihood of the data across expression levels and variances for each gene. BAGEL reconciles multiple cDNA microarray comparisons among multiple samples to produce estimates and credible intervals for gene expression levels. This kind of analysis of a replicated design is robust to the selective absence of data for a gene due to low signal in a particular hybridization. Accurate estimates of expression level with lesser precision could be calculated even when spots were unacceptable for as many as five of the 10 hybridizations. Thus, only 210 out of 6,188 yeast ORFs were not analyzed in this study due to recurrent low signal. Hierarchical clustering of genes by the log2 expression levels estimated by BAGEL was performed with Cluster software (Eisen et al. 1998) using the average linkage algorithm.

Results

We used the conservative criterion that genes with differential expression have nonoverlapping 95% credible intervals from the Bayesian analysis. By this criterion, 433 genes were differentially expressed among some or all of these four natural isolates. Table 1 lists the numbers of genes that had nonoverlapping credible intervals between each pair of natural isolates, demonstrating that some strains are more divergent in gene expression than others. M2-8, for instance, showed more genes that are significantly different when compared with any given strain than did any other isolate. M5-7 was the next most divergent in gene expression. In contrast, M1-2 and M7-8 had very few differences in gene expression. The three genes that were detected as significantly more abundantly expressed in M7-8 compared with M1-2 were HEM1 (5-aminolevulinate synthase, up 14%), and MFA1 (mating factor a1, up 33%), and the ORF YJL070c (unknown function, up 17%). The two genes that were detected as significantly more abundantly expressed in M1-2 compared with M7-8 were ARR2 (arsenate reductase [see below]) and YKR087c (unknown function, up 18%).

Hierarchical clustering of the significantly different genes by their log2 expression levels graphically displays the variation among strains (fig. 2). The most notable division is between genes meagerly (nodes 2 to 3) and abundantly (nodes 4 to 6) expressed in M2-8. A subset of the genes abundantly expressed in M2-8 appear abundantly expressed in M5-7 (node 4). A few genes are meagerly expressed only M5-7 (nodes 1 and 6) and expression levels appear largely consistent between M1-2 and M7-8.

Amino Acid Biosynthesis

M2-8 was expected to show differences in expression level of amino acid biosynthetic genes, owing to the large differences in expression found for these genes in its homozygous progeny (Cavalieri, Townsend, and Hartl 2000). Significantly different amino aid biosynthesis genes, except as noted, all clustered by expression levels in node 5 of figure 2. The expression levels of the genes LYS20 and LYS21, whose products initiate the lysine pathway by converting acetyl-CoA and 2-oxoglutarate to homocitrate, were higher in M2-8 by factors of 2 and 3 (fig. 3). Moreover, the conversion, unique to fungi, of α-aminoadipate to lysine, is completed by the products of LYS9 and LYS1, and these genes were more abundantly expressed in M2-8 compared with other strains by factors > 4 and > 6, respectively.

In the homozygous progeny from M2-8, import of amino acids and synthesis of amino acids appear to be complementary approaches for acquiring the basic units to construct proteins at a comparable rate (Cavalieri, Townsend, and Hartl 2000). Figure 4 shows the expression of LYP1, encoding a lysine permease; KRS1, encoding the lysyl-tRNA synthetase; and LYS14, encoding a lysine pathway transcription factor. Intriguingly, the lysine permease mRNA was abundant in M2-8, as was the lysyl-tRNA synthetase mRNA. M2-8 clearly appears to have an up-regulated lysine pathway. However, the obvious candidate transcription factor, LYS14, encoding a positive regulator that binds to a DNA motif present in LYS1, LYS2, LYS4, LYS9, LYS20, and LYS21 (Becker et al. 1998), showed an abundance profile that suggests there is little to no difference in expression among strains.

Seven other genes in amino acid biosynthesis pathways had nonoverlapping 95% credible intervals in M2-8 compared with other strains (fig. 5), including three genes in the histidine pathway. Broad-affinity amino acid permeases BAP2 and BAP3 were abundant in the same strain with up-regulated amino acid biosynthesis, as was oligopeptide uptake via OPT2 (fig. 6). The gene SSU1 was less abundant in M2-8 compared with other strains by at least a factor of eight or more (fig. 2, node 2 and fig. 7). It encodes a sulfite exporter; mutants are sensitive to sulfite (Avram and Bakalinsky 1997). Interestingly, its well-studied positive transcriptional regulator, the five zinc-finger FZF1 (Avram and Bakalinsky 1996), was estimated to be expressed about 30% less in M2-8 (data not shown), although credible intervals overlap among all strains.

Protein Degradation

M2-8 appears to have up-regulated amino acid biosynthesis and oligopeptide uptake. Does this strain degrade more proteins as well? Significantly differently expressed genes involved in protein degradation clustered together in node 5 of figure 2. Intriguingly, M2-8 and M5-7 share a molecular phenotype of ubiquitin degradation. Table 2 lists estimates of the gene expression levels of four components of ubiquitin degradation. These are the ubiquitin precursor protein UBI4 (which is posttranslationally processed into ubiquitin), the ubiquitin-protein ligase UBA1, the ubiquitin fusion degradation protein UFD1, and the ubiquitin-specific protease UBP5. Transcripts of these genes were consistently estimated at higher levels in M5-7 and M2-8, frequently with nonoverlapping 95% credible intervals. The degree of difference was not very large (about 50%). However, considering the importance of ubiquitin degradation to the metabolism of proteins in the cell, the difference is possibly biologically significant.

Much of protein degradation is accomplished in yeast by the 22 components of the 20S proteasome, estimates of the expression levels of which are listed in table 3. Once again, there was remarkable consistency of abundance of the transcripts of these components among strains, which are very likely to be expressed stoichiometrically within the cell. The difference in expression between the two low-expression and the two high-expression isolates averaged to about 30%. As with the ubiquitin degradation component gene expression levels, this percent increase is low. It is, however, extraordinarily encouraging to have the power to find significant effects of this magnitude, for it is unlikely that effects on protein assemblies of such broad importance to the cell would ever be very large.

Metal Ion Transport

Gene expression levels did not differ much between isolates M1-2 and M7-8, but a number of differences in expression in metal ion transport genes could be found among the four isolates. Both M1-2 and M7-8 differed from M2-8 and M5-7 in that they had increased expression levels of the gene FTR1, an iron transporter, and of the gene FIT2 (fig. 2, node 1). FIT2 mutants have defects in iron-uptake (Protchenko et al. 2001). A zinc transporter, ZRT2, showed a slight (but significant) complementary difference (fig. 2, node 4 and fig. 8). M5-7 abundantly expressed two copper transporters, CTR1 and CTR3, compared with all other isolates (fig. 1, node 3 and fig. 9). The other copper transporter in yeast, CTR2, has the lowest affinity for copper ions and showed no statistically significant difference among strains. Finally, M1-2, M5-7, and M7-8 demonstrated abundant expression of three out of four genes associated with arsenic resistance (fig. 10). ARR1 is a transcription factor for ARR3 and is significantly differentially expressed; ARR3 is an active membrane transporter associated with arsenic resistance (Wysocki, Bobrowicz, and Ulaszewski 1997). Although credible intervals for ARR3 expression were broad, the estimated difference in expression by a factor of 3 to 5 is very large and consistent in profile to that observed in ARR1. ARR2, an arsenate reductase required for detoxification of arsenic, showed a similar profile among strains, whereas ARR4, which had been functionally categorized by similarity to the Escherichia coli ArsA ATPase component of its arsenite-antimonite efflux pump, showed no significant differences among strains. A prediction from these microarray results is that M2-8, when grown on agar plates laced with arsenic compounds, would show retarded growth. Comparison of the four isolates grown on control, 5 μM H3AsO4–supplemented, 10 μM H3AsO4–supplemented, and 15 μM H3AsO4–supplemented rich agarose plates visibly demonstrated this conditional fitness difference (data not shown). The strain M2-8, which meagerly expresses arsenate resistance genes ARR1, ARR2, and ARR3, grew poorly compared with the other strains on media containing high concentrations of H3AsO4.

Pseudohyphal Growth

Interestingly, the pseudohyphal determinant gene, PHD1, up-regulation of which is associated with a filigreed phenotype in progeny of M2-8 (Cavalieri, Townsend, and Hartl 2000), was estimated as abundantly expressed in M2-8 compared with the three other isolates (fig. 1, node 5 and fig. 11). Although credible intervals were nonoverlapping only between M1-2 and M2-8, this criterion is highly conservative. For instance, M2-8 had higher expression than M1-2 for PHD1 in 99% of the BAGEL full-posterior distribution of expression levels and had higher expression than M5-7 and M7-8 in 96% and in 97% of the BAGEL full-posterior distribution, respectively. In the framework of a classical hypothesis test, these correspond to P values for the hypotheses of M1-2, M5-7, and M7-8 being expressed at a higher level than M2-8. These one-tailed P values would be 0.01, 0.04, and 0.03, respectively, not accounting for multiple tests.

Transposable Elements

Among natural isolates, differential expression included the coding sequences of the yeast Ty element, a retrotransposon that may occur in 30 or more copies per genome. Figure 12 charts the expression levels measured for the ten Ty element ORFs that have nonoverlapping 95% credible intervals among the four isolates. Isolate M2-8 always showed the lowest expression of Ty messenger RNA, and each other natural isolate appeared to express Ty messenger RNA at a strain-specific level, with a range of a factor of 8. Note that, although the range among isolates for a given Ty open reading frame varied considerably, the relative levels of expression for each Ty open reading frame were remarkably consistent; an exception is YBL005W-B, which showed less variation among isolates. The consistency could reflect a common retroelement regulatory mechanism, but it is more likely to be an artifact of cross-hybridization due to the high degree of sequence similarity among Ty messenger RNAs. Regardless, there is extensive variation in the net expression of retroelement mRNA among natural isolates, due to variation in either regulation or copy number.

Discussion

Purely genetic differences in expression profile, where environment and developmental time are held constant, have been studied in a several ways. Galitski et al. (1999) have studied the effect of ploidy on gene expression. Despite dramatic phenotypic changes, including a 10% to 15% decrease in growth rate, their analysis showed just 17 genes that were significantly correlated or anticorrelated with ploidy. Strains used were all engineered from a single haploid strain. Several groups have systematically investigated genetic changes on a smaller scale. Compendia of expression profiles are being constructed for large-scale deletion and mutational analysis (Winzeler and Davis 1997; Hughes et al. 2000). Accessible databases for comparison of these profiles with current experiments are appearing; results from these and other studies intriguingly hint at multiple pleiotropic affects from seemingly small changes, and “give the lie to our tendency to discount biological phenomena that cannot, by themselves, cripple an animal or kill a cell” (Cho and Campbell 2000).

Whereas DNA microarray technology has enabled the analysis of global patterns of gene expression and revealed diverse networks of coordinated function, the genetic differences examined have been primarily differences between growth conditions or between mutant strains containing the limited genetic background of laboratory yeast (Mortimer and Johnston 1986). Although these studies explore the impact of environment and developmental state upon gene expression, their generality may be limited by having focused on the restricted and somewhat artificial range of genotypes available among laboratory strains. There is great potential for understanding molecular population genetics and evolution by the study of gene expression in yeast strains isolated from natural environments. In contrast, because dramatic differences in transcriptional profile may be observed between the progeny of a single natural isolate, it is unlikely that the continuous phenotype of gene expression level will be of much utility in phylogenetic reconstruction. This inference is consistent with the results of Dickinson (1987) and Thorpe and Dickinson (1988) who showed that constructions of the phylogeny of the Hawaiian picture-winged Drosophila, based on regulatory characters for the most part, do not reconstruct the known phylogeny. This, in turn, reinforces the conception that the introduction of regulatory variation is frequent.

Here we have shown variation in gene expression level in genes associated with amino acid metabolism, protein degradation, metal ion transport, growth phenotype, and transposable element activity. What can we take from this? Cavalieri, Townsend, and Hartl (2000) demonstrated extensive differences in expression profile of the progeny of M2-8. These differences, largely among genes of amino acid biosynthesis pathways, may represent, in part, segregating differences in how high levels of sulfite are dealt with. It is known from extensive enology experiments attempting to reduce hydrogen sulfide production in wine fermentation that supplementation or increased production of amino acids gives a nitrogenous substrate for processing of excess sulfite. The implication is that whatever difference is segregating in the M2-8 strain, the characteristics that make the two metabolic phenotypes distinct in offspring are both present in the parent strain, although to a considerably modulated degree and manifested in slightly different form. The segregating molecular phenotype appears to consist partly of two different ways of compensating for a homozygous characteristic of M2-8: the most dramatically differentially expressed gene among the four isolates, SSU1.

Transposable elements vary in expression among these four isolates and may generally, inadvertently, play an important role in introducing regulatory variation (Paigen 1986). Transposable elements have been implicated in extensive genetic rearrangement (Rachidi, Barre, and Blondin 1999). Moreover, the insertion of a transposable element near a gene frequently leads to changes in the level or developmental timing of expression of that gene (McClintock 1984).

Paigen (1986) reviews evidence that there is also regulatory polymorphism in translation and degradation rates. Here we demonstrate that the variation present in mRNA abundance is considerable, quite apart from downstream variation in translation and degradation. Furthermore, variation in downstream regulatory systems is implicit in the observed variation among strains in expression levels of genes in the ubiquitin pathway and the 20S proteasome (tables 2 and 3). These results encourage further exploration of variation at the proteomic level, as well as continued development of technological and statistical methodologies allowing inference of differences in mRNA and protein concentration of a factor below 2. The remarkable stoichiometric consistency of estimated expression levels among genes whose products make up the 20S proteasome demonstrates the potential to detect differences of low magnitude reliably using properly analyzed replication.

Such differences in expression of small magnitude may be of considerable importance as a kind of variation that may be selected upon during adaptive change. How much does the average gene differ, from isolate to isolate, in gene expression? Figure 13 shows the relative frequency at which various ratios in gene expression were observed among statistically significant differences in gene expression, based on all 12 pairwise comparisons of these four isolates. This histogram indicates the frequency of differences in gene expression as a function of the magnitude of difference. Note that the vast majority of differences in gene expression among natural isolates are below the twofold level. The fact that small differences in gene expression are more difficult to detect implies that many more genes are differentially expressed at these very low levels but were not detected. This reinforces the point that most of the variation in gene expression levels within a natural population is slight. Note also that much of that variation in gene expression may be composed of the coordinated cellular effects of a few genetic changes (Wilton et al. 1982).

Gene expression arrays tell only which genes are variable in expression, not why they are variable. It is possible (in fact, likely) that many of the differences in expression that are observed may have negligible effects on fitness, provided, for instance, that they have sufficiently small control coefficients in the metabolic pathways in which they participate (Hartl, Dykhuisen, and Dean 1985). On the other hand, some of the expression differences affect genes whose metabolic role makes them good candidates for affecting fitness. For a given gene, this is only a hypothesis and needs separate experimental support. A direct approach is through examination of genotype-phenotype correlations. Methods such as direct competition experiments with DNA-tagged strains may help to distinguish the circumstances in which such slight differences are neutral or adaptive. Retrospectively, indirect approaches may also prove useful, such as statistical tests of haplotype frequencies (Kreitman and Hudson 1991).

With regard to the role that regulatory variation can play in providing the raw material for adaptive evolution, Brem et al. (2002) demonstrate clearly the heritability of transcription, and Ferea et al. (1999) demonstrate rapid change in gene expression level in response to selection. These results, combined with the evidently considerable variation in gene expression level in natural populations disclosed here, argue for renewed attention to the role that regulatory variation plays during adaptive evolution.

Supplementary Material

Text tables of ratios for each gene in each hybridization, and BAGEL output files with expression levels and credible intervals for each gene in each sample, are available online at the Molecular Biology and Evolution Web site.

Present address: Department of Plant and Microbial Biology, University of California at Berkeley.

Pierre Capy, Associate Editor

Fig. 1.

Digital photograph of colonies and diagram of microarray hybridizations performed between natural isolates of S. cerevisiae from Montalcino, Italy. Each line represents one microarray comparison. The arrowhead points toward the strain labeled with the cyanine-5 fluorophore

Fig. 2.

Hierarchical clustering of significantly different genes by the average linkage algorithm on the log2 of the relative expression level estimated by BAGEL. Black or dark gray regions below the strain name represent genes meagerly expressed in that strain, whereas white or light gray regions below the strain name represent genes abundantly expressed in that strain. Note that use of relative expression levels from BAGEL obviates any need for color in this display

Fig. 3.

Expression levels of the genes in the lysine biosynthesis pathway, charted in the order of their action, among four natural isolates of Saccharomyces cerevisiae

Fig. 4.

Gene expression levels of three genes relevant to lysine biosynthesis

Fig. 5.

Gene expression levels of the amino acid biosynthesis genes, other than lysine biosynthesis genes, that have nonoverlapping credible intervals among four natural isolates of S. cerevisiae

Fig. 6.

Gene expression levels for two peptide transporters and two amino acid transporters that are differentially expressed among four natural isolates of Saccharomyces cerevisiae

Fig. 7.

Expression level of the sulfite exporter SSU1 among four natural isolates of Saccharomyces cerevisiae

Fig. 8.

Expression levels of FTR1, an iron transporter; FIT2, a facilitator of iron transport; and ZRT2, a zinc transporter, among four natural isolates of Saccharomyces cerevisiae

Fig. 9.

Gene expression levels of three copper transport genes in four natural isolates of Saccharomyces cerevisiae

Fig. 10.

Gene expression levels for four genes associated with arsenate/arsenite resistance, among four natural isolates of Saccharomyces cerevisiae

Fig. 11.

Gene expression levels for PHD1, the pseudohyphal growth determinant transcription factor, in four natural isolates of S. cerevisiae

Fig. 12.

Expression of transposable element open reading frames (ORFs) among four natural isolates. Note that the high sequence homology of these ORFs probably results in cross-hybridization. Thus, the repeated pattern of expression level across isolates observed here probably indicates the overall abundance of TE mRNA, rather than abundance of mRNA corresponding to each ORF

Fig. 13.

The frequency of significant differences in gene expression levels among four natural isolates of S. cerevisiae at various ratios of gene expression, when the ratio is tallied as that of the larger gene expression level over that of the smaller. Frequencies are derived from 433 genes that have nonoverlapping 95% confidence intervals in at least one pairwise comparison, for a total of 832 significant ratios

Table 1

Number of Genes Significantly Differently Expressed Among Four Natural Isolates.

aExpression higher.

bExpression lower.

Table 1

Number of Genes Significantly Differently Expressed Among Four Natural Isolates.

aExpression higher.

bExpression lower.

Table 2

Gene Expression Levels of Four Components of the Ubiquitin Degradation Pathway.

Note.—* Indicates that credible intervals are nonoverlapping with the credible interval of the lowest other expression level; ** indicates that credible intervals are nonoverlapping with the credible intervals of the lowest two other expression levels.

Table 2

Gene Expression Levels of Four Components of the Ubiquitin Degradation Pathway.

Note.—* Indicates that credible intervals are nonoverlapping with the credible interval of the lowest other expression level; ** indicates that credible intervals are nonoverlapping with the credible intervals of the lowest two other expression levels.

Table 3

Expression Levels of the Genes of the 20S Proteasome.

Note.—* Indicates that credible intervals are nonoverlapping with the credible interval of the lowest gene expression level; ** indicates that credible intervals are nonoverlapping with the credible intervals of the lowest two gene expression levels.

Table 3

Expression Levels of the Genes of the 20S Proteasome.

Note.—* Indicates that credible intervals are nonoverlapping with the credible interval of the lowest gene expression level; ** indicates that credible intervals are nonoverlapping with the credible intervals of the lowest two gene expression levels.

We thank Mario Polsinelli and Federico Sebastiani from the Department of Animal Biology and Genetics of the University of Florence for wine yeast strains, for useful information, and for hosting one of us (J.P.T.) as a visiting scholar. We would also like to thank Carlotta De Filippo of the Department of Pharmacology of the University of Florence for assistance during revision and also thank three anonymous reviewers for their thoughtful suggestions. A presentation of this work by J.P.T. was awarded the Walter M. Fitch prize for young investigators at the meetings for the Society for Molecular Biology and Evolution in Athens, Georgia in 2001.

Literature Cited

Avram, D., and A. T. Bakalinsky.

1996

. Multicopy FZF1 (SUL1) suppresses the sulfite sensitivity but not the glucose derepression or aberrant cell morphology of a grr1 mutant of Saccharomyces cerevisiae.

Genetics

144

:

511

-521.

Avram, D., and A. T. Bakalinsky.

1997

. SSU1 encodes a plasma membrane protein with a central role in a network of proteins conferring sulfite tolerance in Saccharomyces cerevisiae.

J. Bacteriol.

179

:

5971

-5974.

Becker, B., A. Feller, M. El Alami, E. Dubois, and A. Pierard.

1998

. A nonameric core sequence is required upstream of the LYS genes of Saccharomyces cerevisiae for Lys14p-mediated activation and apparent repression by lysine.

Mol. Microbiol.

29

:

151

-163.

Brem, R. B., G. Yvert, R. Clinton, and L. Kruglyak.

2002

. Genetic dissection of transcriptional regulation in budding yeast.

Science

296

:

752

-755.

Cavalieri, D., C. Barberio, E. Casalone, F. Pinzauti, F. Sebastiani, R. Mortimer, and M. Polsinelli.

1998

. Genetic and molecular diversity in Saccharomyces cerevisiae natural populations.

Food Technol. Biotechnol.

36

:

45

-50.

Cavalieri, D., J. P. Townsend, and D. L. Hartl.

2000

. Manifold anomalies in gene expression in a vineyard isolate of Saccharomyces cerevisiae revealed by DNA microarray analysis.

Proc. Natl. Acad. Sci.USA

97

:

12369

-12374.

Cho, R. J., and M. J. Campbell.

2000

. Transcription, genomes, function.

Trends Genet.

16

:

409

-415.

Dickinson, W. J.

1987

. Gene regulation and evolution. Pp. 181–202 in L. V. Giddings, K. Y. Kaneshiro, and W. W. Anderson, eds. Genetics, speciation, and the founder principle. Oxford University Press, New York.

Dickinson, W. J.

1991

. The evolution of regulatory genes and patterns in Drosophila.

Evol. Biol.

25

:

127

-173.

Eisen, M. B., and P. O. Brown.

1999

. DNA arrays for analysis of gene expression.

Methods Enzymol.

303

:

179

-205.

Eisen, M. B., P. T. Spellman, P. O. Brown, and D. Botstein.

1998

. Cluster analysis and display of genome-wide expression patterns.

Proc. Natl. Acad. Sci. USA

95

:

14863

-14868.

Ferea, T. L., D. Botstein, P. O. Brown, and R. F. Rosenzweig.

1999

. Systematic changes in gene expression patterns following adaptive evolution in yeast.

Proc. Natl. Acad. Sci. USA

96

:

9721

-9726.

Galitski, T., A. J. Saldanha, C. A. Styles, E. S. Lander, and G. R. Fink.

1999

. Ploidy regulation of gene expression.

Science

285

:

251

-254.

Hammer, M. F., and A. C. Wilson.

1987

. Regulatory and structural genes for lysozymes of mice.

Genetics

115

:

521

-533.

Hardwick, J. S., F. G. Kuruvilla, J. K. Tong, A. F. Shamji, and S. L. Schreiber.

1999

. Rapamycin-modulated transcription defines the subset of nutrient-sensitive signaling pathways directly controlled by the Tor proteins.

Proc. Natl. Acad. Sci. USA

96

:

14866

-14870.

Hartl, D. L., D. E. Dykhuisen, and A. M. Dean.

1985

. The limits of adaptation: the evolution of selective neutrality.

Genetics

111

:

655

-674.

Hughes, T. R., M. J. Marton, and A. R. Jones, et al. (22 co-authors).

2000

. Functional discovery via a compendium of expression profiles.

Cell

102

:

109

-126.

Kreitman, M., and J. M. Comeron.

1999

. Coding sequence evolution.

Curr. Opin. Genet. Dev.

9

:

637

-641.

Kreitman, M., and R. R. Hudson.

1991

. Inferring the evolutionary histories of the Adh and _Adh_-dup loci in Drosophila melanogaster from patterns of polymorphism and divergence.

Genetics

127

:

565

-582.

Laurie-Ahlberg, C. C.

1985

. Genetic variation affecting the expression of enzyme coding genes in Drosophila: an evolutionary perspective.

Isozymes-Curr. T. Biol. Med. Res.

12

:

33

-88.

Laurie-Ahlberg, C. C., A. N. Wilton, J. W. Curtsinger, and T. H. Emigh.

1982

. Naturally occurring enzyme activity variation in Drosophila melanogaster. I. Sources of variation for 23 enzymes.

Genetics

102

:

191

-206.

Macintyre, R. J.

1982

. Regulatory genes and adaptation—past, present, and future. Pp. 247–285. Evolutionary biology. Plenum Publishing Corporation, New York.

McClintock, B.

1984

. The significance of responses of the genome to challenge.

Science

226

:

792

-801.

Mortimer, R. K., and J. R. Johnston.

1986

. Genealogy of principal strains of the yeast genetic stock center.

Genetics

113

:

35

-43.

Paigen, K.

1979

. Acid hydrolases as models of genetic control.

Ann. Rev. Genet.

13

:

417

-466.

Paigen, K.

1986

. Gene regulation and its role in evolutionary processes. Pp. 3–23 in S. Karlin and E. Nevo, eds. Evolutionary processes and theory. Harcourt Brace Jovanovich, Orlando.

Paigen, K.

1989

. Experimental approaches to the study of regulatory evolution.

Am. Nat.

134

:

440

-458.

Protchenko, O., T. Ferea, J. Rashford, J. Tiedeman, P. O. Brown, D. Boststein, and C. C. Philpott.

2001

. Three cell wall mannoproteins facilitate the uptake of iron in Saccharomyces cerevisiae.

J. Biol. Chem.

276

:

49244

-49250.

Rachidi, N., P. Barre, and B. Blondin.

1999

. Multiple Ty-mediated chromosomal translocations lead to karyotype changes in a wine strain of Saccharomyces cerevisiae.

Mol. Gen. Genet.

261

:

841

-850.

Stern, D. L.

2000

. Evolutionary developmental biology and the problem of variation.

Evolution

54

:

1079

-1091.

Thorpe, P. A., and W. J. Dickinson.

1988

. The use of regulatory patterns in constructing phylogenies.

Syst. Zool.

37

:

97

-105.

Townsend, J. P., and D. L. Hartl.

2002

. Bayesian analysis of gene expression levels: statistical quantification of relative mRNA level across multiple treatments or samples.

Genome Biol.

3

:

research0071.1

-0071.16.

Wang, D., J. L. Marsh, and F. J. Ayala.

1996

. Evolutionary changes in the expression pattern of a developmentally essential gene in three Drosophila species.

Proc. Natl. Acad. Sci. USA

93

:

7103

-7107.

Wilson, A. C.

1985

. The molecular basis of evolution.

Sci. Am.

253

:

164

-173.

Wilson, A. C., S. S. Carlson, and T. J. White.

1977

. Biochemical evolution.

Ann. Rev. Biochem.

46

:

573

-639.

Wilton, A. N., C. C. Laurie-Ahlberg, T. H. Emigh, and J. W. Curtsinger.

1982

. Naturally occurring enzyme activity variation in Drosophila melanogaster. II. Relationships among enzymes.

Genetics

102

:

207

-221.

Winzeler, E. A., and R. W. Davis.

1997

. Functional analysis of the yeast genome.

Curr. Opin. Genet. Dev.

7

:

771

-776.

Wysocki, R., P. Bobrowicz, and S. Ulaszewski.

1997

. The Saccharomyces cerevisiae ACR3 gene encodes a putative membrane protein involved in arsenite transport.

J. Biol. Chem.

272

:

30061

-30066.

Zuckerkandl, E.

1963

. Perspectives in molecular anthropology. Pp. 243–272 in S. L. Washburn, ed. Classification and human evolution. Wenner-Gren Foundation for Anthropological Research, New York.

Society for Molecular Biology and Evolution