We wished to study the impact of growth rate on the total complement of mRNA molecules, proteins, and metabolites in S. cerevisiae, independent of any nutritional or other physiological effects. To achieve this, we carried out our analyses on yeast grown in steady-state chemostat culture under four different nutrient limitations (glucose, ammonium, phosphate, and sulfate) at three different dilution (that is, growth) rates (D = μ = 0.07, 0.1, and 0.2/hour, equivalent to population doubling times (T_d) of 10 hours, 7 hours, and 3.5 hours, respectively; μ = specific growth rate defined as grams of biomass generated per gram of biomass present per unit time). We then looked for changes that correlated with growth rate under all four nutrient-limiting conditions, using principal components analysis (PCA; see Materials and methods). Trends that appear in all four nutrient-limited series, including carbon-limited cultures with equivalent glucose concentrations, cannot be attributed to variations in residual substrate concentrations (for example, different levels of glucose repression). Instead, they must be due to intrinsic growth-rate-related processes.

Gene expression at the mRNA level was investigated by transcriptome analysis using Affymetrix hybridization arrays. Proteomic studies were performed using isotope tags for multiplexed relative and absolute quantification (iTRAQ) 3435. In this case, the four tags and labeling schema applied (see Materials and methods) allowed us to test and compare the proteomes of cells grown at μ = 0.1/hour (T_d = 7 hours) with those of cells grown at μ = 0.2/hour (T_d = 3.5 hours) for all four nutrient limitations. We were able to detect and quantify a significant proportion of the yeast proteome (around 700 proteins per nutrient-limiting condition; 1,358 proteins in total; see Materials and methods). For the metabolome, which is the closest genomic level to the cell's phenotype 3637, gas chromatography coupled to time-of-flight mass spectrometry (GC/TOF-MS) was used to analyze the complement of intracellular and extracellular metabolites, that is, the endo- and the exometabolomes 3839.

Principal components analyses (PCA) of transcriptome, proteome, and endo- and exometabolome data showed clear growth-rate-associated trends for all omic levels (Figure 1). In the case of the endo- and exometabolomes, these trends are clearly revealed after independent analysis of the carbon-limited and carbon-sufficient datasets (see Figure 1d,f). This is because, in contrast to all other nutrient-limited steady states, the endo- and exometabolomic profiles from cells in glucose-limited steady-state cultures showed no clear growth-rate trend. We infer from this that yeast cells are well-adapted to growth under carbon-limited conditions and are able to adjust the individual fluxes through their metabolic network to regulate overflow metabolism whatever overall flux is imposed by the external supply of carbon substrate. This result is congruent with our data from category 2 experiments (D. Delneri and S.G.O., unpublished work) in which we have examined the effect that reducing the copy number of individual genes in diploid cells has on flux by performing competition experiments, in chemostat cultures, between yeast strains heterozygous for individual gene deletions.

For all three levels of 'omic analysis, the data show a clear distinction between carbon-limited and carbon-sufficient cells (Figure 1). Once the data from the carbon-limited steady states have been excluded, both the endometabolome and the exometabolome data from all three carbon-sufficient cultures show a clear and consistent growth-rate trend (compare Figure 1c,e with 1d,f). In addition, for the endometabolome data, the second principal component separates the ammonium-limited cells from those grown under phosphate and sulfate limitation (Figure 1d).

Figure 1a shows that the transcriptome data from nitrogen-limited cells at the lowest growth rate studied (0.07/hour) do not obey the general growth-rate trend. Uniquely among all the cultures that we analyzed, cells from these cultures had a pseudohyphal, rather than a budding, growth pattern; these data should allow us to define those genes whose expression is specifically associated with filamentous growth. We did not examine the proteome at μ = 0.07/hour and so do not know whether this difference is reflected at the protein level. However, the proteomic data from all steady-state cultures at μ = 0.1/hour and 0.2/hour show the same clear discrimination between carbon-limited and carbon-sufficient cells and the same growth-rate-associated trend as was found with the metabolome and transcriptome data. The fact that all 'omes' studied display a growth-rate-associated trend suggests a multilevel control underlying global regulation of cell growth, and we now examine these levels in some detail.

Hybridization-array technology was used to determine how the levels of gene transcripts changed with both flux (growth rate) and nutrient environment. While the transcriptomes of cells grown under each of the four nutrient-limiting conditions have their own characteristics (see Additional data files 1 (Figures S1 and S2), 2 (Tables S1 and S2), and 3), there is a common qualitative and quantitative response to increasing growth rate that is independent of the specific nutrient limitation (see Figure 1a, and Additional data file 1 (Figures S3 and S4)).

We performed an analysis of covariance (ANCOVA) in order to identify those genes whose transcription was significantly and consistently upregulated or downregulated with growth rate in all four nutrient-limitation conditions studied (see Additional data file 1 (Figure S3)). These genes were ranked by estimates of false discovery rate (FDR), in this case the q-value 40 of the ANCOVA model (obtained from the p value, after multiple testing correction; see Additional data file 4), which represents the relative significance in the (condition-independent) change in gene expression with growth rate. Taking these q-values, we applied a cut-off of 5% (q = 0.05 40; see Materials and methods). This produced a set of 493 genes whose expression is significantly upregulated with increasing growth rate (q < 0.05; see also Additional data file 4), and 398 genes that exhibited significant and concomitant downregulation with increasing growth rate, independent of the culture conditions (see Additional data files 1 (Figure S4) and 2 (Tables S3 and S4)).

Essential genes, that is, genes whose deletion results in a failure to grow on rich glucose-containing medium 4142, are statistically overrepresented in the list of genes significantly upregulated with growth rate (161 out of 493 (32.6%); the fraction of all yeast genes that are essential is around 17%), whereas they are significantly underrepresented in the downregulated list (22 out of 398 (5.5%, again compared to 17%)). The proportion of essential open reading frames (ORFs) in the downregulated set (5.5%) is significantly different from the proportion of essential ORFs that we find not to be subject to growth-rate control (16.8%). In fact the fraction of essential ORFs in this non-growth-regulated set is indistinguishable from the proportion of all yeast ORFs that are essential to growth (16.6%).

Despite the fact that genes that are downregulated with increasing growth rate are rarely essential on rich medium 4142, the central role of all growth-regulated genes in cell growth is confirmed by independent studies on deletion mutants. This applies to both the essential and the non-essential genes in both the up- and downregulated sets (Figure 2a). Thus, null mutations in many of the genes that we have identified as growth-regulated have been reported to either be lethal or produce a severe growth defect (84.0% in the upregulated set; 64.6% in the downregulated set) 4142 (see Additional data file 2 (Tables S3 and S4)). In all, our studies have revealed the importance of nonessential genes whose expression is growth-rate regulated in determining whether yeast can grow at normal rates. This applies to genes whose expression is downregulated with increasing growth rate, as well as those under positive growth-rate regulation.

From all these studies, a significant number of genes (891; 15% of the protein-encoding genes in the genome) have their transcript levels determined by growth rate (Figure 2a). While many of these genes (198, 22.2%) correspond to ORFs of so far unknown function (Figure 2a; see also Additional data file 2 (Tables S3 and S4)), according to Affymetrix (12 July 2006) and Gene Ontology (GO) annotations 43, an examination of the functions determined by the remainder is instructive. Using two different GO analysis tools (GoMiner 44 and GenMAPP 45; see Additional data files 1 (Figures S5-S16) and 2 (Tables S5-S11)) we showed that the 435 genes of known function that are upregulated with growth rate (see Figure 2a and Additional data files 1 (Figure S4) and 2 (Table S3)) include a significant proportion whose products are involved in the biological processes of translation initiation, ribosome biogenesis and assembly, protein biosynthesis, RNA metabolism, nucleobase, nucleoside, nucleotide and nucleic acid metabolism, nucleus import and export and proteasome function (see Additional data files 1 (Figures S5 and S11) and 2 (Tables S3 and S5)). The corresponding analysis of GO molecular functions for the same gene set showed the following to be overrepresented: translation initiation factor activity and nucleic acid (RNA) binding, structural constituent of ribosome activity, ligase activity forming aminoacyl-tRNAs and DNA-directed RNA polymerase activity (see Additional data files 1 (Figures S6 and S12) and 2 (Table S6)). At the level of cellular components, GO studies indicated that the most representative upregulated processes occur in a variety of subcellular compartments (cytosol, exosome, and nucleus) and complexes (for example, eukaryotic translation initiation complexes, nucleolus, ribosome subunits, and the proteasome core complex; see Additional data files 1 (Figures S7 and S13) and 2 (Table S7)). For a comprehensive analysis of processes upregulated with increasing growth rate, see Additional data file 5.

GO analysis of the set of 258 genes of known function whose transcription was significantly downregulated with increasing growth rate (see Figure 2a and Additional data files 1 (Figure S3) and 2 (Table S4)) shows that a high proportion of these genes correspond to the following biological processes: response to external stimulus, cell communication and signal transduction, autophagy, homeostasis, response to stress, vesicle recycling within Golgi (see Additional data files 1 (Figures S8 and S14) and 2 (Table S9)). The most overrepresented GO molecular function categories for this gene set correspond to a variety of catalytic, signal transduction, transcription regulator, and transport activities. These include receptor signaling protein activity, protein kinases, phosphotransferase, oxidoreductase and ATPase activity coupled to transmembrane movement of ions, and phosphorylation mechanisms (see Additional data files 1 (Figures S9 and S15) and 2 (Table S10)). At the level of cellular component, downregulated processes occur at the level of the plasma membrane, the vacuole, and the repairosome (see Additional data files 1 (Figures S10 and S16) and 2 (Table S11)). Although essential genes are under-represented in this list (22 out of 398; see Additional data file 2 (Table S4) and the Saccharomyces Genome Database 42), the fact that 64.6% of the downregulated genes have been reported to result in growth defects or inviability in gene deletion studies (see Additional data file 2 (Table S4) and 42) points to a crucial role of these genes in growth-related processes that has yet to be elucidated. All of the 22 essential genes in this set are of known function, but only 11 of them have been reported previously as being directly related to cell growth and maintenance. For a comprehensive analysis of the role of most relevant downregulated processes regulating cell growth at the transcriptional level, see Additional data file 5.

Genes that are downregulated with increasing growth rate are probably involved in maximizing the efficient utilization of cellular resources at each different growth rate and culture condition, particularly when nutrients are scarce. Our data indicate that this is a poorly understood aspect of the cell's economy since a significant number of these genes (140/398; 35.2%) are of as-yet-undetermined function. This is despite the fact that nutrient scarcity is likely to be a common circumstance in the organism's natural environment 46. Among the genes of known function that are upregulated at low growth rates are those involved in mobilization and storage of available resources at the level of the vacuole (see Additional data file 1 (Figure S20)). Another interesting example of genes that are upregulated at low growth rates are those involved in autophagy (see Additional data file 1 (Figure S21)). Autophagy is a major system of bulk degradation of cellular components. It participates in the coordinate degradation of cytoplasmic components, including proteins, large complexes and organelles whose turnover is important in the control of cell growth. Autophagy mediates the shrinkage of the ribosome pool, thus slowing cell growth when nutrients are limiting 47.

Autophagy in yeast has been reported to be a TOR-mediated response to nutrient starvation 48, and we have demonstrated previously the induction of autophagy genes in stationary phase 19. Autophagy genes are well conserved from yeast to mammals, suggesting that it is a fundamental activity of eukaryotic cells, being implicated in processes such as homeostasis, development and differentiation 47. Other genes that are upregulated at low growth rates are those encoding specific transcriptional repressors whose action results in the activation of alternative routes for the assimilation of substrates and/or as an adaptation to the environment.

In all, the data on the downregulated genes present a picture of the yeast cell at low growth rates activating pathways involved in the response to external stimuli, maintenance of homeostasis, vacuolar transport and storage, and autophagy; the whole being directed towards a more efficient use of scarce resources. Finally, we have found that genes that were annotated previously as being involved in 'response to stress' 424950 are upregulated at low growth rates. Moreover, we have confirmed these findings at the proteome level (see proteomic studies (Table 1)). This demonstrates that a large part of what others have termed the 'generalized stress response' may more properly be viewed as a slow-growth response.

A high percentage of the proteins encoded by the up- and downregulated genes are highly conserved in a variety of 'model' eukaryotes (Ashbya gossypii, Caenorhabditis elegans, Arabidopsis thaliana, Drosophila melanogaster and Homo sapiens) 5152, which points to the existence of an essentially conserved 'core' protein machinery governing cell growth in the Eukarya. Thus, 75% of the protein products of yeast genes upregulated with growth rate have orthologs in humans, whereas 52% of the downregulated set have human orthologs (which is not significantly different to the figure of 48% for all S. cerevisiae proteins 51; see Figure 2a and Additional data file 2 (Tables S3 and S4)). Many of these proteins are built into complex machines 53. Proteins encoded by the upregulated genes participate in a large number of interactions with each other (876 interactions as compared with 287 expected by chance), whereas those encoded by the downregulated genes rarely interact with one another (89 compared with the 193 expected by chance; see Additional data files 2 (Tables S12 and S13) and 4).

The TOR signal transduction pathway is a central controller of the eukaryotic cell, sensing cellular environment and linking nutrient assimilation with translation initiation and ribosomal protein synthesis to control cell growth 3433545556. Many genes responsible for central growth processes (for example, translation initiation, ribosome biogenesis, autophagy, stability of biosynthetic components) are regulated at the transcriptional level (see Additional data file 2 (Tables S3 and S4)) and are under the direct or indirect control of TOR 3233 (see Additional data file 1 (Figure S22)). The exact mechanisms by which the TOR pathway controls these processes are not known, but appear to be mediated (at least, in part) by GATA-type, zinc-finger and forkhead transcription factors 323357585960. We decided to test the generality of the hypothesis that TOR, more specifically the TOR signaling branch that mediates temporal control of cell growth (TORC1) complex 32, is the major regulator of yeast gene expression in response to nutrient availability, and hence of growth rate 3313233. To do this, we examined the impact of rapamycin, a specific inhibitor of the TORC1 complex 32, and widely used to elicit the TOR control response 3261, on the yeast transcriptome 14.

The results of this examination should be approached with caution for two reasons. First, few inhibitors are completely specific in their action and thus our analysis is likely to be complicated by side-effects of rapamycin on processes other than TOR action. Second, as the addition of the inhibitor would necessarily disturb the steady state of a chemostat culture, we performed this experiment in batch. We have shown previously that the use of batch culture introduces a number of confounding variables to transcriptome analyses that are avoided by the use of chemostats 1419. Thus, it may be predicted that the rapamycin-inhibition experiment would show more genes affected than were found to be subject to growth-rate control in our chemostat studies. This, indeed, proved to be the case (Figure 2b). Remarkably, the rapamycin and growth-rate data showed more than 70% of growth-rate-regulated genes to be members of the TOR-responsive sets. We found 397 growth-rate upregulated genes to be downregulated by rapamycin, and 249 genes downregulated by growth rate were upregulated in response to the drug. Thus, 646 growth-rate-regulated genes (72.5%) appear to be specifically controlled by TOR (Figure 2b; see also Additional data files 1 (Figure S23) and 2 (Tables S15 and S16)). Our studies are also in good agreement with previous transcriptional studies on the effect of rapamycin on yeast cultures, showing a characteristic global response, with translational initiation, aminoacyl-tRNA synthetases, RNA polymerases, ribosome biogenesis and proteasome subunits among the most significantly affected biological processes (see Figure 2b and Additional data file 2 (Tables S15-S17) and 6162). These are key processes in which our sets of growth-rate-regulated genes are involved.

In our results, none of the genes specifying the components of the TORC1 complex 3263 appears significantly regulated at the level of transcription (see Additional data file 1 (Figures S24 and S25)), in agreement with previously reported studies (SGD; ORF expression connection studies 42). Evidence is accumulating that post-transcriptional mechanisms play an important role in the global regulation of cell growth 246465 (see also the section on translational control, below). As an example, many genes reported to be involved in control of cell size or coordination between cell growth and division 9 do not appear regulated at the transcriptional level (see Additional data files 2 (Tables S3 and S4) and 5), showing that it is important to extend these studies to the proteomic level.

Most global gene-expression studies have been entirely at the transcriptome level and often assume that changes in transcript levels should correlate with changes at the protein level. However, there is ample evidence that this is a dangerous assumption 212223246566676869. We extended our study to the proteome level using iTRAQ 3435, covering a significant proportion of the yeast proteome (around 700 proteins per nutrient-limiting condition; 1,358 in total; see Materials and methods). For example, we examined the differences in protein levels (proteomic signatures) between cells growing at μ = 0.1/hour and those growing at 0.2/hour under carbon limitation (Figure 3 and Additional data file 2 (Table S18)), and found a number of proteins and biological processes to be significantly up- and downregulated under these conditions (Table 1 and Additional data file 2 (Tables S19 and S20)). Remarkably, as with the transcriptome profiles, these proteomic signatures appear to be characteristic for each nutrient-limiting condition, but there is also a common pattern that represents the proteomic response to a growth-rate shift from μ = 0.1 to 0.2/hour (see Figure 3a and Additional data files 1 (Figure S26) and 2 (Tables S18 and S21)). Relative changes in proteome levels of proteins participating in relevant biological processes are shown in Figure 3b. Again, in common with the transcriptome data, most of the changes in protein levels lie in a range between a less than twofold decrease and a less than twofold increase (Figure 3a and Additional data file 1 (Figure S26)). Similar analyses (that is, ANOVA) to those performed on the transcriptome data can be applied to identify groups of proteins that are consistently and significantly up- or downregulated with growth rate (see Additional data file 4).

Among the groups of proteins whose levels appear consistently up- or downregulated with growth irrespective of the specific nutrient limitation (see Figure 3 and Additional data files 1 (Figure S26) and 2 (Tables S22 and S23)) are proteins of the translational machinery (for example, translation initiation and elongation factors, ribosomal proteins, aminoacyl-tRNA synthetases), enzymes involved in methionine and methyl cycle metabolism, and regulatory enzymes of amino-acid and other relevant biosynthetic pathways. Selected groups of proteins are marked in color in Figure 3a. As a relevant example, proteomic studies reveal different responses in the levels of the two S-adenosylmethionine synthetases, Sam1p and Sam2p (see Figure 3a and Additional data file 1 (Figure S26)). This, and the fact that the SAM2 gene was significantly upregulated at the transcriptional level (Additional data file 2 (Table S3)), are in accordance with previous reports 70.

Finally, nutrient-independent changes in levels of metabolic enzymes (see Figure 3a; the most relevant are included in Additional data file 2 (Table S24)) with growth rate will be of particular importance for the elucidation of the yeast cell's strategies for the control of central metabolic fluxes during cell growth, and for the identification of groups of metabolic enzymes consistently up- and downregulated at the protein level (for example, amino-acid biosynthetic enzymes; Table 2). These studies have direct implications for the design of new comprehensive metabolic engineering strategies, and will be discussed in the section below on metabolic control, where (for example) the role of the Sam1p and Sam2p isoenzymes is considered.

Because our transcriptome and proteome data had been obtained from the same samples of cells from chemostat cultures in steady state at growth rates of both 0.1 and 0.2/hour, and as these data had been normalized and statistically analyzed in the same way, we were able to make a realistic determination of the congruence between the level of any gene transcript and its cognate protein product(s). Example results are presented in Figure 4 for the glucose-limited steady states. Overall, the correlation coefficients (r) for each nutrient-limiting condition (C, N, P and S limitation) lie between 0.4 and 0.7, indicating only a moderate global congruence between transcript and protein levels (see Additional data file 6), in agreement with some previous studies 656667686971. The fact that mRNA changes do not generally correlate with protein changes suggests a widespread role for post-transcriptional mechanisms in the control of yeast's growth rate (see below). Most transcripts show a relative change in their level, between both growth rates of 0.1/hour and 0.2/hour, that is within a twofold range up and down, and the same is true for their cognate proteins. However, there are a number of transcript-protein pairs that are significant outliers, cases in which changes in transcript levels do not result in comparable changes at the protein level (for example, ADH4/Adh4p and ADO1/Ado1p in Figure 4); examples of these outliers are shown more clearly in Figure 5, and are discussed in the following section.

A number of post-transcriptional mechanisms might be involved in modulating the cellular concentration of a given protein relative to that of the mRNA species that encodes it. These include mRNA recruitment from the nucleus and p-bodies, polyadenylation states, level of polysomal occupancy per transcript, and rates of protein degradation 21222324727374. To encompass all of these mechanisms of translational control and quantify their overall effect, we define the effective 'translational control efficiency' (Trlc Eff_i) of a given messenger RNA in terms of its P/R ratio [protein_i]/[mRNA_i] (see Materials and methods and Additional data file 7), and show that the ratio of relative change in the level of a protein to the relative change in its cognate mRNA (obtainable from proteome-transcriptome studies; see above) is equal numerically to the ratio of relative changes in translational control efficiencies between the two conditions studied (see Materials and methods and Additional data file 7).

By this means, and on a genome-wide scale, we can quantify the relative changes in the overall translational control efficiencies of mRNA molecules corresponding to a shift from μ = 0.1 to 0.2/hour (that is, a doubling in specific growth rate). The results are presented in Figure 5 (for just the carbon-limited steady state) and in Additional data file 8. The pattern of changes suggests that the translational control efficiencies of particular mRNAs are modulated selectively in order to fine-tune protein activities and metabolic fluxes of relevant biological processes during cell growth (Figure 5a,b). The pattern of changes in translational control efficiencies is dependent on the specific nutrient-limiting condition, with most transcripts showing a less than twofold change (up or down) in their translational efficiencies, but a few undergo much larger relative changes (Figure 5a, see also Additional data files 1 (Figure S27) and 8).

This metric of the relative change in translational control efficiency allowed us to make a quantitative estimate of the relative contribution of post-transcriptional control mechanisms to a change in growth rate. For each nutrient-limiting condition, more than 35% of all transcripts were found to change their translational efficiency to a significant (greater than 20%) extent. Further studies, including analysis of post-translational modifications across the proteome (for example, phosphorylation and glycosylation), will provide a more complete picture of the role of post-transcriptional control during cell growth.

From all these data, we were able to extract groups of transcripts whose translational control efficiencies are consistently up- or downregulated with growth rate, irrespective of the limiting nutrient. Transcripts in this category include those encoding components of the translational machinery, enzymes subject to covalent or allosteric regulation that are involved in amino acid and other biosynthetic pathways, and regulatory proteasome subunits. Selected cases are marked in bold in Figure 5a and summarized in Additional data file 2 (Table S25). As an interesting example, the relative level of the transcript of CPA1 (encoding the small subunit of the multimeric enzyme carbamoyl phosphate synthetase (CPSase) in the arginine biosynthetic pathway) does not change with growth rate (see Additional data file 2 (Table S3)), but the overall efficiency with which this mRNA is translated goes up significantly with growth rate (see Figure 5a, and Additional data file 8). Although CPSase activity has been found to be subject to regulation at the transcriptional, translational and metabolic levels 75767778, under the specific conditions tested (synthetic medium under nutrient-limited conditions, with ammonium as sole nitrogen source), it appears to be regulated mainly at the translational level.

How are the metabolic fluxes characteristic of an increase in the rate of biomass accumulation actually controlled? To what extent are these fluxes regulated by gene expression (enzyme expression levels) or by metabolic regulation? To answer these questions, the quantitative proteomic data must be integrated with those on the metabolome. This is, without doubt, the most difficult challenge in data integration that exists in functional genomics or systems biology. To a large extent, it is because the metabolome, in contrast to the transcriptome and the proteome, has no simple, direct connection to the genome 79. We have recently developed statistical approaches with which to integrate transcriptome data with those for a small number of key metabolites (for example, glucose, ethanol, CO₂) 80, but we have yet to extend this to the entire metabolome. This is a field in which many different strategies are likely to be required and, indeed, are starting to be developed - for instance, metabolic network topology 81.

In the current study, we used the ANOVA analysis applied to the iTRAQ proteomic data to identify proteins whose levels were consistently up- or downregulated with growth rate (see Figure 3 and Additional data file 2 (Tables S22-S24)). This analysis highlighted two growth-related metabolic processes: the coupling of carbon and nitrogen fluxes towards the synthesis of amino acids, for example, glutamine and arginine (Figure 6); and the flux of methionine and S-adenosylmethionine (SAM through the methyl cycle 8283 (Figure 7).

The diploid Saccharomyces cerevisiae strain FY1679 (MATa/MATα ura3-52/ura3-52 leu2-1/+trp1-63/+his3-D200/+ho::kanMX4/ho::kanMX4 was used for all the experiments. Conditions for chemostat cultivation in a mineral medium under C, N, P and S nutrient limitation have been described previously 14120 and are given in Additional data file 4). For the rapamycin study, a culture growing at mid-exponential phase was divided into two. Rapamycin (200 ng/ml) was added to one half, and the drug's solvent to the other, as the control. Samples were taken at 0, 1, 2 and 4 h after treatment (see Additional data file 4).

Biomass was harvested and total RNA extracted as previously described 14. For growth-rate dependence studies, the microarray experimental design consisted of four nutrient-limiting conditions grown at three growth rates. These 12 conditions were analyzed in quadruplicate using Affymetrix Yeast Genome S98 GeneChip oligonucleotide arrays (Affymetrix, Santa Clara, CA). For the rapamycin study, samples were analyzed in duplicate using YG_S98 arrays (Affymetrix). Arrays that passed outlier data-quality assessment using dChip software 121 were normalized with RMAExpress 122. For each probe set the coefficient of variation (CV) was calculated for each condition (%CV = (standard deviation/mean) × 100). The mean CV (variance in transcriptional studies) was calculated, being in the range between 2.4-3.9% in all cases. The data were submitted in MIAME-compliant format to the ArrayExpress public repository 123 under accession numbers E-MEXP-115 (growth-rate studies) and E-MAXD-4 (rapamycin studies).

Statistical analyses (PCA, t-tests, ANOVA/ANCOVA and false-discovery rate estimation) to identify significantly regulated genes were performed with Matlab (MathWorks, Natick, MA) 124, Q-value 40, GeneSpring (Agilent Technologies, Santa Clara, CA) 125 and maxdView software (available from 126); see Additional data file 4 for details. Transcriptome results were validated by comparison of the patterns of gene expression of genes from equivalent chemostat experiments using an independent macroarray technique 19 and, in addition, triplicate analyses of eight genes were performed by QRT-PCR.

Yeast were grown in chemostat culture as described above. Cycloheximide (final concentration 100 μg/ml) was added to 1-liter steady-state chemostat cultures. Cells (10 × 80 ml) were harvested and centrifuged at 5,000 rpm for 5 minutes at 4°C. The pellet was resuspended in 1 ml ice-cold double-distilled water and transferred to a 1.5-ml microcentrifuge tube. Cells were repelleted by centrifugation at 10,000 rpm, the supernatant discarded, and the yeast pellet frozen in dry ice and stored at -80°C.

Sampling, quenching, and efficient extraction for endo- and exometabolome analyses of hundreds of intra- and extracellular metabolites, including the main glycolytic intermediates, nucleotides, pyridine nucleotides, and organic acids (for example, pyruvate, citrate, and succinate) were carried out as described previously 131. Direct quenching of the culture was performed by fast sampling of a volume of culture equivalent to 30 mg dry weight 131; extracts and culture supernatants were stored at -80°C. Samples were prepared for MS immediately before the analysis was carried out.

We sought to minimize sources of systematic error, first by careful experimental design (see above) and the application of the following strategies: use of steady-state chemostat cultures ensuring carefully controlled environmental conditions at each constant growth rate 14151617; avoidance of prolonged cultivation studies (that is, keeping to below 60 generations) to eliminate risks of strain variability and/or mutational effects; fast sampling and growth-arrest methods avoiding environmental disturbances for proper transcriptome, proteome and metabolome analyses; for each omic analysis, all samples were processed by the same specialist researcher; careful analytical strategies and normalization methods (see Additional data file 4).