We used LC-MS/MS to analyze >180 compounds, representing a substantial fraction of the >600 compounds of the known yeast metabolome (Herrgård et al., 2008
). Of the analyzed compounds, reliable relative quantitation (i.e., fold change across biological conditions), was obtained for ~100 of these (see Supplemental Table S1). These compounds include most central carbon metabolites, amino acids, and nucleotides.
Intracellular metabolites were extracted from chemostat cultures using two methods. One involved mixing culture medium containing cells directly into cold methanol, followed by isolation of the cells by centrifugation and subsequent extraction of the cell pellet (de Koning and van Dam 1992
). The other involved isolation of the cells by vacuum filtration followed by quenching and extraction of the filter-trapped cells. In both methods, acetonitrile:methanol:water (40:40:20) was used as the extraction solvent, as for most compounds it gave equal or greater yields to methanol:water (the previous literature standard; Maharjan and Ferenci 2003
; Villas-Bôas et al., 2005
). In particular, acetonitrile:methanol:water was substantially more effective at extracting nucleotides from filter-trapped cells (Supplemental Figure S1A and Supplemental Table S2). Although the filtration method gave larger absolute signals for nucleotide triphosphates (Supplemental Figure S1B), relative metabolite concentration changes across biological samples (i.e., mass spectrometry signals normalized to the volume of the extracted cells and then to the geometric mean across all chemostat conditions) were robust to the sampling method (Supplemental Figure S2). Because both methods gave comparable relative quantitation, the data were pooled in all subsequent analysis. A complete summary of the pooled data, expressed as log2
ratios of the relative metabolite concentration changes, is provided in Supplemental Dataset 1. Similar data clustering and quantitative conclusions are obtained from independent analysis of either the methanol-quenched cells or the vacuum-filtered cells.
Whole Metabolome View
We used hierarchical clustering (Pearson correlation; Eisen et al., 1998
) of the metabolites to view the relative differences and similarities in metabolite concentrations among conditions (, for alternative color scheme see Supplemental Figure S2). This display suggests several general conclusions. First, metabolites belonging to the same class, such as pyrimidine intermediates, amino acids, or tricarboxylic acid (TCA)-cycle intermediates, formed remarkably consistent clusters. Second, the levels of almost all metabolites depended strongly on the identity of the limiting nutrient, with profound differences across conditions that for several compounds even exceeded 100-fold. Third, the differences between nutrients were most pronounced at the slowest growth rate, at which the limitation was most stringent.
Although the concentrations of only a few measured metabolites correlated with growth rate regardless of the nature of the nutrient limitation, within each particular nutrient limitation, many metabolite levels did change consistently with respect to the growth rate. Generally, the compounds that showed the strongest growth rate effect were directly related to the limiting nutrient. For example, most amino acids were depleted in nitrogen limitation, especially at slow growth rates, and rose with increasing growth rate. Similarly, nucleotide triphosphates were depleted in phosphorus limitation and rose with faster growth rate. The inverse was true for nitrogenous bases and nucleosides, whose concentrations were elevated in phosphorus limitation and declined with faster growth rate. The general trend toward amino acid depletion in nitrogen limitation and nucleotide triphosphate depletion in phosphorus limitation is consistent with previous literature showing that nitrogen limitation restricts protein synthesis and that phosphorus limitation restricts nucleic acid synthesis in Enterobacter aerogenes
(Cooney and Wang 1976
Identification of Growth-limiting Intracellular Metabolites
As a prelude to searching for intracellular metabolites that might limit growth, we confirmed that the growth rate was related to the extracellular concentration of the limiting nutrient. As expected, growth rate was a monotonic function of the concentration of the limiting nutrient, as shown for glucose, uracil, and leucine limitation in Supplemental Figure S3. The estimated values for μmax
(Supplemental Figure S3), derived from the Michaelis–Menten relationship (Eq. 1
), approximate the empirically observed maximum growth rate in exponential batch culture (Pronk 2002
For depletion of an intracellular metabolite to limit growth, we reasoned that its concentration must 1) be uniquely low in the nutrient condition where it is growth-limiting; and 2) within that condition, rise when the limitation is partially relieved, i.e., with increasing growth rate. In each limitation regime, we identified several metabolites that qualitatively met these criteria. Some examples include glutamine, a central nitrogen assimilation metabolite, in nitrogen limitation; ATP, the free energy currency and RNA building block, in phosphorus limitation; pyruvate, the glycolytic end product, in carbon (glucose) limitation; and UTP, the RNA building block, in uracil limitation. These four examples are illustrated in . In each case, there is a well-understood connection between the limiting nutrient and the potentially growth-limiting intracellular metabolite.
Figure 2. Examples of metabolites that are potentially limiting growth under glucose limitation, ammonium limitation, phosphate limitation, and uracil limitation (from top to bottom). Metabolite concentrations are plotted on a log2 scale and mean-centered as per (more ...)
To systematically identify the full set of potential growth-limiting metabolites, we quantitatively assessed the impact of nutrient condition and growth rate on intracellular metabolite concentrations. The data for each metabolite–nutrient condition pair were fit to the following simple model:
is the metabolite's concentration in yeast growing at rate μ with limiting nutrient n, [M]0
is the geometric mean concentration of the metabolite across all conditions, μ is the growth rate, μ0
is the geometric mean growth rate, and mn
are model parameters (see for an illustration of the model). The parameter bn
(the “nutrient mean effect”) captures whether limitation for nutrient n generally enhances or depletes the metabolite M. The parameter mn
(the “growth rate slope”) captures the effect of growth rate within that nutrient limitation, i.e., whether [M] increases or decreases when the nutrient limitation is partially relieved and growth rate rises. Although used here primarily as a statistical tool, Eq. 2
also approximates typical functions used to relate metabolite concentrations and growth rate in cells. For example, if growth rate is a saturable function of the concentration of a limiting metabolite M as per Eq. 1
, then the slope mn
in Eq. 2
is the inverse of the Hill coefficient of the function in the growth-limited regime (see Supplemental Material derivation of this relationship).
Figure 3. Model-based determination of the nutrient mean effect and growth rate slope, using arginine as an example metabolite. Arginine concentration data (plotted using the same conventions as in ) were fit to Eq. 2; bn is the nutrient mean effect and (more ...)
plots nutrient mean effects versus growth rate slopes for various metabolites in nitrogen, phosphorus, and carbon (glucose) limitation, respectively. The metabolites shown are those with statistically significant effects on both dimensions at an FDR of 0.1. Supplemental Figure S4 shows analogous data for the auxotrophic limitations. These plots are also available in an interactive format on the website at http://growthrate.princeton.edu/metabolome/
Figure 4. Growth-limiting and overflow metabolites. Data for all metabolites were fit to the model exemplified in . Resulting plots of growth rate slope versus nutrient mean effect are shown here. (A) Nitrogen (ammonium) limitation. (B) Phosphorus (phosphate) (more ...)
For each nutrient condition, candidate growth-limiting species were those with relatively low concentrations (negative nutrient mean effect) that rose with growth rate (positive growth rate slope). Such metabolites are found in the upper left quadrant of the plots in , with roughly 10 candidates found per natural limitation. In contrast, for uracil limitation, we found only three candidates, two of which were UTP and CTP, the biopolymer precursors most directly linked to pyrimidine auxotrophy (Supplemental Figure S4). The identification of UTP and CTP in uracil limitation supports the validity of the analytical method. For leucine limitation, the analysis was less informative, as we were unable to resolve leucine from its structural isomer, isoleucine, in the present LC-MS/MS method.
In addition to identifying candidate growth-limiting metabolites, Eq. 2
also identified compounds that increase in response to limitation for a particular nutrient. Such “overflow metabolites” are characterized by a positive nutrient mean effect and negative growth rate slope and are found in the lower right quadrant of the plots in . Although somewhat fewer “overflow” than “growth-limiting” species were found, those identified tended to be clearly related to the nutrient limitation and often lacked the limiting element, e.g., nucleosides and nitrogenous bases in phosphorus limitation.
It should be noted that the nutrient mean effect term of the linear model can be skewed by metabolites that are extreme in one condition. For example, pyrimidine intermediates accumulated so greatly in the uracil auxotroph that their concentrations in all other conditions were significantly below the overall mean, i.e., a strong positive nutrient mean effect in uracil limitation resulted in the “artifactual” appearance of a negative nutrient mean effect in the other conditions. To avoid such skewing due to the auxotrophs, we repeated the above-mentioned analyses using only data for carbon, nitrogen, and phosphorus limitation (the natural limitations). With the exception of pyrimidine intermediates (which are omitted from on this basis), the results were qualitatively the same; however, due to smaller dataset size, statistical significance of some effects was reduced. For plots using only the natural limitation data, see Supplemental Figure S5 or http://growthrate.princeton.edu/metabolome/
As one approach to identify particularly interesting potential growth-limiting metabolites, for each nutrient limitation, we ordered the metabolites by statistical significance, first considering only the nutrient mean effect and then considering only the growth rate slope. Metabolites were then prioritized based on the sum of their ranks in these two dimensions (). This prioritization, based on statistical significance, favors metabolites that both fit the model closely and have strong nutrient mean effects and growth rate slopes. For example, in carbon (glucose) limitation, although arginine has the strongest nutrient mean effect and growth rate slope (C), pyruvate better fits the model, rising more steadily with increasing growth rate (compare carbon limitation data in and ), and accordingly rises to the top of .
Candidate growth-limiting metabolites under carbon, nitrogen, phosphorus, leucine, and uracil limitation
Growth-limiting Species in Nitrogen (Ammonium) Limitation
A large cluster of amino acids were decreased in nitrogen limitation (). Amino acids are logical candidates for limiting cellular growth. For example, low amino acid levels could limit the rate of tRNA charging. This in turn could lead to the accumulation of uncharged tRNA and impaired protein synthesis.
Every compound identified as potentially growth limiting in ammonium limitation was an amino acid (A, top left quadrant). Amino acids are linked to ammonium via glutamine (produced by reaction of ammonia with glutamate) or glutamate (produced by reaction of ammonia with α-ketoglutarate). Glutamine, but not glutamate, was identified as potentially growth-limiting. The growth-limiting pattern of glutamine was strong, highest in the ranked list of metabolites depleted in nitrogen limitation (). In addition, three of the four proteinogenic amino acids receiving nitrogen from glutamine (arginine, histidine, and tryptophan) were identified as potentially growth limiting (the fourth, asparagine, just missed the cut-off for statistical significance). Interestingly, nucleotides, which are also products of glutamine, were not similarly decreased.
Examination of the growth rate slope across all amino acids revealed that the concentration of every amino acid dropped with increasingly severe ammonium limitation, with the strongest response for glutamine, histidine and arginine, which contain two, three, and four nitrogens, respectively (). Both arginine and histidine receive nitrogen from glutamine, further implicating glutamine in control of nitrogen-limited growth.
Figure 5. Growth rate slope for amino acids under nitrogen limitation. The positive growth rate slope found for every amino acid implies that, under nitrogen limitation, each amino acid's intracellular concentration increases with faster cellular growth rate (i.e., (more ...)
Glutamine is preferred among amino acids as a nitrogen source for Saccharomyces
. Moreover, genetic or pharmacological inhibition of glutamine synthesis induces transcription of nitrogen-responsive genes (Mitchell and Magasanik 1984
; Crespo et al., 2002
; Zaman et al., 2008
). This central role of glutamine and/or its derivatives in indicating nitrogen limitation seems to be evolutionarily conserved, occurring also in bacteria (Ikeda et al., 1996
). Glutamine itself need not serve as the limiting species, however: another amino acid might instead (such as arginine, which also topped the ranked list). Dissecting the relative contributions of glutamine and its amino acid products to growth control will require additional experiments, e.g., direct measurement of tRNA loading, examination of strains with altered expression of amino acid biosynthetic enzymes or tRNAs.
Growth-limiting Species in Phosphorus (Phosphate) Limitation
The principal reaction of phosphate assimilation is phosphorylation of ADP to ATP. During phosphate limitation, ATP was identified as a potential growth-limiting species ( and B and ). Low levels of ATP can affect a great number of reactions, and a variety of direct and indirect products of ATP also showed growth-limiting patterns. These included NAD+
, UDP-glucose, and various sugar-phosphates, as well as UTP and CTP (but interestingly not GTP, which may play a greater regulatory role in bacteria; see, e.g., Krasny and Gourse, 2004
Among these species, the nucleotide triphosphates are the most directly related to biopolymer synthesis and thus particular appealing candidates to be growth limiting. Between ATP, UTP, and CTP, genetic evidence points to ATP being the most likely intracellular signal of phosphate status: two enzymes of ATP metabolism, adenylate kinase and adenosine kinase, negatively regulate the PHO pathway (Auesukaree et al., 2005
; Huang and O'Shea 2005
). One of these, adenylate kinase, catalyzes the conversion of two ADP molecules into ATP and AMP and thus serves to maintain cellular ATP levels.
During phosphorus limitation, when ATP was strongly decreased, ADP fell only slightly and AMP accumulated. Adenylate energy charge (AEC), defined as follows:
strongly fell (). It is unclear, however, whether the cells were actually energy limited, as the free energy of ATP hydrolysis differs from adenylate energy charge in being sensitive to the concentration of free Pi
but not [AMP]:
As the concentration of free phosphate in the cell is presumably very low during phosphorus limitation, it is likely that the cells maintained a relatively constant free energy of ATP hydrolysis despite their putative low adenylate energy charge. In contrast, during carbon (glucose) limitation, where energy availability might be expected to limit growth, adenylate energy charge remained high. Other stimuli that block energy production, however, such as sudden shift of respiring yeast to anaerobic conditions (Abbott et al., 2009
), or iron depletion (Thomas and Dawson 1977
), do alter energy charge. Accordingly, although “adenylate energy charge” indeed sometimes reflects energy-generating capabilities, it is also a measure of phosphorus charge.
Figure 6. Adenylate energy charge across conditions and growth rates. Conventions are as per : limiting nutrients are C, limitation for the carbon source, glucose; N, limitation for the nitrogen source, ammonium; P, limitation for the phosphorus source, (more ...)
Growth-limiting Species in Carbon (Glucose) Limitation
As noted above, we did not observe a decrease in ATP or a large drop in adenylate energy charge during carbon (glucose) limitation (, , and ). Similar results have previously been obtained in Escherichia coli
fed different carbon sources (Schneider and Gourse 2004
). Moreover, in yeast, acute relief of glucose limitation depletes, rather than increases, ATP levels (Somsen et al., 2000
). Thus, we suggest that, at least under aerobic conditions, glucose limitation leads to intracellular limitation for carbon, not energy.
In addition to signaling nitrogen limitation, depletion of amino acids is a possible reflection of carbon limitation. Two amino acids, threonine and arginine, showed a growth-limiting pattern (C), with threonine more sensitive to carbon than nitrogen availability. Histidine, although just missing the statistical cut-off for being growth limiting in glucose, showed a similar pattern to arginine, and these two amino acids cluster tightly together in . Interestingly, histidine biosynthesis is intertwined with purine biosynthesis, whereas arginine is intertwined with that of pyrimidines. It is accordingly possible that the depletion of these amino acids during nitrogen and carbon limitation plays a role in maintaining nucleotide pools when either carbon or nitrogen is scarce (e.g., by leading to impairment of protein synthesis and thereby decreased RNA synthesis).
Other metabolites showing a growth-limiting pattern in carbon (glucose) limitation included dihydroxyacetone phosphate and N
-acetyl-glucosamine-1-phosphate, both of which were also low in phosphorus limitation. Dihydroxyacetone phosphate is a key glycolytic intermediate. N-acetyl-glucosamine-1-phosphate is a precursor to UDP-N
-acetyl-glucosamine, a substrate in protein glycosylation and chitin biosynthesis. Although UDP-N
-acetyl-glucosamine did not show a growth-limiting pattern, the identification of such a pattern in N
-acetyl-glucosamine-1-phosphate is intriguing, as a closely related pathway in Bacillus subtilis
was recently implicated in nutrient control of cell size (Weart et al., 2007
). Moreover, chitin localizes to the yeast bud, a location especially relevant to cell division (Cabib and Bowers 1971
Pyruvate, which topped the list of potential intracellular indicators of glucose limitation (), was more specifically depleted in glucose limitation. Although pyruvate is not a direct biopolymer precursor, it does play a critical role in setting the balance between fermentation and respiration. Pyruvate dehydrogenase, which leads to acetyl-CoA and respiration, has a higher affinity for pyruvate than pyruvate decarboxylase, which leads to fermentative ethanol production. Low concentrations of pyruvate therefore favor respiration (Kresze and Ronft 1981
; Postma et al., 1989
; Nalecz et al., 1991
; Pronk et al., 1996
). Consistent with this, the glucose-limited cultures at low dilution rates, where no ethanol was produced (Supplemental Figure S6), had the lowest pyruvate concentrations across all conditions. In contrast, the fastest growing of the glucose-limited chemostats, where ethanol secretion occurred, had substantially higher pyruvate. The switch from respiratory to respiro-fermentative growth was accompanied by a sharp drop in biomass, as expected based on previous literature (Supplemental Figure S7) (Barford and Hall 1979
; Avigad 1981
; Postma et al., 1989
; Moore et al., 1991
; Goncalves et al., 1997
A Simple Quantitative Model of Growth Control
The above-mentioned analysis identified intracellular metabolites whose depletion might hinder cellular growth during nutrient limitation. A simplified model for this process involves envisioning growth as the assembly of these critical metabolites into biopolymer. The rate of growth could then be approximated as a maximum rate, which decreases whenever any of the critical components is scarce:
where μ is the growth rate, μmax
the maximum growth rate in the absence of nutrient limitation, xn
the concentrations of the limiting intracellular metabolite associated with nutrient limitation n
, and kn
. To test this model, we selected pyruvate, glutamine, ATP, and UTP as the putative limiting intracellular metabolites in carbon (C), nitrogen (N), phosphorus (P), and uracil (U) limitation, respectively. These selections are based on the particularly significant depletion of these species in the relevant nutrient condition (), their clear connection to the limiting nutrient, and their involvement (with the exception of pyruvate) in RNA or protein synthesis. The model did not consider leucine limitation, as leucine was not differentiated from isoleucine in our analyses. The model, using data from only four metabolites, fit the experimental data well (cross-validated R2
of 0.75, p = 3 × 10−7
). Importantly, this good fit did not require selective sensing of specific metabolites in particular nutrient conditions, but the more physically realistic case where all potential limiting species interact to control growth rate.
Although not candidates to limit growth, metabolites that accumulate during nutrient limitation are also informative regarding metabolic regulation. Early evidence for feedback inhibition in metabolic regulation came from the observation that pyrimidine intermediates accumulate during uracil limitation of a pyrimidine auxotroph (Pardee and Yates 1956
). Here, we recapitulate this finding, with dihydroorotate and orotate levels highest in severely pyrimidine-limited cells. It is possible that these high intermediate levels, not just decreased levels of end products, may adversely impact growth rate.
In leucine limitation, although we did not measure any pathway intermediates, we found a large number of other metabolites whose levels were elevated, including most amino acids (). When growth is slowed due to an auxotrophic limitation, if feedback inhibition is the main means of regulation, all metabolic end products unrelated to the auxotrophy are expected to accumulate. Quantitative analysis (see Supplemental Material) suggests that the concentrations of these end products should be a function of growth rate but independent of the nature of the auxotrophy (as long as the auxotrophy is in a pathway separate from that producing the end product). Experimentally, however, amino acids accumulate much more strongly in leucine versus uracil limitation (), pointing to the importance of other modes of regulation, e.g., general up-regulation of amino acid biosynthetic enzymes in leucine limitation due to Gcn4p activation (discussed below). In addition to amino acids, compounds accumulating in leucine limitation included phenylpyruvate, choline, glycerate, and pyruvate. The accumulation of pyruvate is consistent with the wasting of glucose that occurs when yeast are limited for an auxotrophic requirement but not for an elemental nutrient (Brauer et al., 2008
). In addition, pyruvate provides the carbon skeleton for leucine biosynthesis.
In natural limitations, lack of availability of substrates (e.g., glutamine in nitrogen limitation, ATP in phosphorus limitation) may impair biosynthesis. Accordingly, overflow of end products was not predicted, and it did not generally occur (). In nitrogen limitation, metabolites instead accumulated in the part of the TCA cycle directly upstream of ammonium assimilation (). The accumulation of α-ketoglutarate presumably underlies similar accumulation of phenylpyruvate, which is linked to α-ketoglutarate by transamination. In nitrogen limitation, trehalose (which rose in concentration during slow growth in every nutrient condition) was present at particularly high levels. This may reflect cellular efforts to store carbon when growth is nitrogen limited (Lillie and Pringle 1980
), or to compensate for decreased osmotic pressure when amino acid concentrations fall.
In phosphorus limitation, a large number of nucleosides and nitrogenous bases were found to overflow. These compounds are not involved in de novo nucleotide biosynthesis but instead are generated when phosphate moieties are scavenged from nucleotide monophosphates, e.g., via adenosine kinase, whose knockout favors activation of the PHO pathway (Auesukaree et al., 2005
; Huang and O'Shea 2005
). The overflow also of pyrimidine nucleosides and bases argues that, in addition to its annotated activity as a pyrimidine salvage enzyme, uridine kinase can function as a phosphate scavenge enzyme. Interestingly, glycerate and choline, two compounds that overflowed for unknown reasons during leucine limitation, also did so during phosphorus limitation. One possibility is that the overflow of choline (and possibly glycerate) indicates increased recycling of phospholipids: phosphatidylcholine can be deacylated by NTE1
, yielding glycerophosphocholine (Zaccheo et al., 2004
), which can subsequently be used by S. cerevisiae
as a phosphate source (Fernández-Murray and McMaster, 2005
Although essentially all cellular metabolites contain carbon, glucose limitation nevertheless resulted in the significant accumulation (i.e., positive nutrient mean effect) of certain species: mainly, multiply phosphorylated nucleotides such as triphosphates, UDP-glucose, and NAD+ (). Among these, the abundance of NAD+ also significantly decreased with faster growth, meeting our criteria for overflow (C). The accumulation of NAD+ when glucose is low may favor efficient respiration. In addition to nucleotide derivatives, glutamate, and its downstream product, proline, tended to rise in glucose limitation ( and C). The accumulation of glutamate and proline was surprising, given that most amino acids levels were decreased in carbon limitation. It is possible that elevated glutamate in severely glucose-limited cells reflects an overabundance of nitrogen relative to carbon, which in turn may drive the reductive amination of α-ketoglutarate.
Metabolites Correlated to Growth Rate across Different Nutrients
Although the concentrations of most metabolites were highly dependent on the limiting nutrient, some metabolites did show a general trend to increase or decrease with growth rate ( and ). Metabolites whose abundance was strongly correlated with growth rate, irrespective of nutrient limitation, were identified on the basis of their goodness-of-fit (r) to an analogue of Eq. 2
with all nutrient-specific terms removed:
Metabolites showing a statistically significant positive growth rate slope across all nutrient conditions (Bonferroni–Holm corrected p value <0.05, which corresponded to r > 0.64) included the lower glycolytic intermediates dihydroxyacetone-phosphate and bisphosphoglycerate and the nucleotide precursor ribose phosphate. When we considered only natural limitation conditions, taking the same goodness-of-fit cut-off, we found that two pyrimidine intermediates (orotate and dihydroorotate) and the arginine biosynthetic intermediate argininosuccinate also increased significantly with growth rate.
Figure 7. Metabolites with consistent growth rate responses across conditions. Conventions are as per : limiting nutrients are C, limitation for the carbon source, glucose; N, limitation for the nitrogen source, ammonium; P, limitation for the phosphorus (more ...)
Metabolites whose abundance varies with growth rate regardless of the limiting nutrient
The tendency for the glycolytic intermediates to increase with faster growth may reflect the rate of glucose metabolism being adjusted to meet growth requirements. Similarly, the concentrations of pyrimidine intermediates may relate to de novo pyrimidine biosynthetic flux, argininosuccinate to de novo arginine biosynthesis, and ribose phosphate to overall nucleotide biosynthetic activity. In each case, the higher intermediate concentrations may play a role in driving flux: for unsaturated enzymes, flux increases linearly with increasing metabolite concentrations. Thus, increasing concentrations of pyrimidine intermediates could drive biosynthesis without requiring increased synthesis of the corresponding enzymes. As we measured only a few biosynthetic intermediates outside of the pyrimidine pathway, other pathway intermediates may also increase with growth rate in a nutrient-nonspecific manner.
Only two metabolites showed a statistically significant trend to decrease with faster growth rate: trehalose and glutathione disulfide () (the reduced form of glutathione also displayed a similar trend, with a Bonferroni–Holm corrected p value of 0.12.) Both trehalose and glutathione are produced by enzymes that are targets of the Msn2p and Msn4p transcription factors, whose activities increase with slower cell growth. Of these two metabolites, trehalose concentrations were most tightly correlated to growth rate (). This observation is consistent with trehalose biosynthesis and utilization being temporally compartmentalized within the cell cycle, with synthesis occurring in G1 and use in S phase (Kuenzi and Fiechter 1969
; Paalman et al., 2003
; Tu et al., 2005
). As faster growth involves a shortening of G1 but not S phase (Hartwell et al., 1974
; Unger and Hartwell 1976
), the concentration of trehalose should decrease as the growth rate increases, as we observed. The accumulation of trehalose during the G1 phase may contribute to cell cycle control, with burning of trehalose helping to drive nutrient-limited cells through “Start”, the entry point of the yeast cell division cycle (Futcher 2006
). A possible benefit of this arrangement would be that cells would pass Start only when adequate internal carbon sources were available to make it back to G1, thereby protecting cells from being stranded in the cell cycle if environmental nutrient availability were to dry up (Chen et al., 2007
Previously, it was observed that transcripts up-regulated with faster growth tended to be involved in biosynthesis and protein translation, whereas down-regulated transcripts were enriched for genes that play a role in the stress response (Brauer et al., 2008
). An increase in trehalose has been suggested to be protective in osmotic stress (Hounsa et al., 1998
) and heat shock (Singer and Lindquist, 1998
), and glutathione plays a central role in the response to oxidative stress (reviewed in Penninckx, 2002
). This suggests that “biosynthetic” metabolites may increase while “stress-related” metabolites are depleted with faster growth, mirroring the functional classifications observed in the transcriptome.
Correlating Metabolite and Transcript Levels
Brauer et al. (2008)
have reported transcript data under the same conditions that we have used in the current study. In analyzing these data, Brauer et al. (2008)
focused on responses that were independent of the identity of the limiting nutrient. To find nutrient-specific responses, we determined the nutrient mean effect and growth rate slope for transcripts, resulting in analogous scatter plots to those in Figures and (for these plots, see http://growthrate.princeton.edu/metabolome/
). The top left and bottom right quadrants of these plots have different meanings for transcripts than for metabolites. For transcripts, those most induced during nutrient limitation appear in the lower right quadrant, e.g., transporters expressed to cope with scarcity of their substrates. We term such transcripts “limitation induced.” Those most repressed during nutrient limitation appear in the top left quadrant, and we term these transcripts “limitation repressed.”
Compared with the measured metabolite abundances, substantially fewer transcripts (20% at an FDR of 0.1 vs. 38% of metabolites at the same cut-off) showed significant limitation-induced or -repressed patterns. Such transcripts were significantly enriched for genes with known metabolic functions: the Gene Ontology terms “nitrogen compound metabolic process,” “metabolic process,” “pentose metabolic process,” “nucleoside, nucleotide, and nucleotide metabolic process,” and “water-soluble vitamin metabolic process” were all significant at an FDR of 0.01. In general, the functions of these transcripts corresponded well to their expression patterns. For example, in nitrogen limitation, the genes encoding the high-affinity, but energy-inefficient, ammonium assimilation pathway GS-GOGAT (GLN1
) were induced. These genes are regulated by the transcription factors Gln3p and Gat1p, which are regulated by nitrogen availability through target of rapamycin (TOR) complex 1 (Coffman et al., 1996
; Valenzuela et al., 1998
Many genes involved in oxidative metabolism were induced in carbon (glucose) limitation, e.g., ACS1
, and POX1
, all of which are regulated by the glucose sensitive transcription factor Adr1p (Young et al., 2003
; Tachibana et al., 2005
). In contrast, HXK2
was repressed by glucose limitation, consistent with its being a major glucose kinase in high glucose conditions and a mediator of glucose repression (Ma and Botstein 1986
; Diderich et al., 2001
). Similarly, the high-affinity glucose transporters HXT6
were induced by glucose limitation, whereas the low and intermediate affinity transporters (HXT1
, and HXT4
) were repressed (reviewed in Kruckeberg 1996
; Boles and Hollenberg 1997
Targets of the Leu3p transcriptional regulator (Kohlhaw 2003
; Boer et al., 2005
, and ILV3
were induced by leucine limitation, as were many genes regulated by Gcn4p, which activates transcription in response to uncharged tRNA (Hinnebusch 1992
; Natarajan et al., 2001
). Among genes with a positive nutrient mean effect in leucine limitation, 27% (49 of 182) were documented targets of Gcn4p (Teixeira et al., 2006
), versus 9% of all measured genes, a significant enrichment (p < 10−13
by Fisher's exact test). This induction of amino acid biosynthesis pathways under leucine limitation, when ample carbon and nitrogen were available, presumably accounts for the rampant amino acid accumulation in these cells ().
In phosphorus limitation, one of the most striking transcriptional patterns was, intriguingly, not directly related to phosphate metabolism, but instead to sulfate. SUL1
, and SAM1
were strongly repressed by phosphorus limitation, with SUL1
in phosphorus limitation showing the most positive growth rate slope of any gene in any condition. This suggests a strong relationship between phosphate limitation and sulfur metabolism. The repression of the high-affinity sulfate permease SUL1
when phosphate is scarce (see also Tai et al., 2005
), is probably a consequence of nonspecific sulfate transport through the highly expressed phosphate transporters. Nonspecific transport has been described previously for ammonium through potassium channels (Hess et al., 2006
), where uncontrolled ammonium influx led the yeast to excrete amino acids in an effort to detoxify ammonia. Consistent with cells needing to decrease the intracellular concentration of free sulfate, in phosphorus limitation we observe modestly higher methionine and glutathione concentrations. Interestingly, clones isolated after prolonged phosphate limitation often display increased expression of sulfur assimilation pathways (Gresham et al., 2008
). This would suggest that high intracellular sulfate (or a high intracellular ratio of sulfate to phosphate), resulting from the increased sulfate influx through phosphate transporters, is detrimental to yeast, which evolve the ability to more rapidly assimilate sulfate to deal with this toxicity.
Divergent Metabolome, Homeostatic Transcriptome
A striking feature of the metabolome response was the magnitude of the concentration changes (). This partial breakdown of homeostasis is perhaps not unexpected, given the relatively direct connection of metabolites to environmental nutrients. Environmental nutrient concentrations are outside the realm of cellular control, and in our experiments, extracellular glucose concentrations varied from 0.1 to 118 mM across different chemostats, corresponding to an ~1000-fold range, whereas leucine (in leucine-limited cells) and uracil (in uracil-limited cells) varied over a more modest, but still large range (~40-fold) (Supplemental Table S3). The median metabolite concentration also varied over a large range (14-fold) but not as large as the nutrients. In contrast, the median transcript varied over a substantially smaller range (3.2-fold), comparable with the experimental range of growth rates (5.5-fold). Thus, cellular regulatory systems partially dampen environmental variability at the level of the metabolome, and more strongly at the level of the transcriptome ().
Figure 8. Dynamic range of extracellular and intracellular small molecules, transcripts, and cellular growth rate. Dynamic range refers to the maximum fold-change across all experiments. Reported values for nutrients, metabolites, and transcripts are the median (more ...)
For metabolic enzymes, a key objective of transcriptional regulation is to control, via enzyme concentrations, metabolic flux. Although we did not measure metabolic flux here, in the absence of futile cycling, many fluxes are expected to scale linearly with growth rate. In this light, it is sensible that the experimental range of transcripts and growth rate are similar. Nevertheless, in preliminary efforts to relate enzyme expression to growth rate, we find many complexities. For example, enzymes catalyzing different steps in linear biosynthetic pathways sometimes show opposing patterns of regulation. These observations point to the importance of other modes of metabolic regulation, such as active site competition, allostery, and enzyme covalent modification, which require further dissection (see, e.g., Yuan et al., 2009
). In such efforts, the existence of a consensus reconstruction of yeast metabolism will provide a valuable roadmap (Herrgård et al., 2008
Another important question is how regulatory systems ultimately link the metabolome to growth rate and the transcriptome. The most straightforward possibility is that metabolite availability directly controls growth rate by limiting the availability of substrates for biomass synthesis. Such a view is consistent with the ability of cells to tailor their growth rate to auxotrophic requirements, and with our ability to model growth rate using a simple Michaelis-Menten approach based on a few limiting nutrients. Hence an appealing possibility is that the expression of transcripts showing generic growth rate effects is regulated downstream of growth rate, rather than solely via integration of signals from various nutrient-specific sensing pathways, including Ras/protein kinase A for glucose, TOR complex 1 for nitrogen and Pho80/Pho85 for phosphate (reviewed in Zaman et al., 2008
). This possibility would account for the similarity between the transcriptional responses to auxotrophic and to natural nutrient limitations, even when no system for sensing the auxotrophic limitation exists. One potential means by which transcription could be directly controlled by growth rate involves steady production of a transcriptional activator or repressor (at a rate independent of growth rate), with the intracellular concentration of that activator or repressor controlled by dilution by cell growth. Another possibility is that the total rates of protein and/or RNA synthesis might be sensed by the cell.
In addition to the nature of the mechanism linking the metabolome to the transcriptome, the precise pathway by which nutrient limitation controls growth also requires further investigation. One issue is whether the limiting intracellular metabolites actually fall to levels that directly limit biosynthesis, or instead just to levels that alter the activity of nutrient-sensing systems such as TOR. Another issue is that each elemental nutrient limitation resulted in decreases in the concentrations of multiple possible growth-limiting metabolites. This observation contrasts with a theoretical expectation that only a single metabolite should be growth limiting at any instant, as, even if the production of many metabolites is slowed by insufficient nutrient concentrations, only the one whose production is impaired the most will be limiting (Wingreen and Goyal, personal communication). Even though many species fall in concentration in each elemental nutrient limitation, it is possible that only a single species actually falls to limiting levels (e.g., in nitrogen limitation, glutamine might be growth limiting, with the fall in histidine, arginine, and other amino acids essentially incidental). More complex scenarios are also possible, however. One involves the limiting species varying between cells depending on inter-cell variation in enzyme expression. To this end, single-cell measurement of specific metabolite concentrations over time would be a valuable advance (Deuschle et al., 2005
). Combined with genetic perturbations and direct measurement of tRNA loading, such experiments hold the promise to assemble the candidate limiting metabolites found here into well-validated growth-control pathways.