Metabolic Consequences of Polyphosphate Synthesis and Imminent Phosphate Limitation

ABSTRACT Cells stabilize intracellular inorganic phosphate (Pi) to compromise between large biosynthetic needs and detrimental bioenergetic effects of Pi. Pi homeostasis in eukaryotes uses Syg1/Pho81/Xpr1 (SPX) domains, which are receptors for inositol pyrophosphates. We explored how polymerization and storage of Pi in acidocalcisome-like vacuoles supports Saccharomyces cerevisiae metabolism and how these cells recognize Pi scarcity. Whereas Pi starvation affects numerous metabolic pathways, beginning Pi scarcity affects few metabolites. These include inositol pyrophosphates and ATP, a low-affinity substrate for inositol pyrophosphate-synthesizing kinases. Declining ATP and inositol pyrophosphates may thus be indicators of impending Pi limitation. Actual Pi starvation triggers accumulation of the purine synthesis intermediate 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR), which activates Pi-dependent transcription factors. Cells lacking inorganic polyphosphate show Pi starvation features already under Pi-replete conditions, suggesting that vacuolar polyphosphate supplies Pi for metabolism even when Pi is abundant. However, polyphosphate deficiency also generates unique metabolic changes that are not observed in starving wild-type cells. Polyphosphate in acidocalcisome-like vacuoles may hence be more than a global phosphate reserve and channel Pi to preferred cellular processes.

turnover of poly(P) inside the organelle is regulated and coordinated with the release of P i into the cytosol is unknown.
To explore mechanisms that contribute to an early recognition of forthcoming P i limitation, we explored the metabolic consequences experienced by yeast cells that are at the brink of phosphate limitation, that is, where P i becomes scarce but not yet limiting for growth. Furthermore, we analyzed the metabolic significance of acidocalcisome-like vacuoles and their poly(P) pool.

RESULTS
Optimization of P i starvation conditions for metabolomic analysis. Yeast cells distinguish scarcity of P i (where they induce genes to optimize P i scavenging and maintain normal growth) from P i starvation that reduces growth and induces genes to facilitate P i recycling from internal sources (6, 18-20, 24, 51, 52). We used nontargeted metabolomic analysis to explore whether and how cells react to beginning P i scarcity in comparison to profound P i starvation. To this end, we first identified conditions that bring our strain background to the brink of P i limitation of growth. Batch cultures of cells were grown logarithmically overnight until the optical density at 600 nm (OD 600 ) was around 1 (4.6 Â 10 7 cells/mL). A small inoculum of these logarithmically growing cells was transferred into synthetic complete (SC) liquid medium containing 0 mM to 10 mM P i (OD 600 = 0.05; 2.3 Â 10 6 cells/mL), and the OD 600 was followed over 24 h. Compared to cells growing in high-P i medium (10 mM P i ), growth was normal at 1 mM and 0.5 mM P i , and we observed only a mild retardation of growth in 0.25 mM and 0.1 mM P i . Incubation in 0 mM P i arrested growth (Fig. 1A). We measured the activity of secreted acid phosphatase, which is activated by P i starvation (21), as a readout of the PHO pathway. In P i -free medium, the activity of secreted acid phosphatase gradually increased up to 8 h and then maintained a similar level up to 24 h, suggesting that the PHO pathway was maximally activated at 8 h (Fig. 1B). In 0.1 mM and 0.25 mM P i , the activity of acid phosphatase was partially induced at 8 h but fully induced at 24 h, probably reflecting the gradual depletion of P i from the medium over that period. In 0.5 mM P i medium, acid phosphatase activity increased to half of the maximal value on 0 mM P i . There was no significant growth delay under this condition, indicating that the cells managed to compensate the limited availability of P i , probably through induction of the PHO pathway (Fig. 1A). To assess the P i consumption of the cells under each condition, we monitored the amounts of P i remaining in the medium using the malachite green assay (Fig. 1C). In 10 mM P i , more than 90% of P i remained even after 8 h of incubation. In 0.5 mM P i medium, 70% of the P i was consumed after 6 h, and none remained after 8 h. We monitored this induction by tagging the Pho4 transcription factor with green fluorescent protein (GFP), which is translocated into the nucleus when cells lack P i . Pho4-GFP was predominantly in the nucleus after 8 h of incubation without P i (Fig. 1D), and 0.5 mM P i led to partial nuclear accumulation of Pho4-GFP, suggesting that the PHO pathway was moderately activated. We further measured the transcript level of the PHO pathway marker genes Pho5 and Pho84 using quantitative real-time reverse transcription-PCR (qRT-PCR). The gene expression of Pho5 and Pho84 was highly induced after 8 h of P i starvation ( Fig. 1E and F). Under P i scarcity (0.5 mM P i ), however, only Pho84 showed a mild, yet statistically significant, induction, whereas induction of Pho5 remained below 1% of the maximal value and was not significant. In agreement with earlier studies (18,53), this suggests that partial induction of the PHO pathway allows the cells to compensate for reduced availability of P i to a level that supports normal growth.
Under P i -limiting conditions, vacuolar poly(P) is degraded (18,34,45,46,54). It is assumed that resulting P i is exported from the vacuoles to replenish the cytosolic P i pool. In 0.25 mM, 0.125 mM, and 0 mM P i , poly(P) was completely degraded (Fig. 1G). However, after 8 h of growth in 0.5 mM P i , cells retained 40% of poly(P) compared to P i -replete conditions. We hence chose the described scheme of 8 h of growth in 0.5 mM P i as a condition for metabolomic analysis under P i limitation because it partially activates the PHO pathway, has no significant effect on cell growth, and allows partial maintenance of the vacuolar poly(P) pool.
Complete P i starvation induces broad metabolic changes. For metabolome analyses, logarithmically growing cells were transferred into medium providing abundant P i (10 mM), P i scarcity (0.5 mM), or P i starvation (0 mM P i ). After 8 h of incubation in these medium formulations, 0.5 OD 600 units of cells were harvested using vacuum filtration through a polytetrafluoroethylene polymer (PTFE) membrane and immediately FIG 1 Response of S. cerevisiae under different P i starving conditions. (A) Growth curves of yeast cells in synthetic complete medium supplemented with different concentrations of P i from 10 mM to 0 mM. Cells were inoculated at an OD 600 of 0.05 and cultured for 24 h. The means of triplicates are shown with standard deviation. (B) Acid phosphatase activities of yeast cells grown as in A. The means of triplicates are shown with standard deviation. (C) Concentrations of remaining P i in the medium during cell growth. P i concentration in the medium was monitored every 2 h for 8 h using the malachite green assay. The means of triplicates are shown with standard deviation; ***, P , 0.001; **, P , 0.01; *, P , 0.05; ns, not significant by Student's t test; nd, not detected. (D) Fluorescence microscopy of live yeast cells producing Pho4 genomically tagged with GFP as the sole source of this protein. Cells were incubated for 8 h in 10 mM, 0.5 mM, and 0 mM P i medium as in A before observation. (E and F) Relative gene expression levels of PHO5 (E) and PHO84 (F). Cells were grown in 10 mM, 0.5 mM, and 0 mM P i medium for 8 h and harvested for RNA extraction and qRT-PCR. Fold change values were normalized with internal control TAF10. The means of three biological replicates are shown with standard deviation; ***, P , 0.001; **, P , 0.01; *, P , 0.05; ns, not significant by Student's t test.
(G) Polyphosphate levels in different P i -containing medium. Cells were incubated for 8 h as in A and harvested for polyphosphate measurement. The means of triplicates are shown with standard deviation; ***, P , 0.001; **, P , 0.01; *, P , 0.05 by Student's t test; a.u., arbitrary units. frozen in liquid nitrogen (55). To analyze the metabolic effects of P i limitation, untargeted metabolomic analyses were conducted through hydrophilic interaction liquid chromatography (LC)-mass spectrometry (HILIC-MS). A total of 169 metabolites were identified. Individual metabolic features were normalized by the median of each sample, transformed to log 10 , and centralized to the mean by an autoscaling method for further statistical analyses. Partial least-squares discriminate analysis (PLS-DA) was performed to condense the metabolomic data into a simple plot, allowing easy comparison of overall metabolic features ( Fig. 2A). This revealed a clear separation of the three growth conditions. Component 1, which comprises the largest difference of the total variance in metabolites (53.3%), placed the 0 mM P i samples far from 10 mM and 0.5 mM P i samples, suggesting that the metabolic changes caused by P i starvation were greater than by P i limitation ( Fig. 2A). In the loading plot, which visualizes the contribution of individual metabolites to components 1 and 2, most phosphate-containing metabolites (red open circle) negatively contributed to component 1, indicating that they decreased under P i starvation (Fig. 2B). Only two phosphate-containing metabolites, AICAR and 39,59-cyclic GMP (39,, positively contributed to component 1. For further analysis, we compared variable importance in projection (VIP) scores of each metabolite, which represent the contribution of variables to the PLS-DA model encompassing 10 mM, 0.5 mM, and 0 mM P i data. ATP showed the highest VIP value among the top 20 most influential metabolites (Fig. 2C), underscoring an interrelation of P i and ATP metabolism (56).
Pearson coefficients were calculated as a measure for the correlation between metabolites. Two metabolic groups (group 1 and group 2) were negatively correlated with each other (Fig. 3A). The relative abundance of metabolites in 0 mM P i medium increased for group 1 and decreased for group 2 ( Fig. 3B and C). Purine and pyrimidine pathway metabolites, nucleosides, and nucleobases increased (Table 1), mirrored by a decrease of nucleotides. Metabolites of the citrate cycle (TCA) increased, such as citric acid, isocitric acid, and oxoglutaric acid, whereas the levels of glycolytic intermediates declined, suggesting an altered strategy for energy production under P i starvation. Nicotinate and nicotinamide metabolites also decreased. By contrast, metabolites of the tryptophan degradation pathway, which are involved in NAD de novo synthesis, accumulated.
Pathway analysis of P i starvation. To identify metabolites that accumulated differentially and in a statistically significant manner, we compared their relative abundance by volcano plots. Profound metabolic effects occurred under complete P i starvation; 31 metabolites increased and 49 decreased in a statistically meaningful way, representing 47% of the detected metabolites ( Fig. 4B and D; Data Set S1 in the supplemental material). Pathway analysis of these 80 metabolites identified the most affected pathways as purine metabolism, pyrimidine metabolism, nicotinate and nicotinamide metabolism, glycolysis/gluconeogenesis, citrate cycle, and cysteine and methionine metabolism, with a 2log (P value) of .1.5 ( Fig. 5B; Table 2).
(ii) Nicotinate and nicotinamide metabolism. The P i -containing metabolites nicotinic acid mononucleotide, NAD 1 , and NADP 1 decreased 2-to 5-fold under P i starvation (Fig. S2). By contrast, metabolites of NAD 1 de novo synthesis, also known as the kynurenine pathway, accumulated more than 6-fold. Based on these results, it can be speculated that the impaired NAD 1 synthesis from the kynurenine pathway will affect intracellular redox homeostasis.
(iii) Cysteine and methionine metabolism. S-Adenosylmethionine (SAM) is the major methyl donor for modifications of various biomolecules, including proteins, DNA, RNA, and metabolites, producing S-adenosylhomocysteine (SAH) as a byproduct of the reaction. The relative abundance of SAM and of methionine salvage pathway metabolites increased while that of SAH decreased under P i starvation (Fig. S3). Cystathionine, which can be produced from SAH via homocysteine, diminished under P i starvation, whereas metabolites derived from cystathionine, such as glutathione and taurine, were not significantly affected. P i starvation also affected another SAM-dependent branch, the synthesis of polyamines, because a byproduct of polyamine synthesis, 59-methylthioadenosine (MTA), accumulated strongly. These results suggest that yeast cells change their strategy of SAM utilization under P i deprivation. The reduction of nucleic acid synthesis, which accompanies growth arrest, may reduce consumption of SAM for nucleotide synthesis and promote accumulation of this compound.
(iv) Purine and pyrimidine metabolism. Nucleosides and nucleobases, such as adenosine, inosine, xanthosine, guanosine, uridine, cytosine, and xanthine, significantly accumulated under P i starvation ( Fig. S4A and B), whereas P i -containing nucleotides, nucleoside diphosphates, and nucleoside triphosphates decreased. By contrast, the amount of AICAR increased. This metabolic change may contribute to triggering the PHO pathway under P i starvation in two ways. First, AICAR inhibits the production of IP 8 (57), which itself is a potent suppressor of the PHO pathway (9). Second, AICAR stabilizes the interaction of the association of the transcription factors Pho4 and Pho2, which is necessary for full induction of the PHO pathway (17).
Moderate P i limitation causes few potentially diagnostic metabolic changes. Few significant changes (2log [P] . 1) occurred under P i -limiting conditions; only 17 metabolites increased and 10 decreased more than 1.5-fold ( Fig. 4A and C and 5A; Data Set S1). ATP decreased in a statistically significant manner, which suggests that the ATP level is more sensitive to P i availability than most other metabolites, making ATP a bona fide early indicator of P i scarcity. In line with this, one class of key enzymes for signaling the intracellular P i state, IP 6 kinases, have an unusually high K m for ATP, which is close to the normal cellular concentration of this compound (58). To test this possibility, we measured the products of these enzymes, inositol pyrophosphates, under different P i conditions using capillary electrophoresis-coupled electrospray ionization mass spectrometry (CE-ESI-MS). Under P i starvation, 1,5-IP 8 was not detected at all, 5-IP 7 and 1-IP 7 decreased by 80%, and IP 6 decreased by 60% ( Fig. 6A to D). In contrast to the PP-IPs, the decrease of IP 6 occurred only with a lag phase of 2 h (Fig. S8A to D). Even under mild P i scarcity, all four compounds significantly declined in comparison with P i -replete conditions, by 60% for 1,5-IP 8 , 30% for 5-IP 7 , and 50% for 1-IP 7 and IP 6 . This is consistent with the hypothesis that even moderate decreases in ATP levels under P i limitation can be translated into decreased PP-IP levels, which then activate SPX domain-based signaling to stabilize cytosolic P i . Acetyl-CoA, fructose-6-phosphate, glyceraldehyde-3-phosphate, glucose-6-phosphate, 2/3-phosphoglyceric acid, phosphoenolpyruvic acid Cysteine and methionine 1-Aminocyclopropanecarboxylic acid, ophthalmic acid, S-adenosylhomocysteine, cystathionine, 3-sulfinoalanine, pyroglutamic acid Nicotinate and nicotinamide NAD 1 , NADP 1 , nicotinic acid mononucleotide Glycerolipid sn-Glycerol-3-phosphate, ethanolamine, glycerol Arginine N-Acetylputrescine, proline, N-acetylglutamic acid, citrulline, dimethylarginine Others Guaiacol, aminoisobutyric acid, 2-hydroxyglutaric acid, aspartic acid, glutamic acid, glycine, histidine, lysine, serine, threonine, mevalonic acid, N 2 -acetyllysine, TMP, UDP-hexose, UDP-Nacetylhexosamine Poly(P) in acidocalcisome-like vacuoles contributes to P i homeostasis even under P i -replete conditions. Yeast cells contain acidocalcisome-like vacuoles, which can convert the g -phosphate from ATP into inorganic polyphosphate. Thereby, they can store phosphate units at concentrations in the hundreds of millimolar in an osmotically inactive form (40,41). At the same time, vacuoles contain polyphosphatases and P i exporters, which Same as in C but shows metabolites changing at least 2-fold under P i starvation (0 mM P i ). The relative abundance of metabolites is represented as log 2 (fold change) through a color code.
can hydrolyze poly(P) and export the liberated P i to the cytosol (46). This system is powerful enough to influence P i homeostasis of the cells and the P i starvation response when it is dysregulated (18,37,38,54). To investigate the metabolic role of vacuolar polyphosphates, we analyzed the metabolic profiles of the Dvtc4 mutant, which lacks the poly(P)synthesizing complex VTC, under P i -replete conditions and P i starvation. To maximize the chance of observing poly(P)-dependent differences, we restricted the starvation period to 2 h because wild-type cells mobilize their poly(P) pool over the first 2 to 3 h of P i starvation (18)(19)(20). A PLS-DA plot showed that the metabolic features of Dvtc4 cells were clearly distinct from those of the wild type under P i -rich conditions and even more under P i starvation (Fig. 7A). The loading plot of PLS-DA revealed that almost all phosphate-containing  metabolites (red open circles) were abundant under P i -rich conditions, except AICAR, consistent with the results above ( Fig. 2B and 7B).
To analyze how polyphosphate synthesis and P i concentration in the medium affect metabolic features, a two-way analysis of variance (ANOVA) was performed. The relative abundances of 92, 116, and 78 metabolic features were affected by poly(P), P i concentration, and their interaction, respectively ( Fig. 8A; Data Set S2). A total of 65 metabolites were affected by both poly(P) and P i , of which 66% (43 features) were additionally affected by their interaction. Metabolite set enrichment analysis was performed using these 43 metabolites. Most metabolic pathways affected by P i starvation shown in the previous pathway analysis, such as pyrimidine metabolism, nicotinate and nicotinamide metabolism, glycolysis, and purine metabolites, were again ranked statistically high, showing consistency between the analyses ( Fig. 8B; Table 3). The changes of these 43 metabolites were visualized by heatmap analysis (Fig. 8C). Dvtc4 cells showed much more pronounced changes than wild-type cells in several respects, including the decrease of P i -containing purines and pyrimidines (CMP, UMP, AMP, and dGMP) ( Fig. 8C; Fig. S5A and B), the increase of nucleosides and nucleobases (cytosine, cytidine, guanosine, uridine, and inosine) ( Fig. 8C; Fig. S7 and S8), and the reduction of NAD 1 and NADP 1 (Fig. S6).

Metabolomics of Phosphate Scarcity mBio
Dvtc4 cells grown on P i -replete medium showed numerous metabolic features of P istarved wild-type cells. They did not, however, simply phenocopy a P i starvation response at a reduced scale. Intermediates of tryptophan degradation, such as kynurenine, kynurenic acid, 3-hydroxykynurenine, and the NAD 1 precursor quinolinic acid, which did not change significantly in P i -starved wild-type cells, were increased in Dvtc4 cells already under P i -replete conditions and increased further (up to 5-fold) after P i starvation ( Fig. 8C; Fig. S6). In Dvtc4 cells, the later glycolytic intermediates dihydroxyacetone phosphate and phosphoenolpyruvate were more abundant than in wild-type cells in P i -replete medium, and 2/3-phosphoglyceric acid underwent a much more pronounced reduction than in wild-type cells after P i starvation ( Fig. 8C; Fig. S7). These results indicate that poly(P) synthesis by VTC has a significant effect on the metabolic profile of cells. Poly(P) synthesis dampens the metabolic consequences of P i starvation, which is consistent with its proposed role as a P i reserve, but it also has, so far, unrecognized metabolic functions under P i -replete conditions, as indicated by metabolic features of Dvtc4 cells that cannot be recapitulated by P i scarcity or P i starvation of wild-type cells.

DISCUSSION
Our results extend earlier studies of P i starvation (56,59,60). In agreement with these studies, we observed reductions in nucleotides and late glycolytic intermediates and increased nucleoside and nucleobase levels. These changes can be explained by simple mass action (60, 61) because they reduce phosphate-containing metabolites under conditions of intracellular P i shortage. By contrast, early TCA cycle metabolites, such as citric acid, isocitric acid, and oxoglutaric acid, strongly increased during P i starvation. Furthermore, oxygen consumption of yeast increases after P i starvation (62). This suggests that mitochondrial respiration may become activated as an alternative mechanism for energy production because it fixes less P i in metabolic intermediates than ATP production based on glycolysis. In line with this, we made the side observation that P i starvation also caused mitochondrial fragmentation (data not shown). Mitochondria fragment in medium favoring respiration, such as nonfermentable carbon sources or glucose-limited medium (63,64).
Following P i starvation, two P i -containing metabolites increased, AICAR and cGMP ( Fig. 2B; Fig. S4A in the supplemental material). Little is known about the roles of cGMP in yeast so far (65); however, it provides a potential link to protein kinase A signaling, which is involved in P i homeostasis and signaling through P i transporters in yeast (66)(67)(68)(69)(70)(71)(72). cGMP can also inhibit DNA polymerase (73), which might reduce P i consumption by cells and avoid P i depletion during S phase (47). AICAR is an intermediate of de novo purine biosynthesis and activates a master regulator of energy homeostasis, AMP-activated protein kinase (AMPK), in mammalian cells (74), but the yeast AMPK Snf1 does not depend on it. Snf1 is activated by ADP (17,75). After a block in nucleoside synthesis, accumulating AICAR stimulates the PHO transcription pathway by stabilizing the interaction of the responsible transcription factors Pho4 and Pho2 (17). Furthermore, AICAR reduces the activity of mammalian PPIP5 kinase (57), suggesting that it might contribute to the decline of IP 7 and IP 8 after P i starvation that also occurs in these cells (76). However, it has remained unknown how AICAR behaves under P i starvation. Our analysis now shows that AICAR accumulates during P i starvation. AICAR accumulation should be favored by the decrease of ADP and ATP levels following P i starvation because these nucleotides exert feedback inhibition on the first step in purine synthesis and thereby on AICAR synthesis (77). Thus, AICAR may promote expression of PHO genes following P i starvation. However, under P i scarcity, when ATP decreases less severely than under P i starvation, AICAR did not increase. This corresponds to the only partial activation of the PHO pathway under P i scarcity. We hence propose that an increase in AICAR may contribute to switching the PHO pathway from partial to full activation when cells transit from P i scarcity to starvation. Glucose-6-phosphate is used by the pentose phosphate pathway (PPP) to convert NADP 1 to NADPH, which is essential for cellular redox homeostasis (78,79). Although NADPH was not detected in our metabolomic analysis, NADP 1 decreased, and we hence assume that NADPH should decrease as well. NAD 1 , nicotinic acid mononucleotide, and nicotinic acid, which are precursors of NADP 1 , were all reduced by P i starvation, but intermediates of the kynurenine pathway, also known as the de novo NAD 1 synthetic pathway, were significantly accumulated (Fig. S2). These changes may result from the accumulation of AICAR, which stimulates the expression of enzymes involved in the kynurenine pathway (80). The last metabolite of the kynurenine pathway, quinolinic acid, is converted to nicotinic acid mononucleotide by consuming phosphoribosyl pyrophosphate (PRPP). Because PRPP is produced from ribose-5-phosphate, this reaction may be impaired through the decrease in ribose-5-phosphate under P i starvation, favoring the observed accumulation of quinolinic acid. In addition, the PHO pathway directly promotes the catabolism of NAD 1 by inducing the vacuolar phosphatase Pho8, which removes P i from nicotinic acid mononucleotide and nicotinamide mononucleotide (81). The decrease in the NAD 1 and NADP 1 pools is expected to affect intracellular redox homeostasis. This may increase the dependence of cells on the oxidative stress response for surviving P i starvation, which had been previously observed (62). In line with this, cells with perturbed P i and inositol pyrophosphate homeostasis induce the environmental stress response (82,83).
We observed increased SAM and decreased SAH under P i starvation. SAM is synthesized from methionine and ATP, releasing P i and PP i . SAM provides methyl groups for methyl transfer reactions, generating SAH as a byproduct (84). Histone methylation affects global gene expression patterns by changing the structure of chromatin through interactions with various chromatin remodeling factors and transcription regulators (85,86). The expression of PHO genes is also under the control of histone methylation. Expression of Pho5 and Pho84 is induced in the Dset1 mutant, which affects methylation of Lys4 of histone H3 (87,88). In addition, the methyltransferase Hmt1 promotes expression of several P i -responsive genes (89). Thus, the changes of SAM following P i starvation might alter the intracellular methylation status and thereby provide a further route of input for P i -dependent gene expression.
An interesting question is how cells distinguish P i scarcity from P i starvation. This is challenging because P i scarcity can be corrected by cells through partial induction of the PHO pathway. The resulting improved capacity for P i scavenging (e.g., through expression of high-affinity P i transporters) apparently allows them to maintain sufficient metabolic performance to support normal growth. Furthermore, positive feedback loops involving Spl2, a small regulator of the P i transporter Pho90, may stably commit cells to activation of the starvation program, even if this reestablishes sufficient intracellular P i supply (20,51,52). Nevertheless, the P i starvation program is not launched in a simple all-or-none fashion. Certain genes are activated at different levels of P i shortage, as exemplified by the gene for the high-affinity transporter Pho84, which is activated earlier than the secreted acid phosphatase Pho5 (18)(19)(20). Our analyses of the low-P i state, which were performed in cells that induced Pho84 but not Pho5, provide hints on metabolic changes that might be used by cells to distinguish P i scarcity from P i starvation.
Under P i scarcity, few metabolites changed in a statistically meaningful way, but the levels of ATP, ADP, and PP-IPs decreased by 30 to 50% compared to P i -replete conditions. Known properties of the enzymes involved in the production of ATP and in the synthesis of PP-IPs support the following working hypothesis on the reasons for these declines (Fig. 9). Glycolysis and oxidative phosphorylation both contain enzymes that use P i as a substrate, glyceraldehyde-3-phosphate dehydrogenase and F-ATPase, respectively. Both enzymes have K m values for P i of around 1.5 to 2 mM (90, 91), rendering them susceptible to declines of P i below the 1 to 2 mM threshold that cells normally maintain on P i -replete medium. This can reduce ATP production by both pathways following P i scarcity. PP-IP production is sensitive to both ATP and P i . The IP 6 kinases synthesizing inositol pyrophosphates have K m values for ATP of 1 to 2 mM, which is in the range of the cellular ATP concentration under P i -replete conditions (58). While PPIP5 kinase activity, with its K m for ATP of 0.1 mM, is shielded from changes in ATP and weakly stimulated by P i (92,93), the opposing phosphatase activity of this bifunctional enzyme is increasingly inhibited in the P i range from 0.1 to 2 mM, covering the normal cellular concentration range (93). The combination of these enzymatic properties may allow an even moderate decrease in P i to significantly reduce PP-IP levels. As we recently observed that the PHO pathway is repressed by PP-IPs through the SPX domain of Pho81 (9), we propose that this decline of PP-IPs following P i scarcity reduces Pho85/Pho80 activity and promotes partial Pho4 nuclear translocation and partial activation of the PHO pathway, allowing cells to maintain normal growth.
Yeast stores P i in the form of polymers in acidocalcisome-like vacuoles. Under P i -limiting conditions, polyphosphatases degrade polyphosphate and liberate P i , which could potentially be brought back to the cytosol through the vacuolar P i transporter Pho91 (34,45,46). In this way, the vacuolar poly(P) pool might buffer the cytosol against sudden drops in P i and delay the onset of the P i starvation response (6,18). In line with this, our metabolomic analysis revealed exaggerated metabolic changes when the poly(P)-deficient Dvtc4 mutant was starved for P i . This supports a significant role of acidocalcisome-like vacuoles in P i homeostasis, which may provide the proposed buffer for cytosolic P i . Such a buffer is of obvious relevance for an ordered transition into P i starvation and cell cycle arrest (48,94). Surprisingly, metabolic features of P i starvation were observed in the Dvtc4 mutant already under high P i conditions. This phenotype may again reflect the buffering function of poly(P). This function may FIG 9 Working hypothesis on the translation of cytosolic P i concentration into changes of PP-IPs. The scheme illustrates the inhibitory (red) and stimulatory (green) influences of P i on key enzymes of ATP and PP-IP production, which are postulated to result from the high K m and half-maximal inhibitory concentration (IC 50 ) values of GAP-DH, F-ATPase, IP 6 kinase, and PPIP5 kinase. Details are discussed in the main text.
become important even on high-P i medium because the duplication of all nucleic acids and phospholipids generates a very high need for P i in S phase. It has been proposed that this need can transiently exceed the maximal import capacity of cells, necessitating the vacuolar poly(P) store to cover the deficit (47,48,54,95). In line with this, we observed that the amounts of 1,5-IP 8 , 5-IP 7 , and 1-IP 7 were reduced in the Dvtc4 mutant in high-P i medium (Fig. S8E to H). An explanation of the starvation features of Dvtc4 cells from this perspective is, however, only partially satisfactory for our data set because Dvtc4 cells on high-P i medium shows numerous, but not all, features associated with P i starvation. The trend for some metabolites was even inversed, such as for adenosine, guanosine, orotic acid, acetyl-CoA, and phosphoenolpyruvate. This suggests that poly(P) may have additional metabolic functions that go beyond those of a P i buffer for the cytosol. There is potential for this because poly(P) has a significant role for the storage of cations, such as for Zn 21 , Ca 21 , Mn 21 , and Mg 21 (96)(97)(98)(99), and also for cation uptake, as shown for Mg 21 (100). Furthermore, poly(P) may affect cellular signaling, influencing the stress response (101)(102)(103)(104). Here, it may even have direct impact, such as through polyphosphorylation of lysine residues, which modifies yeast topoisomerase 1 (Tpo1) and nuclear signal recognition factor 1 (Nsr1) (105,106).
In sum, our observations favor a model where lack of P i induces differential metabolic changes, which together promote the P i starvation response. Beginning P i scarcity could be diagnosed through moderate declines in ATP and inositol pyrophosphates, leading the cells to partially activate P i scavenging systems and maintain normal growth. Profound P i starvation entails numerous additional metabolic changes, such as through AICAR, SAM, and strong reductions in ATP and inositol pyrophosphates. These changes may fully stimulate the transcriptional P i starvation and stress responses in a combinatorial manner. We consider the latter point as an attractive potential solution of the specificity problem that is inherent in the task of measuring a very abundant metabolite. Nuclear magnetic resonance (NMR) studies of yeast found total P i concentrations in the cell to be around 20 mM, of which approximately one-fourth is cytosolic (38,107). This allows for the estimation of cytosolic P i under P i -replete conditions at 5 mM, declining to 1 mM after P i starvation. Given that P i is present in the cytosol in millimolar concentrations, it is difficult to envision that it be "measured" by specific binding to a very low-affinity receptor, which might be susceptible to competition by numerous other compounds. Coincidence detection through a network of P i -dependent metabolic reactions could, however, generate such specificity, even for a highly abundant ligand such as P i . Therefore, we favor this model of P i detection.

MATERIALS AND METHODS
Yeast strains. The S. cerevisiae strains used in this study are listed in Table 4. Endogenous GFP tagging was performed as described previously (108). The yEGFP-CaURA3 was PCR amplified from the plasmid pKT209 by introducing 40-bp homology before and after the stop codon of the Pho4 gene. Gene deletion was conducted based on the CRISPR-Cas9 system as described previously (109). The single guide RNA (sgRNA) was cloned into the sgRNA expression vector and cotransformed into yeast cells with hybridized double-stranded oligonucleotides, which contain 40-bp homology of each side before the start codon and after the stop codon of the Vtc4 gene as the templates for homologous recombination. After transformation, positive colonies were selected by colony PCR and sequencing. PCR primers used for genetic manipulation are listed in Table 5.
Synthetic complete (SC) medium was prepared using yeast nitrogen base without phosphate (Formedium, UK). The desired phosphate concentration was adjusted by adding KH 2 PO 4 . The potassium concentration was controlled by adding KCl instead of KH 2 PO 4 .
Media and cell growth. All media were prepared with ultrapure, UV-treated water from an ultrapurification system (SG, Germany). SC medium was prepared using yeast nitrogen base without phosphate (Formedium, UK). The phosphate concentration was adjusted by adding KH 2 PO 4 , and the potassium concentration was kept constant by substituting KCl for KH 2 PO 4 . For assays of growth, acid phosphatase, and malachite green, yeast cells were logarithmically grown overnight in 50 mL of SC medium containing 10 mM P i up to an OD 600 of 1 (4.6 Â 10 7 cells/mL). Cells were sedimented in a table-top centrifuge (3,000 Â g) and washed with SC medium containing different concentrations of P i . After two washing steps, the OD 600 was measured, and cells were inoculated in 100-mL Erlenmeyer flasks containing 20 mL of SC medium with the desired P i concentration (OD 600 = 0.05; 2.3 Â 10 6 cells/mL). Cells were incubated at 30°C with shaking at 210 rpm in a shaking incubator (Climo-shaker ISFL-X, Kühner, Switzerland). To assess growth, the OD 600 was monitored at different time points (0, 2, 4, 8, and 24 h).
For microscopic analysis, RNA extraction for qRT-PCR, poly(P) measurement, inositol pyrophosphate extraction, and yeast metabolite extraction, yeast cells were prepared in the same manner as described above except that they were transferred into 0 mM P i medium after the washing steps (OD 600 = 0.2; 9.2 Â 10 6 cells/mL).
Acid phosphatase assay. An acid phosphatase assay was performed as previously described (110). Cells were grown in the same manner for growth assays as described above. At each time point, 0.2 OD 600 units of cells (9.2 Â 10 6 cells) were harvested by centrifugation in a bench-top centrifuge (3,000 Â g) and resuspended in 250 mL of 0.1 M sodium acetate (pH 4.2) and 250 mL of freshly prepared 9 mg/mL p-nitrophenyl phosphate. The mixture was incubated at 37°C for 9 min, and 800 mL of 1.4 M Na 2 CO 3 was added to stop the reaction. After centrifugation, the OD 420 was measured from the supernatant as acid phosphatase activity.
Malachite green assay. Logarithmically grown cells were inoculated into SC medium containing different concentrations of P i as described above. At different time points (0, 2, 4, 6, and 8 h), 1 mL of cell culture was transferred into a microcentrifuge tube and sedimented by centrifugation at 13,000 Â g for 1 min. The supernatant was transferred into a new tube and diluted with P i -free SC medium to be within a linear range of detection (200-fold dilution for 10 mM samples, 10-fold dilution for 0.5 mM samples, and no dilution for 0 mM samples). Fifty microliters of diluted samples was mixed with 32 mL of 0.1 mM malachite green solution containing 0.35% (mass/vol) of polyvinyl alcohol (molecular mass 85,000 to 124,000 Da) and 43 mL of 4.48 mM ammonium molybdate solution containing 12.5% (vol/vol) H 2 SO 4 . After incubation for 15 min at room temperature, the absorbance was measured at 620 nm on a SpectraMax M3 plate reader (Molecular Devices, USA) in a 96-well clear plate with a flat bottom.
Fluorescence microscopy. Cells in the logarithmic phase were inoculated in SC medium and grown as described for the growth assay above. Fluorescence images were obtained with a Nikon Eclipse Ti2/Yokogawa CSU-X1 spinning-disk microscope with two Prime BSI scientific complementary metal oxide semiconductor (sCMOS) cameras (Teledyne Photometrics, USA), a LightHub Ultra laser light (Omicron Laserage, Germany), and an Apo total internal-reflection fluorescence (TIRF) Â100/1.49 oil lens (Nikon, Japan). Experiments were repeated at least three times. Representative images are shown in the figures.
RNA extraction and qRT-PCR. Total RNA was extracted from 10 OD 600 units of yeast cells (4.6 Â 10 8 cells) with RNeasy kits (Qiagen, Germany) according to the manufacturer's instructions. One microgram of total RNA was used for cDNA synthesis using RevertAid reverse transcriptase (Thermo Fisher Scientific, USA). Gene expression levels were quantitatively monitored using real-time PCR (LightCycler 480, Roche, Switzerland) with SYBR Green I master mix (Roche, Switzerland). Gene expression was normalized by using TATA-binding protein-associated factor Taf10 transcript as an internal control. Primers used for qRT-PCR are listed in Table 5. The mean and standard deviation values of gene expression were calculated from three biological replicates with three technical replicates.
Poly(P) measurement. Poly(P) levels were evaluated from cells using the direct 49-6-diamidino-2-phenylindole (DAPI) assay (62). Cells were logarithmically grown in P i -rich SC medium and transferred to SC medium containing different concentrations of P i , as described above. After 8 h of incubation, 0.5 OD 600 units of  for column reequilibration. The flow rate was 400 mL/min, column temperature was 25°C, and the sample injection volume was 2 mL. ESI source conditions were set as follows: dry gas temperature of 290°C, nebulizer of 35 lb/in 2 and flow of 14 L/min, sheath gas temperature of 350°C and flow of 12 L/min, nozzle voltage of 0 V, and capillary voltage of 2,000 V. Dynamic multiple reaction monitoring (DMRM) was used as acquisition mode with a total cycle time of 600 ms. Optimized collision energies for each metabolite were applied. In negative mode, a SeQuant ZIC-pHILIC (100-mm, 2.1-mm i.d., and 5-mm particle size; Merck, Germany) column was used. The mobile phase was composed of A (20 mM ammonium acetate and 20 mM ammonium hydroxide in water, pH 9.7) and B (100% acetonitrile). The linear gradient elution was from 90% (0 to 1.5 min) to 50% B (8 to 11 min) down to 45% B (12 to 15 min). Finally, the initial chromatographic conditions were established as a postrun during 9 min for column reequilibration. The flow rate was 300 mL/min, the column temperature was 30°C, and the sample injection volume was 2 mL. ESI source conditions were set as follows: dry gas temperature of 290°C and flow of 14 L/min, sheath gas temperature of 350°C, nebulizer of 45 lb/in 2 and flow of 12 L/min, nozzle voltage of 0 V, and capillary voltage of 22,000 V. DMRM was used as an acquisition mode, with a total cycle time of 600 ms. Optimized collision energies for each metabolite were applied. Data preprocessing. Raw LC-MS/MS data were processed using the Agilent Quantitative analysis software (version B.07.00 MassHunter, Agilent Technologies, USA). Relative quantification of metabolites was based on extracted ion chromatogram (EIC) areas for the monitored MRM transitions. The obtained results were exported to R software (http://cran.r-project.org/), and signal intensity drift correction was done within the LOWESS/Spline normalization program followed by noise filtering (Coefficient of Variance [Quality Control features] of .30%).
Statistical analysis of metabolite profiling. Statistical analyses of metabolomic data were performed by MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/) (117). Before analysis, signal intensity data were median normalized, log transformed, and mean centered using the autoscaling method. PLS-DA of the first metabolic profiling with different P i conditions was conducted by considering the P i concentration order. The heatmap of the correlation matrix between metabolites of different P i conditions was calculated by the Pearson r correlation coefficient. Volcano plot analysis was performed by a two-sample t test. The metabolites showing a P value of ,0.1 with an absolute value fold change (jFCj) of .1.5 (10 mM P i versus 0.5 mM P i ) or jFCj of .2 (10 mM P i versus 0 mM P i ) were considered statistically meaningful metabolites. Results of the volcano plot analysis were exported and visualized with GraphPad Prism 9 (GraphPad Software, USA). For the second metabolic profiling analysis, using wild-type and Dvtc4 cells, the prominent outliers from the PLS-DA were removed before further analyses. A two-way ANOVA followed by false discovery rate correction (P , 0.05) was performed to investigate metabolite variabilities between two different factors, genotype (wild type and Dvtc4), and P i conditions (10 mM P i and 0 mM P i ) and their interaction. A hierarchical clustering heatmap was generated using the Euclidean distance measure with Ward's clustering method. The 50 most significantly changed metabolites according to the ANOVA were selected for visualization.
Pathway analysis and metabolite set enrichment analysis. Pathway analysis was performed using MetaboAnalyst 5.0 based on the metabolites that statistically significantly increased or decreased under 0.5 mM P i (jFCj . 1.5; P , 0.1) or 0 mM P i (jFCj . 2; P , 0.1) conditions compared to under 10 mM P i conditions. Hypergeometric test and relative betweenness centrality were used for the enrichment method and topology analysis, respectively. Metabolite set enrichment analysis was performed by MetaboAnalyst 5.0 based on 84 metabolite sets of KEGG human metabolic pathways. Results of pathway analysis and metabolite set enrichment analysis were exported and visualized with GraphPad Prism 9.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. DATA SET S1, XLSX file, 0.01 MB. DATA SET S2, XLSX file, 0.01 MB.