Genome-wide Analysis Reveals Inositol, Not Choline, as the Major Effector of Ino2p-Ino4p and Unfolded Protein Response Target Gene Expression in Yeast*

In the yeast Saccharomyces cerevisiae , the transcription of many genes encoding enzymes of phospholipid biosynthesis are repressed in cells grown in the presence of the phospholipid precursors inositol and choline. A genome-wide approach using cDNA microarray technol-ogy was used to profile the changes in the expression of all genes in yeast that respond to the exogenous presence of inositol and choline. We report that the global response to inositol is completely distinct from the effect of choline. number of genes the genes the UAS INO Ino2p Ino4p. by OPI1 , Ino4p,

In the yeast Saccharomyces cerevisiae, the transcription of many genes encoding enzymes of phospholipid biosynthesis are repressed in cells grown in the presence of the phospholipid precursors inositol and choline. A genome-wide approach using cDNA microarray technology was used to profile the changes in the expression of all genes in yeast that respond to the exogenous presence of inositol and choline. We report that the global response to inositol is completely distinct from the effect of choline. Whereas the effect of inositol on gene expression was primarily repressing, the effect of choline on gene expression was activating. Moreover, the combination of inositol and choline increased the number of repressed genes compared with inositol alone and enhanced the repression levels of a subset of genes that responded to inositol. In all, 110 genes were repressed in the presence of inositol and choline. Two distinct sets of genes exhibited differential expression in response to inositol or the combination of inositol and choline in wild-type cells. One set of genes contained the UAS INO sequence and were bound by Ino2p and Ino4p. Many of these genes were also negatively regulated by OPI1, suggesting a common regulatory mechanism for Ino2p, Ino4p, and Opi1p. Another nonoverlapping set of genes was coregulated by the unfolded protein response pathway, an ER-localized stress response pathway, but was not dependent on OPI1 and did not show further repression when choline was present together with inositol. These results suggest that inositol is the major effector of target gene expression, whereas choline plays a minor role.
Phospholipids are the key structural elements of membranebounded organelles and play important roles in signaling and membrane trafficking pathways. Each membrane compartment is composed of a unique set of phospholipids whose biophysical properties contribute to the function of each organelle. Phospholipid metabolism is highly regulated by the cell, ensuring the biogenesis and growth of membranes by coordinating the relative rates of synthesis of individual phospholipids with numerous factors, such as the availability of exogenous supplies of phospholipid precursors, growth stage, and membrane trafficking (1,2).
In many organisms, the regulation of lipid biosynthetic pathways is achieved through control of transcriptional regulatory networks. For example, lipid homeostasis in animal cells is controlled through the sterol regulatory element binding protein (SREBP) pathway (3). In the budding yeast Saccharomyces cerevisiae, a major control mechanism for the coordinated synthesis of inositol-and choline-containing phospholipids is the precise transcriptional control of genes for many of the enzymes required for phospholipid synthesis (2,4,5). The synchronized expression of these genes requires the participation of Ino2p and Ino4p, basic helix-loop-helix transcription factors that bind as a heterodimer to a cis-acting element in the promoters of these genes called UAS INO , as well as Opi1p, a negative regulator of transcription (6 -15). Many of these UAS INO -containing genes are maximally derepressed when the phospholipid precursors inositol and choline are absent from the growth medium in logarithmically growing cultures (16 -25), suggesting a common regulatory mechanism (4).
A recent collaborative study involving our laboratory has begun to reveal how this regulation is controlled by ongoing lipid metabolism (26). When inositol is added to the medium of growing cells, a dramatic change to the pattern of phospholipid synthesis is induced. Phosphatidic acid (PA), 1 a precursor for the synthesis of phosphatidylinositol, is rapidly consumed. This drop in PA levels is directly sensed by Opi1p, a component of an endoplasmic reticulum (ER)-localized lipid sensing complex (27), causing it to dissociate from the ER and translocate to the nucleus, where it participates in the repression of target genes. Although Opi1p does not directly interact with the UAS INO sequence (28,29), it is thought to interact with Sin3p (15), a histone deacetylase that functions as a global transcriptional repressor, and Ino2p, which potentially targets it to the promoters of UAS INO -containing genes. Thus, the transcriptional regulation of UAS INO -containing genes responds not to inositol directly but instead to a metabolic signal induced when inositol participates in lipid metabolism.
Cells grown in the absence of inositol also induce the unfolded protein response (UPR) pathway (30). The UPR pathway is an ER-localized signal transduction pathway that responds to the accumulation of unfolded proteins in the lumen of the ER as well as to secretory stress by up-regulating the expression of target genes (31)(32)(33)(34)(35). Although the mechanism for UPR induction under inositol-limiting conditions is unknown, Cox et al. (30) have suggested that the activation of the UPR might be directly involved in the mechanism by which INO1 transcrip-tion is activated. However, recent work from our laboratory has suggested that under certain conditions, UPR activation is not coupled to activation of UAS INO -containing genes (35,36).
Although much study of the regulatory effects of inositol and choline has focused upon the transcription of UAS INO -containing genes, their effects on the expression of other genes have been largely unexplored. A recent report (37) examined the combined effects of inositol and choline on genome-wide expression in yeast, but the individual contributions of inositol and choline on gene expression were not explored. Because inositol and choline have distinct effects on different branches of phospholipid biosynthetic pathways (4) (Fig. 1), an important unanswered question is whether the presence of these phospholipid precursors individually affects the expression of the same set(s) of genes. To identify all of the genes whose expression is regulated by inositol and choline, we carried out a genome-wide study using cDNA microarrays to find genes whose expression is regulated, both independently and jointly, by inositol and choline. We report that inositol and choline affect the expression of different sets of genes. Furthermore, we show that OPI1 primarily regulates genes that were previously shown by Lee et al. (38) to contain a bound Ino2p and Ino4p, and these genes are distinct from those genes regulated by the UPR pathway.
Our results indicate that inositol, not choline, is the major effector of Ino2p-Ino4p-and UPR-targeted gene expression. Together, these results suggest that distinct pathways contribute to regulate the expression of genes in response to changes in phospholipid metabolism induced by phospholipid precursors.
Cell Growth-All cultures were grown in synthetic complete growth medium lacking inositol and choline or containing combinations of inositol and choline as indicated at 30°C with shaking for at least 12 generations before harvesting at mid-log phase as described below. After overnight growth in indicated media at 30°C, strains were diluted 1:50 in the indicated media and grown for 5-6 h at 30°C. 250 ml of media were inoculated at an A 600 of 0.01 and allowed to grow at 30°C until the cultures reached mid-log phase growth (A 600 ϭ 0.5). Triplicates were prepared for each growth condition. Cells were harvested by filtration, immediately frozen on a dry ice/ethanol bath, and stored at Ϫ80°C. Filtration was chosen as the method for harvesting cells because the recovery of the INO1 transcript was the most reproducible using this procedure.
RNA Isolation and Microarray Analysis-Total RNA was extracted by the high temperature acid phenol method (41,42). For some experiments, total RNA was further purified using the RNeasy RNA purification kit (Qiagen, Inc.), or poly(A) RNA was isolated using oligo dT cellulose as described previously (41,42). mRNA was reverse-transcribed from 20 g of purified total RNA using 2 g of oligo dT (18-mer) primers (New England Biolabs) or 2.5 g of mRNA using 3 g of random (9-mer) primers (Invitrogen) by incubating with 400 units of SuperScript II (Invitrogen) plus 500 M concentrations of dATP, dCTP, and dGTP, 200 M amino-allyl dUTP (Sigma), and 300 M dTTP in a final volume of 40 l at 42°C for 2 h. After cDNA synthesis, the remaining RNA was hydrolyzed by adding 10 l of both 1 M NaOH and 0.5 M EDTA and incubating at 65°C for 15 min followed by neutralization with 10 l of 1 M HCl. The resulting cDNAs were purified using the QIAQuick PCR purification kit (Qiagen, Inc.), substituting kit buffers with KPO 4 buffers. Purified cDNAs were labeled with monofunctional reactive Cy3 or Cy5 dye esters (Amersham Biosciences) in the presence of Na 2 CO 3 pH 9.0, for 1 h at room temperature and subsequently quenched with 1 M NH 2 OH. After an additional QIAQuick purification, 20 pmol of Cy3-and Cy5-labeled cDNA probes were combined and hybridized to Corning CMT Yeast-S288c Gene Arrays (version 1.32) in 5ϫ SSC, 25% formamide, 2.5% SDS, and 100 g/ml salmon sperm DNA at 42°C for 14 -16 h. After washing, hybridized microarray slides were simultaneously scanned with lasers at 532-and 635-nm bandwidths using a GenePix 4000B array scanner (Axon Instruments, Inc.).
Statistical Analysis-Three replicates were analyzed for each experimental condition. Image analysis for each array was processed using the GenePix Pro 4.0 (Axon Instruments, Inc.) software package, which produces (R, G) fluorescence intensity pairs for each gene. After image acquisition, individual data spots on each microarray were visually inspected for size, signal-to-noise ratio, background level, and uniformity. Using these quality control criteria, about 30% of the spots for each triplicate set of experiments were discarded because of poor spot quality, a conventional practice for microarray data analysis (43). Normalization was conducted as follows: let M ϭ log 2 (R/G), A ϭ 0.5* log 2 (R*G).
FIG. 1. Yeast phospholipid biosynthetic pathways. Soluble phospholipid precursors: G-6-P, glucose-6-phosphate; I-3-P, inositol-3-phosphate; C-P, choline phosphate; CDP-C, cytidine diphosphate choline. Phospholipids: PA, phosphatidic acid; DAG, diacylglycerol; CDP-DAG, cytidine diphosphate diacylglycerol; PI, phosphatidylinositol; PS, phosphatidylserine; PE, phosphatidylethanolamine; PMME, phosphatidyl monomethylethanolamine; PDME, phosphatidyl dimethylethanolamine; PC, phosphatidylcholine. Arrows represent routes of metabolic conversion. The names of the structural genes for enzymes catalyzing specific metabolic conversions are shown adjacent to the arrows. Genes that were shown in the present study to be repressed in the presence of inositol and choline (Table II and  The log ratio M is known to be dependent on overall spot intensity A (44). To remove this systematic variation, intensity-dependent normalization was conducted for each array replicate. The intensity-dependent trend was fitted using the LOWESS fit function (45) in S-PLUS, after which the log ratio values (M) were normalized by subtracting the trend values. The student t test was performed on normalized M, for each gene for which three high quality replicates were available using the null hypothesis of no change in expression (i.e. normalization to M ϭ 0). A conservative significance level of p value equal to 0.025 corresponding to a false detection rate of q ϭ 0.075 was chosen. Hence the percentage of false positives among the significant tests is controlled to be at most 7.5%, a conservative approach that is expected to reduce false positive identifications to near zero. See Storey (46) and Storey and Tibshirani (47) for details on relating adjusted p values to false detection rate and the expected number of false positive tests. The t statistic was computed for each gene, and genes were then ranked according to their p values. Genes were considered to have differential expression in the red channel versus the green channel (or vice versa) if they exhibited a 2-fold change and also had a p value below 0.025.
Northern Slot Blot Analysis-Approximately 250 ng of unfractionated mRNA was heated at 65°C for 15 min in 3 volumes of denaturing buffer (66% formamide, 7% formaldehyde, 26 mM MOPS, 66 mM sodium acetate, and 1.3 mM EDTA), placed on ice after addition of an equal volume of ice-cold 20ϫ SSC, and spotted on BrightStar-Plus (Ambion, Inc.) nylon membranes using a manifold slot blot apparatus as described previously (41). Strand-specific 32 P-labeled riboprobes were synthesized from plasmids pJH310-INO1 (16), pSPACT (36), pBDG456-Ty1 (kind gift of D. Garfinkel), pBDG458-Ty2 (kind gift of D. Garfinkel), pSJ29-KAR2, pSJ30-PDI1, pSJ31-VTC3 and pSJ32-HO by in vitro transcription according to manufacturer's instructions (Promega). pSJ29-KAR2 was constructed by PCR-amplifying the KAR2 ORF and inserting the 878-bp HindIII-XbaI fragment into pGEM1 (Promega). pSJ30-PDI1 was constructed by PCR-amplifying the PDI1 ORF and inserting the 861-bp HindIII-SalI fragment into pGEM1. pSJ31-VTC3 was constructed by PCR-amplifying the VTC3 ORF and inserting the 703-bp PstI-XbaI fragment into pGEM1. pSJ32-HO was constructed by PCR-amplifying the HO ORF and inserting the 365-bp PstI-BamHI fragment into pGEM1. Membranes were hybridized with either INO1, KAR2, PDI1, VTC3, HO, ACT1, Ty1, or Ty2 probes in formamide hybridization buffer and washed to a final stringency of 0.2ϫ SSC/0.1% SDS at 60°C as described previously (16,41). Quantitation was performed by analysis on a STORM 860 PhosphorImager (Amersham Biosciences) and analyzed with ImageQuant software. The data were normalized by dividing the total counts per minute for the INO1, KAR2, PDI1, VTC3, HO, Ty1, or Ty2 probe by the total counts per minute for the ACT1 probe and expressed as the fraction of the amount of mRNA from cells grown in the absence of inositol and choline.
Construction of Gene Disruption Alleles-Complete disruptions of VTC1, VTC3, and VTC4 genes were constructed by PCR-mediated gene replacement as described previously (48) in the wild-type strain BY4742. The plasmid pFA6a-His3MX6 (kind gift of M. Longtine) was used as a template to generate PCR fragments for the gene disruptions. The entire open reading frame of each gene was replaced with the HIS3 marker gene. Histidine prototrophs were screened by colony PCR to verify integration at the correct genetic locus.
Assays for Ino Ϫ and Opi Ϫ Phenotypes-To test for Ino Ϫ phenotypes, strains were grown to mid-log phase in synthetic complete media containing 75 M inositol, washed once with water and 10-fold serial dilutions spotted on plates containing synthetic complete medium lacking inositol. Strains were incubated for 3 days at 18°C, 24°C, 30°C, and 37°C. To test for Opi Ϫ phenotypes (49, 50), strains were grown as described above and spotted on plates containing synthetic complete medium lacking inositol and incubated at 30°C and 37°C for 2 days. The plates were then sprayed with a suspension of a diploid tester strain (AID) homozygous for ino1 and ade2 and incubated for an additional 2 days.

Global Analysis of Inositol and Choline on Gene Expression
in Yeast-Previous studies have shown that the presence of phospholipid precursors in the medium of logarithmically growing cultures of Saccharomyces cerevisiae regulates the transcriptional response of a number of genes. These genes include those encoding the structural enzymes for phospholipid biosynthesis (4) as well as genes encoding protein folding chaperones induced by the UPR pathway (51). The goal of the present study was to identify, in a comprehensive manner, the complete set of genes whose expression levels change in response to the presence or absence of the phospholipid precursors inositol and choline using cDNA microarray technology.
The cDNA microarray experiments performed in this study are shown in Table I and are summarized below. First, to determine the individual effects of inositol and choline on genome-wide expression, the relative mRNA abundance from cells grown in the presence of either inositol or choline was measured by comparing with transcript levels from cells grown in the absence of both inositol and choline (Table I, experiments 1 and 2). Next, to characterize the combined effects of inositol and choline on genome-wide expression, the relative mRNA abundance from cells grown in the presence of both inositol and choline was compared with transcript levels from cells grown in the absence of inositol with or without choline (Table I, experiments 3 and 4). For every experiment, fluorescently labeled cDNA probes were synthesized from RNA extracted from cells grown to mid-logarithmic phase in chemically defined synthetic media and competitively hybridized to commercial spotted cDNA microarrays containing 6135 unique Saccharomyces cerevisiae open reading frames (ORFs). Each experiment was performed in triplicate. To assess which ORFs showed differential regulation from each experiment, rigorous statistical criteria were used as described under "Experimental Procedures." In brief, after normalization, the ratio for each spot from the array that showed a p value Յ0.025 over three replicates was considered to be differentially expressed. After statistical analysis, data collected from these experiments were compared and analyzed as described below. The complete data set is available at our laboratory website (www.mbg.cornell.edu/Henry_Lab.cfm).
In the first experiment (Table I, experiment 1), which measured the independent effects of inositol on genome-wide expression, a total of 32 genes showed a 2-fold or greater change in expression that met our statistical criteria ( Fig. 2A). Among these genes, 29 were down-regulated in the presence of inositol compared with cells grown in its absence. These genes are listed in Table II. As expected, the transcript levels of numerous genes previously shown to be affected by inositol levels were down-regulated when inositol was present, including INO1 (16), OPI3 (18, 21), ITR1 (23), and PSD1 (25). INO1 showed the greatest difference in expression, consistent with previous reports that this gene is the most highly regulated of  (2). A large number of genes involved in protein folding and secretion, including genes not previously shown to be regulated by inositol, were also downregulated when inositol was present. We were surprised to find that only three genes (HO, which encodes the endonuclease involved in mating-type switching, and two uncharacterized ORFs) showed the reverse pattern and were up-regulated in the presence of inositol (Table III), suggesting that the presence of inositol induces the expression of a very small number of genes.
In the next experiment (Table I, experiment 2), which measured the independent effects of choline on global gene expression, 32 genes showed a 2-fold or greater change in expression levels ( Fig. 2A). It is interesting that the pattern of expression was reversed from the pattern described in the previous experiment. Whereas a majority of the genes showing differential expression in response to inositol were down-regulated in its presence, a majority of those genes showing a change in expression when choline was present were up-regulated. In all, 31 genes were up-regulated in the presence of choline, whereas only a single gene was down-regulated (Table IV). Functional classification of the genes whose expression is affected by choline did not reveal any discernible pattern. For example, none of the detected genes were related to choline or lipid metabolism, such as phosphatidylcholine metabolism or choline transport. Furthermore, none of the genes whose expression was affected by choline showed any overlap with the genes differentially expressed in response to inositol, suggesting that choline, by itself, has no significant effect on the differential expression of genes repressed or activated in the presence of inositol. Thus, the global transcriptional effect of choline seems to be completely distinct from the effect of inositol.
However, it has been reported that addition of choline to medium containing inositol increases the extent of the repression of certain phospholipid biosynthetic genes compared with inositol alone. For example, it was originally reported that INO1 expression is repressed by an additional 2-3-fold in cells grown in inositol-containing medium when choline is present compared with cells grown in inositol-containing medium when choline is absent (16). Similar simultaneous effects of inositol and choline have also been observed for CHO1 (17,20), OP13 (21), CHO2 (21), CKI1 (52), and CPT1 (53). All of the above genes have been shown to contain an active UAS INO element in their promoters, suggesting that the additive repressive effect of choline in the presence of inositol might be specific to this particular set of genes. For this reason the UAS INO sequence has been referred to as the inositol/choline-responsive element (ICRE) (19). One prediction from the aforementioned studies is that the combination of inositol and choline might have a general repressive effect on the entire set of genes affected by inositol alone. To address this possibility, the combined effects of inositol and choline on genome wide expression were determined by comparing the relative mRNA transcript levels from cells grown in the presence of inositol and choline to transcript levels from cells grown in the absence of inositol and choline (Table I, experiment 3). These are conditions that have typically been used to measure the effect of phospholipid precursors on gene expression (16,19,37,52). The genes identified in experiment 3 were subsequently compared with the set of genes regulated by inositol alone (Table I, experiment 1). The number of genes showing a statistically significant change in expression level by 2-fold or greater increased from 32 when only inositol was present to 93 when choline was present in addition to inositol. The transcript levels of 68 genes were down-regulated, whereas 25 genes were up-regulated in the presence of both inositol and choline compared with the absence of inositol and choline ( Fig. 2A). Comparison of the set of genes regulated by the combined effects of inositol and choline with the set of 29 genes downregulated by inositol alone revealed an overlap of 21 genes that showed at least a 2-fold or greater change in expression levels (Fig. 2B). It is noteworthy that the down-regulated level of a subset of these overlapping genes was greater when cells were grown in the presence of inositol and choline, including INO1 and OPI3 (Table II), in agreement with previous reports (16,21). The presence of choline also had an additional repressive effect on other inositol-regulated genes including ITR1, YEL073C, YIP3, PSD1, and SAH1, which is reported here for the first time. However, other inositol-regulated genes did not show a significant change from their expression level when choline was present (Table II). For example, the relative change in expression of KAR2 and ERO1, genes whose expression is regulated by the UPR pathway (34,51), showed similar levels of repression in response to inositol, regardless of the presence or absence of choline in the growth media. KAR2 and ERO1 were repressed by 2.4-and 2.6-fold, respectively, in cells grown in the presence of inositol alone, and both were repressed by 2.3-fold when grown in the presence of inositol and choline. These results suggest that choline has an additional repressive effect on expression levels when inositol is present in the growth medium, but only on a subset of overlapping genes.
We were surprised to find that only two genes (PET127 and YLR460C) that were up-regulated in the presence of both inositol and choline overlapped with the set of genes that were up-regulated by choline alone (Fig. 2C), suggesting that inositol in combination with choline overwhelms the activating effect of choline alone on gene expression. To further examine the combined effects of inositol and choline on genome-wide expression,  Ϫ2 Actin-related protein the transcriptional response of cells grown in medium containing both inositol and choline was compared with cells grown in medium lacking inositol but containing choline (Table I, (Table II). Furthermore, approximately half of the up-regulated genes identified in experiment 4 overlapped with the set of genes up-regulated in experiment 3 (Table III), suggesting that these genes are affected by inositol but not choline. Many of the up-regulated genes identified in experiments 3 and 4 also overlapped with genes up-regulated in experiment 1, but the level of differential expression in experiment 1 was less than 2-fold in most cases (Table III). However, there was minimal overlap in experiments 3 and 4 with genes up-regulated by choline alone (Table IV). Taken together, these results suggest that inositol, but not choline, is the major effector of target gene expression.
To confirm the results from the cDNA microarray experiments for a subset of genes identified in experiments 1, 3, and 4, INO1, VTC3, KAR2, PDI1, ACT1, HO, Ty1, and Ty2 were chosen for further characterization by quantitative Northern slot blot analysis. ACT1, which is not regulated by inositol and choline, was chosen as a control. The relative levels of mRNA from these genes from wild-type cells grown in the absence of inositol and choline were compared with transcript levels for cells grown in the absence of inositol and presence of choline, presence of inositol and absence of choline, and presence of inositol and choline. In agreement with the data obtained from the microarray experiments, INO1, VTC3, KAR2, PDI1, and Ty2 were down-regulated in the presence of inositol. The relative levels were consistent with the microarray data (Table II, Fig. 3), and the repression levels for INO1 were also comparable with levels previously observed by slot blot analysis (16). In addition, HO was up-regulated in response to inositol, in agreement with the microarray results (Fig. 3). Moreover, unlike Ty2 transcript levels, Ty1 transcript levels were not significantly affected by inositol and choline (Fig. 3, see below), which corresponds with the microarray analysis.
These experiments indicate that the predominant effect on genome wide expression of inositol in the growth media is repressive, and that this effect is augmented by the addition of choline. Integration of the results obtained in experiments 1, 3, Analysis of Genome-wide Expression Regulated by Inositol and 4 revealed that the expression levels of 110 genes were repressed by at least 2-fold by the presence of inositol. The remainder of the report will focus on the analysis of this set of inositol-repressed genes, which are shown in Fig. 4 (Fig. 5). Genes involved in protein folding, including KAR2, SCJ1, ERO1, FPR2, MPD1, and PDI1, which have been shown to be regulated by the UPR pathway (34), were also detected (Fig. 5). Additional functional categories were also identified ( Fig. 5) and will be discussed below.

A Major Subset of Inositol-regulated Genes Interact Directly with Ino2p and Ino4p-Previous studies have shown that
Ino2p and Ino4p act as positive transcription factors by binding to the UAS INO element present in the promoters of a number of phospholipid biosynthetic genes (1). As a rationale for clustering of genes affected by the presence of inositol and choline, we reasoned that Ino2p and Ino4p should be bound to a subset of

Analysis of Genome-wide Expression Regulated by Inositol
the genes that are repressed by growth in inositol-containing medium. Lee et al. (38) have located the genomic binding sites of Ino2p and Ino4p. Throughout the remainder of this report, the set of genes reported by Lee et al. (38) that are bound by both Ino2p and Ino4p will be referred to as Ino2p-Ino4p target genes. We made use of this comprehensive data set by comparing the Ino2p-Ino4p target genes with the set of genes identified in our analysis that are regulated by inositol and choline. A total of 27 genes that were down-regulated by inositol alone or in combination with choline overlapped with the set of Ino2p-Ino4p target genes (Fig. 4). As expected, a number of genes identified in the current study that were previously shown to be regulated by inositol, including INO1 (16), OPI3  (55), were also reported to be bound by Ino2p and Ino4p in the study by Lee et al. (38). It is interesting that almost two thirds of the remaining inositol-repressed genes identified here as overlapping with the set of Ino2p-Ino4p target genes (38) had not been previously shown to be regulated by inositol and choline (Fig. 4). These newly identified inositol repressed genes include those involved in membrane trafficking (YIP3 and SRO77), methionine metabolism (SAH1), and lipid metabolism (OLE1 and ERG20). In addition to its function in lipid metabolism, ERG20, which encodes farnesyl-pyrophosphate synthetase, plays a role in protein prenylation and dolichol synthesis (56). These additional genes represent a novel set that are likely to be co-regulated with phospholipid biosynthetic genes. It is noteworthy that none of the genes that were up-regulated in cells grown in the presence of inositol or choline (Tables III and IV)    ment present in target genes (13). To ascertain the role of Opi1p on the genome-wide regulation in response to phospholipid precursors, the transcriptional response of an opi1⌬ strain was compared with its isogenic wild-ype in the presence of inositol (Table I, experiment 5). Of the 24 genes up-regulated by 2-fold or greater in the opi1⌬ strain, 19 overlapped with the set of genes negatively regulated by inositol, a set that includes genes that are further repressed by the presence of choline in addition to inositol (Table II). Furthermore, more than two thirds of these coregulated genes were reported by Lee et al. to contain a bound Ino2p and Ino4p (38) (Fig. 4), indicating that Opi1p primarily functions through genes regulated by Ino2p and Ino4p. Taken together, these findings corroborate the role of Opi1p as a negative transcriptional regulator of genes repressed by phospholipid precursors (7,26).
Walter and colleagues first reported that the UPR pathway is activated when cells are grown in the absence of inositol (30). When the UPR is triggered, the transcription of UPR target genes is induced to protect cells from secretory stress. Inositol starvation has also been shown to induce the expression of two UPR target genes, KAR2 (30) and PDI1 (57), as well as a UPRE-lacZ reporter construct (35,36). A genome-wide study of UPR target genes (34) identified genes including those required for folding and transport of proteins through every step of the secretory pathway. Comparison of the UPR target genes identified by Travers et al. (34) with the set of genes negatively regulated by inositol revealed 23 UPR target genes that were regulated in response to the presence of inositol (Fig. 4). These included many of the so-called canonical UPR genes involved in protein folding, including KAR2, EUG1, ERO1, PDI1, SCJ1, MPD1, and FPR2 (34), as well as numerous genes of unknown function. Because the expression levels of these genes were not influenced by the presence of choline in the growth media (Table II), these data show that the UPR pathway is activated solely as a consequence of the lack of inositol. Furthermore, the set of UPR target genes induced by inositol starvation did not overlap with the set of inositol-repressed Ino2p-Ino4p target genes (Fig. 4). Two Ino2p-Ino4p target genes (38), INO1 and OPI3, which are repressed in the presence of inositol (Fig. 4, Table II), have also been classified as UPR target genes (30,34). However, because both genes have been previously demonstrated to contain an active UAS INO element (8,21), we chose to classify these genes in the set of Ino2p-Ino4p target genes and omit them from the UPR target gene list for the purpose of our analysis.
Consistent with the previous report by Travers et al. (34), none of the inositol-regulated UPR target genes overlaps with the set of OPI1-regulated genes (Fig. 4), suggesting that activation of the UPR pathway represents a separate response to inositol starvation from the metabolic response that coordinates the regulation of the set of Ino2p-Ino4p target genes. As a further test of this hypothesis, the transcriptional profile of opi1⌬ and wild-type strains grown in the absence of inositol were compared and analyzed (Table I, experiment 6). Under these growth conditions Ino2p-Ino4p target genes will be expressed in both strains. They will be expressed in opi1⌬ cells because the Opi1p negative regulator is missing and in wildtype cells because inositol is absent (2, 7). As expected, none of the Ino2p-Ino4p target genes was differentially expressed to a significant degree in the opi1⌬ strain compared with the wildtype strain grown in the absence of inositol. However, opi1⌬ cells also overproduce and excrete inositol into the growth medium (7,49,58). For this reason, any inositol-dependent but non-Ino2p-Ino4p target genes are expected to be repressed in opi1⌬ cells regardless of inositol supplementation compared with wild-type cells grown in the absence of inositol. Indeed, several UPR genes (ERO1, MCD4, SIL1, ATG17, DOG1, and MNN4 (Table II)) were down-regulated in the opi1⌬ strain compared with the wild-type strain grown in the absence of inositol. Thus, in an opi1⌬ strain, Ino2p-Ino4p target genes are constitutively expressed, but the UPR is not constitutively activated.
In a final control experiment, the gene expression in the opi1⌬ strain grown in the absence of inositol was compared with expression in the same strain grown in the presence of inositol (Table I, experiment 7). As discussed above, it is expected that the addition of inositol to the medium should make little difference to genome wide expression in an opi1⌬ strain because these cells excrete inositol into the growth medium. Indeed, no significant changes in gene expression were detected when opi1⌬ cells were grown in the absence compared with the presence of inositol. Taken together, the results of experiments 5, 6, and 7 suggest that Opi1p negatively regulates Ino2p-Ino4p target genes but not UPR target genes.
Other Categories of Genes Regulated by Inositol-Functional classification of genes repressed in the presence of inositol included a subset of genes that are coregulated by the PHO pathway, including PHO81, PHO84, VTC1, VTC3, and VTC4 (59) (Fig. 5). The PHO pathway, which consists of at least 22 genes (59), is induced under low phosphate conditions to allow the cell to survive when phosphate is limiting. Because the cells were grown under high phosphate conditions in the experiments reported here, these results show that a subset of PHO pathway genes are induced when inositol is limiting. PHO81 encodes a cyclin-dependent kinase that initiates the PHO pathway (60), and PHO84 encodes a high affinity inorganic phosphate transporter (61). VTC1, VTC3, and VTC4 gene products are involved in polyphosphate accumulation by the vacuole (59) but also function together in a complex during late stages of vacuole fusion (62)(63)(64). It is interesting that two additional  (38), OPI1-regulated genes (experiment 5), and UPR target genes (34) are categorized. Criteria for classifying Ino2p-Ino4p and UPR target genes are described under "Results." Genes whose repression levels were enhanced in the presence of choline (Table II) are underlined and in boldface. Genes that did not fit into these categories are classified as "Other." genes, PBI2 and YHR138C, which function in early stages of vacuole fusion (65) but not the PHO pathway, were also identified in the present study. Taken together, these results suggest a combined function for these genes under conditions of nutrient-limited growth. Given their overlapping roles in storage of polyphosphate and maintenance of vacuole morphology, VTC1, VTC3, and VTC4 genes may also play a role in inositol storage in the vacuole. If the vacuole is an important inositol storage compartment, it is possible that disruption of VTC1, VTC3, or VTC4 might lead to growth defects in the absence of inositol. However, neither vtc1⌬, vtc3⌬, nor vtc4⌬ mutants showed diminished growth in the absence of inositol (Ino Ϫ phenotype) or an overproduction of inositol (Opi Ϫ phenotype) (data not shown).
Functional classification also revealed eight Ty2 element TYA GAG ORFs to be regulated by inositol and choline (Fig. 5). Ty elements are members of a family of eukaryotic elements called retrotransposons that resemble animal retroviral genomic RNA in structure (66). S. cerevisiae has five distinct retrotransposon families, designated Ty1-Ty5 (67). It has been reported previously that the expression of Ty1-H3, a Ty1 element retrotransposon closely related to Ty2 elements (67), is regulated by inositol and choline (68) and contains a UAS INO sequence in its promoter (1). However, whereas Northern slot blot analysis confirmed that Ty2 element transcript levels are ϳ4-fold lower in cells grown in the presence of inositol and choline compared with strains grown in the absence of inositol and choline (Fig. 3), Ty1 element transcript levels were not significantly altered in cells grown in the absence or presence of inositol and choline (Fig. 3). Retrotransposons can act as mutagens that may allow cells to adapt to stressful environments (69). Given that the UPR, a stress response pathway, is activated when cells grown in the absence of inositol, it is tempting to speculate that mobilization of Ty elements might also be related to stress produced by growing cells in media lacking inositol. However, because only a single Ty2 GAG ORF was detected as a UPR target (34), activation of the remaining seven elements might indicate a stress response in addition to the UPR pathway in cells grown in the absence of inositol. DISCUSSION We have presented a comprehensive genome wide expression analysis designed to identify the sets of genes whose transcript levels in dividing cells are influenced by the presence of the phospholipid precursors, inositol, and choline. The independent effects of inositol and choline on genome-wide expression have not been examined previously and are reported here for the first time. Our findings indicate that inositol and choline affect the expression of completely distinct sets of genes. Moreover, whereas the effect of inositol on overall gene expression was primarily repressing ( Fig. 2A), the effect of choline on target gene expression was activating ( Fig. 2A). However, when choline was present in medium also containing inositol, many of the genes affected by inositol alone exhibited intensified repression (Table II). Not only was the degree of repression of certain genes enhanced but also a greater number of genes were found to be differentially expressed in the presence of both precursors. This is in agreement with previous reports measuring expression of individual genes, including INO1 (16), CHO1 (17,20), OP13 (21), CKI1 (52). We also show for the first time that ITR1, YEL073C, YIP3, PSD1, and SAH1 match this pattern of regulation. Then again, only a few genes that were up-regulated in the presence of choline alone were also upregulated when both inositol and choline were present (Table  IV). Overall, our results indicate that inositol plays the primary role in repressing the expression of genes sensitive to the presence of exogenous phospholipid precursors, whereas choline plays a minor role. Choline primarily enhances the repressive effect of inositol on a subset of inositol-responsive genes. A possible metabolic mechanism for these observations will be discussed subsequently.
A Distinct Set of Inositol-Repressed Genes Regulated by the Transcription Factors Ino2p and Ino4p-Several distinct classes of genes were repressed in response to the presence of inositol or inositol plus choline in the growth media. One set included genes, previously identified by Lee et al. (38), that contain a bound Ino2p and Ino4p. Ino2p and Ino4p are members of the family of basic helix-loop-helix transcription factors, both of which are required for depression of INO1 and other coregulated genes encoding enzymes of phospholipid biosynthesis (9,11,12). Together, these transcription factors form a heterodimer and bind to the repeated element UAS INO found in the promoters of target genes (1,12). Using genome-wide location analysis (70), a method to identify genomic protein-DNA interaction sites using chromatin immunoprecipitation followed by DNA microarray analysis, Lee et al. (38) identified the genomic binding sites of 106 yeast transcriptional regulators, including Ino2p and Ino4p. The number of promoter regions containing both Ino2p and Ino4p using a p value threshold of 0.01 for both factors was 189 (38). As expected, virtually all of the genes involved in lipid metabolism that were down-regulated in the presence of inositol were also detected by Lee et al. (38) as Ino2p-Ino4p target genes (Fig. 5). Many of these genes FIG. 5. Functional categories of inositol-repressed genes. Inositol-dependent genes shown in Fig. 4 that were functionally annotated in the Saccharomyces Genome Data base (SGD) (www. yeastgenome.org) were categorized based upon their gene ontology annotation. Each section of the pie chart represents a functional category. The numerals refer to the number of members of each particular category. Bold headings indicate functional categories. Genes were assigned to subcategories according to summaries of published data listed in the SGD. (*Ino2p-Ino4p target genes identified by Lee et al. (38); ϩUPR target genes identified by Travers et al. (34)). Genes falling into categories with 2 or fewer members were grouped into the category "Other." Genes with no functional annotation were grouped into the category "Unknown." had previously been shown to contain an active UAS INO element in their promoters (1), confirming that both Ino2p and Ino4p play a pivotal role in the regulated expression of these genes.
A recent and more thorough genome-wide location analysis study conducted by Harbison et al. (71) identified the genomic occupancy of all 203 known yeast transcription factors and the recognition motifs for 116 of these regulators. Harbison et al. (71) refined the statistical analysis of bound genes identified by Lee et al. (38) and confirmed that Ino2p and Ino4p bind to target genes as regulator pairs through the UAS INO sequence (12). Using a more stringent significance threshold for genes bound by Ino2p and Ino4p from the study by Harbison et al. (71), a portion of the Ino2p-Ino4p target genes reported by Lee et al. (38) that are repressed by inositol shown in Fig. 3 did not show significant binding to Ino2p. These genes include CKI1, ERG20, HNM1, PBI2, TYE7, CHO1, NRG2, PHO81, and RMD12. One explanation for the failure to detect Ino2p binding to these genes is that Ino2p levels might be reduced in cells grown in the YPD medium used by Harbison et al. (71). The current study (Table II) and previous studies (55,72) showed that INO2 expression is down-regulated in the presence of inositol, which is present in YPD media. Moreover, CKI1, HMN1, and CHO1 each contain a UAS INO element in their promoters (1) and match the pattern of regulation of other UAS INO -containing genes (17,20,52,54). Thus, it is likely that both Ino2p and Ino4p are required for their regulated expression.
In addition, our results show that Ino2p-Ino4p target genes significantly overlap with the set of OPI1-regulated genes. In the first indication that Opi1p physically interacts with any gene, a subset of the genes we identified as OPI1-regulated were shown by Harbison et al. (71) to be bound by Opi1p, including INO1, CDS1, and ITR1. Opi1p is localized at steady state to the ER in an inactive form as part of a lipid-sensing complex (27). Addition of inositol to the media of growing cells results in a rapid change in lipid metabolism, causing Opi1p to translocate to the nucleus (26). Although Opi1p does not seem to directly interact with the UAS INO element (28,29), Opi1p is potentially targeted to UAS INO -containing genes by a direct interaction with Ino2p, where it represses the transcription of genes by recruiting the histone deacetylase Sin3p (15). These observations suggest that Opi1p plays a central role in sensing metabolism and modulating the regulation of Ino2p-Ino4p target gene expression.
Additional Potential Regulatory Mechanisms That Mediate the Differential Expression of Genes Regulated by Inositol-Aside from their role in regulating the transcription of genes that are repressed in the presence of inositol, Ino2p and Ino4p may also indirectly affect the transcription of genes that are activated in the presence of inositol by functioning through a transcriptional regulatory network. For example, both Ino2p and Ino4p bind to the SWI5 gene in YPD media containing inositol (38). SWI5 encodes a transcription factor that regulates the expression of genes expressed during the G 1 phase and the M/G 1 boundary of the cell cycle (73). Whereas the differential expression of SWI5 was not detected in our study, five genes that were activated in the presence of inositol (HO, PRY3, SCW11, WSC4, and YML131W (Table III)) have been shown by Lee et al. (38) and Dohrmann et al. (73) to be bound by Swi5p. The HO endonuclease gene, which showed the greatest differential expression in the presence of inositol, is regulated by Swi5p (73), and HO, SCW11, and PRY3 are regulated in a cell cycle-dependent manner (73,74). Thus, Ino2p and Ino4p may contribute to the regulation of gene expression by controlling the expression of transcription factors that in turn regulate cell cycle specific genes.
The expression of other genes may be regulated in response to the enzymatic activity of genes directly regulated by Ino2p and Ino4p. For example, the product of the INO1 gene, inositol-3-phosphate synthase, catalyzes the formation of inositol-3phosphate from glucose-6-phosphate in a reaction that is absolutely dependent upon NAD (75). It is interesting that BNA2, which is required for the de novo synthesis of NAD, is highly expressed when cellular NAD levels are depleted (76) and in cells grown in the absence of inositol (Table II). Because INO1 levels are dramatically elevated in cells grown in the absence of inositol (16) (Table II), overall cellular levels of NAD may drop as a result of consumption by increased activity of inositol-3phosphate synthase, activating a previously undetected feedback regulatory mechanism.
In contrast to the results presented here, a recent study by Santiago and Mamoun (37), which investigated the combined effects of inositol and choline on genome-wide expression in yeast, concluded that Ino2p and Ino4p regulate only a minor subset of genes affected by these phospholipid precursors. There are several differences between our study and that of Santiago and Mamoun in the both experimental design and findings. First, Santiago and Mamoun (37) detected none of the UPR target genes and identified only three genes involved in lipid biosynthesis as responding to the combined effects of inositol and choline. Second, they concluded that only two genes, OPI3 and YDL241W, require Ino2p and Ino4p for inositol-and choline-dependent repression. However, neither INO1 nor any other gene besides OPI3, previously demonstrated to contain an active UAS INO element (1), was identified in their study as Ino2p-and Ino4p-dependent. We believe that their failure to identify genes such as INO1, which is clearly Ino2pand Ino4p-dependent, is most probably the result of the design of their experiment. Santiago and Mamoun (37) compared gene expression in ino2⌬ and ino4⌬ cells to wild-type cells, all of which were grown in the presence of both inositol and choline. In the present study, we have shown that Ino2p-Ino4p target genes are fully repressed in wild-type cells grown under these conditions (Table II). Because the ino2⌬ and ino4⌬ strains each lack an activator required for the derepression of target genes, these mutant strains are expected to express only repressed levels of the affected genes, regardless of the growth condition. Santiago and Mamoun (37) obtained the expected, but uninformative, result because the expression levels of Ino2p-Ino4p-dependent genes in ino2⌬ and ino4⌬ cells did not show significant differences compared with wild-type cells. Finally, these authors identified numerous genes that were not detected in our study, including genes involved in biotin synthesis and nutrient transport. For example, GIT1, which encodes a plasma membrane glycerophosphoinositol permease, was detected as an inositol-and choline-responsive gene in their study. However, GIT1 is known to be expressed only when both inositol and phosphate are limiting (77,78). In fact, GIT1 is fully repressed regardless of the presence or absence of inositol in high phosphate media (77,78), such as that based on standard yeast nitrogen base formula media, comparable with the media that we used in the present study. In addition, many other genes identified by Santiago and Mamoun (37) overlap with genes shown by Wodicka et al. (79) to be up-regulated in minimal media. Thus, essential nutrients, including vitamins and phosphate, may have been limiting or absent in the growth media used in the study by Santiago and Mamoun (37).
A Distinct Regulatory Network for Regulating UPR Target Genes-We have shown that a subset of genes reported by Travers et al. (34) to be regulated by the UPR pathway are also activated in the absence of inositol. The UPR is a stress response pathway that is activated under conditions in which insufficient protein folding capacity leads to an accumulation of unfolded proteins in the ER lumen, a condition referred to as ER stress (80). In yeast, UPR activation induces the transcription of target genes that increase the capacity for protein folding and secretion as well as genes for degradation of unfolded polypeptides (34). In all, 381 UPR target genes were identified in the genome-wide analysis by Travers et al. (34) of cells treated with tunicamycin and dithiothreitol, conditions that are known to induce the UPR pathway. In the current study, we found 23 UPR target genes in the set of genes repressed in the presence of inositol. Although we detected many canonical UPR target genes, we did not detect other targets of the UPR pathway, including genes involved in ER-associated degradation or membrane trafficking pathways. It is possible that the UPR response in cells grown in the absence of inositol is less robust than when it is induced by tunicamycin or dithiothreitol.
It is noteworthy that the set of UPR genes that is induced in cells grown in the absence of inositol does not overlap with the set of genes shown to contain a bound Ino2p-Ino4p by Lee et al. (38) (Fig. 4). Furthermore, the expression of the UPR target genes identified here was not affected further by the presence of choline in the growth media. These results strongly suggest that induction of UPR target genes under inositol limiting conditions is through an entirely different mechanism than genes regulated by Ino2p and Ino4p. The relationship between UPR activation with inositol starvation and Ino2p-Ino4p target gene expression is clearly complex and worthy of continued analysis.
A Mechanism for Regulating the Expression of Inositol-responsive Genes by a Metabolic Signal-Previous studies have suggested that the expression of INO1 and other UAS INOcontaining genes are regulated by a common metabolic signal that is produced by the participation of inositol and choline in lipid metabolism (4), even though these precursors enter the pathways for phospholipid biosynthesis by distinctly different routes (Fig. 1). Inositol reacts directly with CDP-diacylglycerol (CDP-DAG) to form phosphatidylinositol (PI), whereas choline is phosphorylated and converted to CDP-choline before it reacts with diacylglycerol (DAG) to form phosphatidylcholine (PC) via the Kennedy (CDP-choline) pathway. PC is also synthesized via the three-step phosphatidylethanolamine methylation pathway (Fig. 1). Phosphatidylethanolamine is derived from phosphatidylserine, which is produced in a reaction that competes directly for CDP-DAG with PI biosynthesis (81). Because DAG and CDP-DAG intermediates are derived from PA ( Fig.  1), PA is the common lipid precursor for the production of PI and both routes of PC biosynthesis. Moreover, PA is sensitive to the presence of inositol and choline because it is consumed in the production of inositol-and choline-containing phospholipids (4,26,81).
Several lines of evidence suggest that PA levels provide the relevant signal for the repression of Ino2p-Ino4p target genes. First, the signal controlling INO1 expression, and presumably other UAS INO -containing genes, is sensitive to ongoing metabolism in all three of the interconnected pathways leading to PI and PC synthesis described above (Fig. 1). Moreover, input from these pathways seems to be additive. As shown in the current study and previous investigations of individual Ino2p-Ino4p target genes, the presence of inositol alone is enough to trigger repression (Table II). However, many Ino2p-Ino4p target genes, including INO1, CHO1, OP13, CKI1, ITR1, YEL073C, YIP3, PSD1, and SAH1 (16,17,20,21,52) (Table II), show enhanced repression when choline is present with inositol, supporting the hypothesis that these two precursors affect a common signal in an additive fashion. Because PA is the common precursor at the branch point for the synthesis of PI from CDP-DAG and PC from DAG (Fig. 1), it is a logical candidate for the relevant signaling molecule.
Second, mutations causing complete or partial blocks at any step in the de novo synthesis of PC from PA through the phosphatidylethanolamine methylation pathway (Fig. 1) lead to the derepression of INO1 and other UAS INO -containing genes even in the presence of inositol (4). In these mutants, derepression of INO1 occurs even when inositol is present, and they have an Opi Ϫ phenotype when inositol is absent. Because each block in the pathway leads to the buildup of precursors upstream of the specific metabolic lesion, including PA, derepression of Ino2p-Ino4p target genes in these mutant strains is correlated to a rise in PA (4). Moreover, supplementation with choline restores PC biosynthesis, eliminates the Opi Ϫ phenotype, and restores INO1 regulation in response to inositol in cho2 and opi3 mutants (4). Because choline enters via the Kennedy pathway and uses DAG as a precursor (Fig. 1), these results indicate that ongoing metabolism in the two routes to PC biosynthesis have additive effects on INO1 expression and that they influence the same signal as inositol. The PA signal hypothesis gained strong support from the finding that sec14 ts mutants that are also blocked in the Kennedy (CDP-choline) pathway overexpress INO1 even in the presence of inositol (82). This was shown to be caused by increased PC turnover by a phospholipase D (Pld1p)-dependent mechanism (83), which leads directly to increased PA levels (Fig. 1).
Finally, a recent collaborative study involving our laboratory has revealed the molecular mechanism whereby PA levels regulate the expression of INO1 (26). Opi1p, which is a member of an ER-localized lipid-sensing complex (27), was shown to interact directly with PA (26), suggesting that Opi1p serves as the relevant PA sensor in the ER. The introduction of inositol was shown to result in a rapid rise in PI biosynthesis and a fall in PA levels. Concurrent with the drop in PA levels, Opi1p rapidly translocated from the ER to the nucleus, which coincided with INO1 repression (26). The results presented in the current study predict that this mechanism can be extended to all Ino2p-Ino4p-bound genes that are coregulated by Opi1p.
A remaining unanswered question is why choline alone has little or no effect on expression of Ino2p-Ino4p target genes in wild-type cells in the absence of inositol. One testable prediction is that the effect of choline on PA levels is less than that of inositol, whereas the combination of inositol and choline produces an enhanced effect on PA levels. Moreover, the relative effects of these two precursors on PA levels should be reversed in mutants such as cho2 and opi3 with respect to their effects on both gene expression and PA levels. We are currently testing these models.