Research Article

Phospholipid Metabolism Regulated by a Transcription Factor Sensing Phosphatidic Acid

See allHide authors and affiliations

Science  11 Jun 2004:
Vol. 304, Issue 5677, pp. 1644-1647
DOI: 10.1126/science.1096083

Abstract

Cells regulate the biophysical properties of their membranes by coordinated synthesis of different classes of lipids. Here, we identified a highly dynamic feedback mechanism by which the budding yeast Saccharomyces cerevisiae can regulate phospholipid biosynthesis. Phosphatidic acid on the endoplasmic reticulum directly bound to the soluble transcriptional repressor Opi1p to maintain it as inactive outside the nucleus. After the addition of the lipid precursor inositol, this phosphatidic acid was rapidly consumed, releasing Opi1p from the endoplasmic reticulum and allowing its nuclear translocation and repression of target genes. Thus, phosphatidic acid appears to be both an essential ubiquitous metabolic intermediate and a signaling lipid.

Lipid synthesis in widely divergent organisms is coordinately regulated by transcriptional feedback loops (14). An important class of transcription factors that regulate lipid metabolism have inactive precursors anchored in the endoplasmic reticulum (ER) (13). Although it is clear that these transcription factors respond to changes in lipid concentrations by being released from the ER and translocating to the nucleus, it is not known precisely how such transcription factors are coupled to changes in the lipid composition of the ER membrane. In yeast, a high proportion of genes involved in phospholipid metabolism are regulated by the same set of three transcription factors (Fig. 1A) (5, 6). A cis-acting inositol-sensitive upstream activating sequence (UASINO) common to these genes is activated by a pair of transcription factors, Ino2p and Ino4p, that in turn are repressed by Opi1p (7). Inositol is a master regulator of this pathway; it activates Opi1p, thereby creating a negative feedback loop because the most highly regulated gene in the pathway is INO1, which codes for the rate-limiting enzyme in inositol synthesis (Fig. 1A). Thus, deletion of OPI1 leads to unrepressible overproduction of inositol (hence called “opi”). Opi1p is a cytoplasmic protein that contains two phenylalanines in an acidic tract (a FFAT motif) that binds the integral ER membrane protein Scs2p and thus targets it to the ER (8). However, both the signal generated by inositol and the mechanism by which the signal is sensed by Opi1p are not known.

Fig. 1.

Rapid transmission of an inositol-based signal to the nucleus by Opi1p. (A) Phospholipid biosynthesis in yeast. Reactions (solid arrows) and membrane transporters (open arrows) that are coordinately activated by Ino2p/Ino4p and repressed by Opi1p are colored red. CL indicates cardiolipin. (B) (Left) Groups of yeast expressing GFP-tagged Opi1p grown without inositol (0 min) or supplemented with inositol (100 μM) and imaged at the indicated times. (Right) Sequential images of two cells between 1 and 6 min after addition of inositol (100 μM), during which time GFP-Opi1p relocalized from the periphery (arrowheads) to the nucleus (arrows). (C) Yeast carrying the temperature-sensitive allele prp20-1 (RAN-GEFts), expressing GFP-Opi1p, and grown without inositol were shifted to the nonpermissive temperature (37°C) and imaged before and 15 min after addition of inositol (100 μM). (D) INO1 mRNA transcripts extracted from yeast after the addition of inositol (75 μM), were quantified on Northern blots and normalized to the amount of ACT1 transcripts. The amount of INO1 transcripts decreased, with a half-life of 10 min and with little initial lag. Results are the average of two experiments (vertical bars indicate range).

Results. In log-phase cells growing without inositol, green fluorescent protein (GFP)–Opi1p was targeted to the ER, which in yeast consists of a peripheral network beneath the plasma membrane, the nuclear envelope (NE) on which GFP-Opi1p was slightly enriched, and occasional linking strands (Fig. 1B) (8). To study the regulation of Opi1p by inositol, we visualized GFP-Opi1p after the addition of inositol. Within 5 min GFP-Opi1p relocalized away from the cell periphery to the nucleus (Fig. 1B). The increase in soluble nuclear Opi1p slightly preceded accumulation on the NE (fig. S1, A to D) (9), the latter remaining for over 2 hours (Fig. 1B). Inhibition of the major pathway of nuclear export (mediated by Crm1p) had no effect on the peripheral localization of GFP-Opi1p (fig. S1E), indicating that Opi1p is not constitutively recycling from ER to nucleus. In contrast, complete inhibition of nucleocytoplasmic traffic reduced the enrichment of GFP-Opi1p on the NE and prevented its translocation in response to inositol (Fig. 1C). Thus, inositol appeared to activate nuclear import of Opi1p, possibly after it had been displaced from the ER into the cytoplasm, and the enrichment of GFP-Opi1p seen on the NE was likely to be intranuclear (i.e., on the inner NE). The time course of inositol-mediated repression is not known (10, 11), so we examined the short-term effects of inositol on transcription of INO1. After addition of inositol, INO1 mRNA amounts declined, with little initial lag time (Fig. 1D). Thus, nuclear translocation by GFP-Opi1p had the same kinetics as inositol-mediated gene repression, suggesting the physiological relevance of the translocation.

We next set out to identify the signal sensed by Opi1p. First, we tested whether Opi1p sensed inositol directly with the use of cells carrying a mutant phosphatidylinositol (PI) synthase gene that limits incorporation of inositol into PI (12). Translocation of GFP-Opi1p was slow and partial (Fig. 2A), indicating that the inositol-derived signal required production of PI. Downstream of PI, yeast synthesize four PI phosphates (PIPs); however, we found that production of PIPs was not required for Opi1p translocation (fig. S2A). Alternatively, the signal detected by Opi1p might be the production of PI or the consumption of its precursors, phosphatidic acid (PA) and cytidyldiphosphate diacylglycerol (CDP-DAG), biosynthetic intermediates (Fig. 1A) that are present at only a few percent of total phospholipid (13). To test whether inositol affected these lipids, we labeled phospholipid in vivo over short time periods with 32P-orthophosphate. In the absence of inositol, within 2 min phosphate had passed through several intermediates to label both PA and CDP-DAG (Fig. 2B). During the first 10 min of labeling in the absence of inositol, label accumulated in CDP-DAG, which was gradually converted to phosphatidylserine (PS), PI, and some phosphatidylethanolamine (PE). However, this pattern changed when inositol was added: Newly synthesized PA and CDP-DAG rapidly declined, and label then accumulated in PI (Fig. 2B). The same pattern was also seen with simultaneous addition of inositol and 32P-phosphate (fig. S2B), indicating that addition of inositol caused consumption of newly synthesized PA and CDP-DAG in favor of PI (Fig. 2C). Thus, translocation of Opi1p was temporally related to large changes in concentrations of newly synthesized PA, CDP-DAG, and PI.

Fig. 2.

Inositol-induced alteration in phospholipid synthesis. (A) Role of PI synthesis in Opi1p translocation. (Top) In response to inositol (50 μM), GFP-Opi1 expressed in wild-type cells translocated completely within 12 min. (Bottom) In cells carrying a hypoactive mutant allele of PI synthase (12), GFP-Opi1p was still seen on peripheral ER at this time and only translocated in a minority of cells at later time points (arrowheads). (B) Newly synthesized glycerophospholipids were quantified 2, 5, and 10 min after addition of 32P-H3PO4 to cells growing in log phase in the absence of inositol. At 10 min, inositol (75 μM) was added (arrow), and cells were sampled 2, 5, and 10 min later. Lipids were extracted and quantified after two-dimensional chromatography (9). (C) Diagram of the rapid effects of inositol on phospholipid metabolism. Addition of inositol (labeled 1) leads to rapid PI synthesis, which consumes the immediate precursors PA and CDP-DAG (labeled 2).

If Opi1p directly or indirectly senses a signal from these changes in PA, CDP-DAG, or PI, the most simple explanation is that Opi1p is itself a lipid sensor. Because of a long-standing prediction that PA concentrations are intimately involved in regulating phospholipid metabolism (6), we tested binding of Opi1p to PA. COS-7 cell lysates containing GFP-Opi1p were incubated with beads coated with PA (14). Opi1p bound to the PA-coated beads, whereas binding to beads coated with another anionic phospholipid, PI(4,5)P2, was negligible (Fig. 3A). Additionally, Opi1p extracted from either COS cells or yeast bound specifically to PA-rich membranes (fig. S3, A and B). To exclude the possibility that this binding was indirect via protein partners, we purified Opi1p from bacteria and incubated it with liposomes made from phosphatidylcholine (PC) and PA. Binding was saturable, with an approximate Kd of 50 μM, whereas there was no binding to liposomes containing PC alone or mixed with CDP-DAG, PI, or PS (Fig. 3B). In addition, there was no interaction with phosphatidylglycerol (PG) or PI4P and a weak interaction with diacylglycerol pyrophosphate (DGPP) (fig. S3C). To test whether PA could recruit Opi1p to membranes in vivo, we microinjected mammalian cells expressing GFP-Opi1p with bacterial phospholipase D (PLD) to convert PC into PA. In uninjected cells, GFP-Opi1p was diffusely intranuclear, but upon injection it translocated to the NE (Fig. 3C). Thus, Opi1p bound specifically to PA and detected membrane pools of PA in vivo.

Fig. 3.

Interactions of Opi1p with phospholipids. (A) Detergent lysate of COS cells expressing GFP-Opi1p was incubated with agarose beads coupled to PA or PI 4,5-bisphosphate (PIP2) as described previously (14). Bound material, together with 5% of the input, was immunoblotted for GFP. (B) Maltose binding protein (MBP)–Opi1 fusion protein (25) was incubated with different concentrations of liposomes containing PC alone or PC mixed with the indicated lipid (9). Bound material, together with 100% of input material, was analyzed on Coomassie-stained SDS–polyacrylamide gels. Concentrations of lipid available for binding were 25, 50, and 100 μM (top) and 25 and 62.5 μM (bottom). (C) COS-7 cells expressing GFP-Opi1p uninjected or after microinjection of bacterial PLD (bPLD), which induced NE translocation, not seen with GFP alone.

Scs2p is required for normal localization of Opi1p (8), and deletion of SCS2 represses INO1, causing a mild inositol auxotrophy (15). Here, we found a molecular mechanism that explains the relationship between SCS2 and INO1: Reduced targeting to the ER, normally mediated by Scs2, activated Opi1p (fig. S4). In a similar manner, reduction of PA-mediated targeting to the ER might be expected to activate Opi1p. To investigate this, we attempted to create a mutant of Opi1p with reduced PA binding. Among known examples of PA binding proteins, there is no motif for PA binding other than the presence of one or more basic residues (14, 1622). To identify the PA binding site, we subdivided Opi1p, first into Opi1N and Opi1C (Fig. 4A). Both halves showed moderate affinity for immobilized PA (fig. S5A). In addition, bacterially expressed Opi1N and Opi1C both bound to PA liposomes (Fig. 4B), indicating the presence of at least two PA binding sites in Opi1p, a feature also found in the protein tyrosine phosphatase SHP-1 (16). On subdivision of Opi1 into quarters (Q1 to Q4), Q2 bound PA in vitro (Fig. 4C), but Q1 did not; neither did Q3 and Q4, making it difficult to dissect the C-terminal PA binding site (fig. S5B). Q2 contains a 30-residue basic domain with 11 lysines and arginines (7), which we divided into groups A, B, and C (Fig. 4A). Substitution with alanine of the three residues in group A (Q2A3m) or the five residues in group B (Q2B5m) had a marginal effect on PA binding (reduced to 71% and 77%, respectively), whereas mutation of three residues in group C (Q2C3m) had a stronger effect (binding at 45% of wild type) (Fig. 4C). Although GFP-Q2 expressed in COS-7 cells was intranuclear like Opi1p, GFP-Q2C3m was mainly cytoplasmic and responded to microinjection of bacterial PLD by translocating to the ER and NE (Fig. 4D) despite its reduced PA binding. Thus, PA could mediate membrane recruitment of Q2, and one of the PA binding sites in Opi1p mapped to the basic domain in Q2.

Fig. 4.

Analysis of PA binding sites in Opi1p. (A) Domains in Opi1p. Two protein interactions have been mapped: the Sin3-interacting domain (Sin3ID) (29) and the FFAT motif (8). Other features are the basic domain (bd), leucine zipper (lz), and polyglutamine repeat (polyQ) (7). (Top) Segments subcloned and (bottom) basic domain sequence (30), with acids and bases colored red and blue, respectively. The 11 bases were grouped together for mutation to alanines (A3, B5, and C3). (B) Binding of MBP-Opi1N and MBP-Opi1C (arrows) to PA liposomes. Binding was weaker than that for full-length Opi1p (Fig. 3B). A breakdown product (asterisk) acted as an internal negative control. (C) Binding to PA liposomes of MBP-Q2 and MBP-Q2C3m as in Fig. 3B, except PA concentrations were 100, 200, and 400 μM. Quantitation of bound protein indicated that the mutation reduced binding by 55%. (D) COS-7 cells expressing GFP-Q2C3m were microinjected with either buffer or bPLD, which induced translocation to a network characteristic of the ER. (E) Yeast expressing GFP-Q2, either wild-type (WT) cells or cells lacking SPO14. (F) Opi1C3m was compared to wild-type Opi1p in a growth assay for OPI1 function. A △opi1 strain (TLY501) was transformed with an empty plasmid (nil; pRS406), a plasmid expressing GFP-Opi1p (pTL211), or a plasmid expressing a mutant with reduced PA binding (GFP-Opi1C3m; pTL226). Transformants were spotted onto medium containing 100 μM or no inositol (10 cells per spot) and grown for 2 days at 37°C (as in fig. S4E). Cells containing the mutant form of Opi1p grew poorly in the absence of inositol (inositol auxotrophy), showing that the mutant Opi1p had increased activity. (G) Model of the negative feedback loop mediated by Opi1p. Step 1. In the absence of inositol, Opi1p is inactivated by interacting with Scs2p and PA on the ER (red). Step 2. Addition of inositol consumes PA in the ER (gray) and increases translocation of Opi1p into the nucleus, possibly by exposing its NLS (blue arrow), which partially overlaps the PA binding site. Step 3. Soluble nuclear Opi1p inhibits Ino2p/Ino4p (boxes), repressing genes such as INO1. It is unclear how binding to Scs2p on the inner nuclear membrane modifies the activity of Opi1p.

To characterize the role of the PA binding site in Q2 in vivo, we initially expressed GFP-Q2 in yeast and found it localized to both the periphery and the nucleus (Fig. 4E). The peripheral targeting was uniform, suggestive of the plasma membrane, not the ER. Rather than being inhibited by inositol, peripheral targeting of Q2 was enhanced (fig. S5D). We next investigated a possible mechanism localizing a PA binding protein to the plasma membrane, namely the action of yeast PLD1, Spo14p, which catalyzes the hydrolysis of PC to PA (Fig. 1A). Deletion of SPO14 did not affect localization or translocation of Opi1p (fig. S5E); however, peripheral targeting of Q2 was inhibited (Fig. 4E). Thus, whereas full-length Opi1p sensed only the PA pool on the ER because of the interaction of its FFAT motif with Scs2p (8), Q2 targeted a metabolically separate SPO14-dependent pool of PA on the plasma membrane. Although Spo14p is itself diffusely localized (23), this result implies that the major site of PA production by Spo14 is the plasma membrane and may explain the plasma membrane recruitment in yeast of the PA binding protein α-synuclein (24).

We next expressed various truncation and alanine mutants of Q2 in yeast (fig. S5F), mapping the minimal PA binding region to 56 residues (amino acids 113 to 168). In addition, we found that the nuclear localization signal (NLS) in Q2 was mainly located in the basic residues of group A and that plasma membrane targeting particularly involved group C (fig. S5F), correlating with the more prominent role of this group in PA binding (fig. S5C). To determine the effect of reduced PA binding on the function of Opi1p, we mutated group C in full-length Opi1p. In the absence of inositol, this mutation did not appear to significantly alter the distribution of GFP-Opi1p. However, in a growth assay for OPI1 function, Opi1C3m caused inositol auxotrophy (Fig. 4F), indicating significant over-activity. An eightfold mutant incorporating mutations of both group C and five hydrophobic residues that may be involved in membrane binding (19) showed even less PA binding in vivo (fig. S5F), was fully translocated to the nucleus in the absence of inositol (fig. S5G), and produced stronger inositol auxotrophy than Opi1C3m (fig. S5H). Thus, reduced binding of the basic domain to PA correlated with activation of Opi1p.

Conclusion. Here, we have shown that Opi1p is inhibited by binding PA on the ER and can rapidly translocate into the nucleus in response to consumption of PA induced by inositol (Fig. 4G). Notable examples of other transcription factors that regulate membrane lipid synthesis also have inactive forms on the ER (13), an important organelle for lipid sensors because this is where the majority of lipids are synthesized (25). Although other transcription factors that target the ER have transmembrane domains from which they are released by proteolysis (13), Opi1p targets the ER by a noncovalent interaction with a separate integral membrane protein subunit, Scs2p (8), and also appears to translocate after release from membranes (fig. S1). Modification of the interaction of Scs2p with Opi1p, for example by phosphorylation (26), may allow other pathways to regulate phospholipid production. The same interaction with Scs2p (or its homologs) is used by a variety of lipid binding proteins to gain access to the ER (8, 27), and we suggest that many cytoplasmic lipid binding proteins with regulatory functions will have targeting determinants to access membrane lipid pools (28). Lastly, although many proteins are known to bind PA (14, 1622), the physiological role of this ubiquitous intermediate in lipid metabolism has not been properly established. Here, we have demonstrated a physiological response to changes in PA, showing that PA is a signaling lipid and suggesting that specific pools of PA may play important roles in signaling in other cells.

Supporting Online Material

www.sciencemag.org/cgi/content/full/304/5677/1644/DC1

Materials and Methods

Figs. S1 to S5

References

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article