Multiple pools of phosphatidylinositol 4-phosphate detected using the pleckstrin homology domain of Osh2p.

Phosphatidylinositol (PtdIns) phosphate (PtdInsP) lipids are used as intracellular signposts for the recruitment and activation of peripheral membrane proteins. Whereas the distribution of most PtdInsPs is restricted to a single organelle, PtdIns(4)P is unique in that it exists in several discrete pools, and so proteins that bind PtdIns(4)P must use extra receptors to achieve a restricted localization. Here we compare the two highly related pleckstrin homology (PH) domains from Osh1p and Osh2p, yeast homologues of oxysterol-binding protein (OSBP), that target membranes using PtdIns(4)P, and in vitro bind both PtdIns(4)P and PtdIns(4,5)P2. We show that Golgi targeting is specified by an additional site on PH(Osh1), which lies on a face of the domain not previously known to interact with receptors. In contrast, PH(Osh2) does not have a demonstrable second site, and targets multiple pools of PtdInsPs, each dependent on a different PtdIns 4-kinase. This lack of a second site in PH(Osh2) allows it to be used as an unbiased reporter for altered distribution of 4-phosphorylated PtdIns. For example, in cells with excess PtdIns(4)P caused by inactivation of the phosphatase Sac1p, PH(Osh2) indicates that PtdIns(4)P accumulates on the plasma membrane, whereas other Golgi-targeted PH domains fail to detect this change.

Phosphatidylinositol (PtdIns) phosphate (PtdInsP) lipids are used as intracellular signposts for the recruitment and activation of peripheral membrane proteins. Whereas the distribution of most PtdInsPs is restricted to a single organelle, PtdIns(4)P is unique in that it exists in several discrete pools, and so proteins that bind PtdIns(4)P must use extra receptors to achieve a restricted localization. Here we compare the two highly related pleckstrin homology (PH) domains from Osh1p and Osh2p, yeast homologues of oxysterol-binding protein (OSBP), that target membranes using PtdIns(4)P, and in vitro bind both PtdIns(4)P and PtdIns(4,5)P 2 . We show that Golgi targeting is specified by an additional site on PH Osh1 , which lies on a face of the domain not previously known to interact with receptors. In contrast, PH Osh2 does not have a demonstrable second site, and targets multiple pools of PtdInsPs, each dependent on a different PtdIns 4-kinase. This lack of a second site in PH Osh2 allows it to be used as an unbiased reporter for altered distribution of 4-phosphorylated PtdIns. For example, in cells with excess PtdIns(4)P caused by inactivation of the phosphatase Sac1p, PH Osh2 indicates that PtdIns(4)P accumulates on the plasma membrane, whereas other Golgi-targeted PH domains fail to detect this change.
Membrane recruitment of peripheral membrane proteins is determined by one or more targeting domains that bind other proteins or membrane lipids. An important class of membrane lipids used to recruit proteins are the PtdInsPs, 1 phosphates being added at positions 3, 4, or 5 of the inositol sugar in any one of 7 combinations. Most PtdInsPs have quite restricted distributions, and can be considered as intracellular signposts or signals that recruit proteins to specific compartments (1). These distributions are created by the targeting of lipid modifying enzymes. For example, PtdIns(3)P is largely restricted to endosomes: it is synthesized by a PtdIns 3-kinase lying on the trans-Golgi network to the endosome pathway, and it is prevented from reaching other compartments by degradation in the vacuole/lysosome (2).
PtdIns(4)P differs from other PtdInsPs as it performs multi-ple roles. Not only is it a signpost for protein recruitment, but it is also an essential intermediate in the production of PtdIns(4,5)P 2 . Furthermore, PtdIns(4)P is synthesized by multiple, separate PtdIns 4-kinases and is therefore likely to have multiple intracellular pools. There are two families of PtdIns 4-kinases: types II and III (type I are now known to be PtdIns 3-kinases (3)). In mammalian cells, the predominant PtdIns 4-kinases are type II, but have only recently been cloned and are still partially characterized (4 -7). In addition, there are two mammalian type III kinases: ␣ and ␤ (8). Analysis of the relative importance of the different kinases in membrane traffic has been carried out in the model organism Saccharomyces cerevisiae, which also has type II and III kinases, but with a different emphasis, because here the type III kinases are dominant (9). The yeast type III ␣ PtdIns 4-kinase homologue, Stt4p, is an essential protein that has been localized to the plasma membrane (10), and has been shown to synthesize the bulk of PtdIns(4)P for cellular requirements. This pool acts as a precursor for PtdIns(4,5)P 2 (11), which is perhaps exclusively on the plasma membrane (12,13), and which is made by a single PtdIns(4)P 5-kinase, Mss4p, that is active on the plasma membrane (14,15). However, there are important roles for PtdIns(4)P beyond being an intermediate in PtdIns(4,5)P 2 production, in particular in budding from the late Golgi. This is the location of Pik1p, another essential PtdIns 4-kinase that is homologous to the human type III ␤ kinase (16,17). A number of peripheral Golgi proteins are now known to be recruited by PtdIns(4)P. Pleckstrin homology (PH) domains from three families of lipid-binding proteins target the Golgi (12, 13, 18 -21). In addition, clathrin adaptor AP1 complex was also shown to interact with PtdIns(4)P on the Golgi (22). Yeast also harbor a third PtdIns 4-kinase, Lsb6p, which is in the type II family, and has no known function (23). The question posed by the presence of these different PtdIns 4-kinases is how might proteins detect one specific pool of PtdIns(4)P among the multiple organelles that have active PtdIns 4-kinases? One plausible explanation supported by previous observations with the PH domain of OSBP is that proteins cannot target a specific compartment by virtue of PtdIns(4)P alone, but achieve restricted targeting by binding a membrane lipid in combination with other membrane receptors (12,24). OSBP is a peripheral membrane protein that targets the late Golgi solely by virtue of a PH domain that binds to PtdIns(4)P and PtdIns(4,5)P 2 with equal affinity (12,18). In addition, PH OSBP binds a Golgi-restricted receptor dependent on the small GTPase Arf1p (12), and has recently been shown to bind weakly to ARF1 (21).
Here, we focus on two PH domains from yeast homologues of OSBP: Osh1 and Osh2. Whereas PH Osh1 targets Golgi membranes, PH Osh2 , a closely related domain (71% identical and 91% homologous) targets membranes weakly (25). We have identified the site within PH Osh1 that binds a second mem-* This work was supported by Wellcome Trust Grant 060537 and Fight For Sight. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

EXPERIMENTAL PROCEDURES
Plasmids-PH domains were defined as follows (alternative gene names together with accession numbers are given in brackets, followed by starting and ending residues, identified using the three letter amino acid code): Osh1p (YAR042W/YAR044W, AY241177) Glu 280 -Arg 384 ; Osh2p (YDL019C, NP_010265) long form: Pro 256 -Pro 424 ; short form: Ser 286 -Lys 391 ; phospholipase C ␦1 (PLC␦1), ceramide transfer protein (CERT, previously called Goodpasture's antigen-binding protein) and phosphoinositol 4-phosphate adaptor protein 1 (FAPP1) as before (12). For chimeras between Osh1p and Osh2p PH domains, 4 restriction sites that do not alter the translated sequence were introduced at roughly equal spacing in both PH domains, to produce 5 segments that shared the same boundary residues in both domains. Plasmids containing PH domains were based on either pRS406 (integrating at URA3) or pRS416 (CEN URA3) and contained the constitutive portion of the PHO5 promoter, followed by GFP, a linker with the Myc epitope (GSMEQKLI-SEEDLRS), then the PH domain and a carboxyl-terminal extension of XNS (where X is encoded by the last two nucleotides from the source DNA and a guanine) as previously described (12,24,25). Plasmids are listed in Table I.
Yeast Strains-For localization in wild-type yeast, plasmids were routinely transformed into strain RS453B. Strains harboring the temperature-sensitive alleles pik1-83, stt4-4, and mss4-2 together with their wild-type controls and ⌬vps34 were as described previously (12). A ⌬sac1 strain (TLY218) was derived from SEY6210 by replacing SAC1 with the heterologous marker gene Schizosaccharomyces pombe HIS5 using the PCR method (26), and deletion of SAC1 was confirmed by PCR and inositol auxotrophy.
Liposome Centrifugation Assays-His 6 -tagged fluorescent PH domains were expressed in bacteria and purified as previously described (12). Small unilamellar liposomes were prepared, incubated with GFP-PH domain and removed from suspension by centrifugation as described previously (12). The GFP content remaining in the supernatant was measured in a LS50 spectrophotometer (excitation 485 nm, emission 515 nm, slit widths 5 nm). Phosphoinositides were assumed to be equally distributed between the two leaflets of the liposome bilayer.
Cell Imaging-GFP-PH domain fusion proteins were visualized in yeast cells growing at mid-log phase (A 600 nm ϭ 1) and in COS cells as previously (12,24). To label yeast with FM4-64, the dye was added at final concentration of 30 M 15 min prior to imaging endosomal/late Golgi compartments. Osh1 for Golgi Targeting-PH Osh1 targets Golgi membranes in yeast (Fig. 1A) and mammalian fibroblasts (24,25), whereas PH Osh2 shows barely discernible membrane targeting in yeast (Fig. 1B) and fibroblasts (data not shown). This differs from a recent study showing stronger Golgi targeting by monomeric PH Osh2 , although in that case a basic patch adjacent to the core PH domain sequence was also included, which may alter targeting (27). The portion of primary sequence responsible for the difference in targeting between PH Osh1 and PH Osh2 was mapped using a series of chimeras in which portions of the Osh1p sequence were exchanged for the homologous Osh2p sequence. For this purpose, the PH domain (105 amino acids in Osh1p) was divided into five approximately equal segments, and chimeras were made that contained mixtures of Osh1p and Osh2p sequence (see "Experimental Procedures"). Substitution of the NH 2 -terminal three-fifths of PH Osh1 with the equivalent region of PH Osh2 had no effect on targeting (Fig. 1C), despite the almost universal location of the binding site in PH domains for PtdInsPs within this NH 2 -terminal region (28,29). In contrast, inclusion of the next (i.e. fourth) segment of PH Osh2 inhibited membrane localization (Fig. 1D). The same requirement for the fourth segment of PH Osh1 was seen for Golgi targeting in mammalian COS cell fibroblasts (data not shown).

Identification of the Second Site in PH
Within the critical fourth segment of 22 residues, PH Osh1 differs from PH Osh2 at 8 positions. By further dissection of this segment, a single residue, histidine at position 79 in PH Osh1 compared with arginine at position 80 in PH Osh2 , was shown to be critical for targeting (Table II). To confirm the role of this single residue, arginine was introduced at position 79 of PH Osh1 , in which it strongly inhibited Golgi targeting (Fig. 1E). Introduction of histidine at the same site in PH Osh2 induced a striking gain of Golgi targeting (Fig. 1F).
Modeling of the sequence of PH Osh1 on known PH domain structures (28,29) shows that this residue lies in the middle of the ␤7 strand. In all known structures, this residue points away from the core of the domain on a side of the molecule at some distance from the membrane interface, assuming that this is defined by the site that binds PtdInsPs (Fig. 2). These results therefore indicate that the second site is unlikely to interact with a membrane lipid receptor, but is more likely to be a protein.
PH Osh2 Binds Both PtdIns(4)P and PtdIns(4,5)P 2 -If PH Osh2 is fully capable of membrane targeting, but just lacks a second site, it should bind PtdInsPs similarly to PH Osh1 . To test this, we compared the binding of GFP-tagged PH domains to liposomes containing 2.5% PtdIns(4)P or PtdIns(4,5)P 2 (12) (Fig.  3). As controls, PH PLC␦1 showed its well documented absolute preference for PtdIns(4,5)P 2 over PtdIns(4)P, and PH FAPP1 bound to both lipids, with a slight preference for PtdIns(4)P (Fig. 3). PH Osh1 and PH Osh2 bound PtdIns(4,5)P 2 with almost identical affinity, greater than that of PH FAPP1 (Fig. 3A); PH Osh1 bound PtdIns(4)P more strongly than PH Osh2 (ϳ2-fold more) (Fig. 3B). Overall, these experiments show that PH Osh2 binds PtdIns(4)P and PtdIns(4,5)P 2 with affinity similar to other PH domains that target membranes efficiently, despite its very weak membrane targeting in vivo. A PH Osh2 Dimer Targets Two Pools of 4-Phosphorylated Pt-dInsPs in Growing Cells-The above data show that PH Osh2 binds 4-phosphorylated PtdIns, but it differs from closely related PH domains by the absence of a second site. Therefore, we reasoned that if a reporter construct could be made to target membranes dependent purely on PH Osh2 , it would reveal the overall distribution of PtdIns(4)P and PtdIns(4,5)P 2 . To achieve membrane localization, we expressed a tandem dimer of PH Osh2 , as has been done for PH PLC␦1 and FYVE domains previously. The presence of two membrane binding sites vastly increases avidity without altering specificity (12,30). GFP-PH Osh2 -dimer showed tight dual localization to punctate internal structures and the plasma membrane, with little cytoplasmic background (Fig. 4A). GFP-PH Osh2 -dimer expressed from either a relatively weak promoter (OSH1), or a rapidly inducible promoter (GAL1) produced the same dual localization (data not shown), indicating that it is independent of duration and level of expression. Although the distribution of GFP-PH Osh2dimer became diffuse after fixation (data not shown), partial co-localization was found in live cells with punctae stained by FM4-64, which traverses the endocytic pathway and also enters late Golgi compartments prior to being sorted back to the cell surface (Fig. 4B) (31). Therefore, we conclude that the punctate structures are late Golgi and/or early endosomes. In addition, the dual targeting of PH Osh2 implies either that it interacts with two different ligands, or alternatively that a single ligand is present in two sites. An indication that both localizations of GFP-PH Osh2 -dimer are PtdInsP-dependent was gained using a dimer in which two conserved positive residues of both PH domain sequences contained mutations that are known to prevent the PtdInsP binding site of the PH domains (12), which had an entirely diffuse localization (Fig. 4C).
To test whether Pik1p plays a role in the localization of PH Osh2 , as it does for PH OSBP (12), GFP-PH Osh2 dimer was expressed in a strain carrying a conditional pik1 allele. Shift to the non-permissive temperature caused a rapid and almost complete loss of internal punctate localization in the mutant cells, with only a slight effect seen in wild-type cells undergoing the same temperature shift (Fig. 5A). By comparison, PH OSBP is only incompletely delocalized under these conditions (12). Therefore, PtdIns(4)P produced by Pik1p is likely to be the Golgi ligand for PH Osh2 . Next, we sought to identify the plasma membrane ligand for PH Osh2 . Initially we looked at the role of PtdIns(4,5)P 2 by expressing GFP-PH Osh2 dimer in cells carrying the mss4-2 ts allele. Little effect on the plasma membrane localization was detected at the non-permissive temperature under conditions where PtdIns(4,5)P 2 levels are known to fall (Fig. 5B) (12). This result does not rule out a role for PtdIns(4,5)P 2 at the plasma membrane, but indicates that there must be an additional ligand present when PtdIns(4,5)P 2 production is blocked.
We next tested whether the other essential PtdIns(4)-kinase, Stt4p, was responsible for synthesizing a pool of PtdInsP on the plasma membrane that is recognized by PH Osh2 . In temperature-sensitive strains carrying the mutant allele stt4-4 ts , the localization of the PH Osh2 dimer was altered, with much reduced plasma membrane localization even at the permissive temperature (data not shown). Plasma membrane targeting was almost completely lost at the non-permissive temperature, whereas punctate localization was preserved (Fig. 5C), and these effects were not seen in wild-type controls (Fig. 5C), indicating that Stt4p synthesizes the 4-phosphorylated PtdIns that PH Osh2 binds on the plasma membrane. We also examined the role of the single PtdIns 3-kinase in yeast, VPS34, because it is possible that PH Osh2 binds either PtdIns(3)P or PtdIns(3,5)P 2 . However, localization of the GFP-PH Osh2 dimer was unaffected in ⌬vps34 cells (data not shown). The combination of this result and the complete loss of punctate localization with the pik1 mutation indicates that the punctate localization of PH Osh2 is not mediated by PtdIns(3)P. Overall, these results show that each of the dual localizations of this PH domain is sensitive to inactivation of a different essential PtdIns 4-kinase, demonstrating that PH Osh2 interacts with two separate pools of 4-phosphorylated PtdIns.
PH Osh2 as a Reporter for Altered PtdIns(4)P Levels in ⌬sac1 Cells-The results above indicate that PH Osh2 is an unbiased reporter of PtdIns(4)P and PtdIns(4,5)P 2 , so we next examined how its distribution was affected by a well characterized mutation that alters PtdInsP levels. Sac1p is one of five PtdInsP phosphatases in yeast, and its inactivation leads to a large increase in cellular PtdIns(4)P, up to 10-fold, together with a fall in PtdIns(4,5)P 2 (16,32). To determine the effect of these FIG. 1. Identification of second site within PH Osh1 that determines Golgi targeting. PH Osh1 , PH Osh2 , and the indicated chimeras of these two homologous sequences were expressed as GFP fusion proteins as indicated in RS453, a wild-type yeast strain, and images taken of cells growing in log-phase. The different PH domains were: A, PH Osh1 (105 residues); B, PH Osh2 (106 residues); C, residues 1-61 of PH Osh2 followed by residues 61-105 of PH Osh1 ; D, PH Osh1 with residues 61-82 replaced with 62-83 of PH Osh2 . The region of 61-82 near the COOH terminus of PH Osh1 was critical for Golgi targeting. Fine dissection of this region then indicated that a single residue is critical (see Table II). E, PH Osh1 with mutation H79R; F, PH Osh2 with mutation R80H. Substitution of arginine into PH Osh1 inhibited targeting, whereas histidine 79 brought about Golgi targeting when inserted into PH Osh2 . Plasmids used in A-F were pTL331, pTL502, pTL503, pTL504, pTL508, and pTL509, respectively. changes in 4-phospholrylated PtdIns, we expressed tagged PH domains in a ⌬sac1 strain. The GFP-PH Osh2 dimer showed much increased plasma membrane targeting, and an apparent decrease in punctate localization (Fig. 6A, compare with Fig.  4A). This change might be caused by either a reduction in ligand concentration on Golgi/endosomes or an increase in ligand on the plasma membrane. To distinguish between these possibilities, we expressed monomeric GFP-PH Osh2 in the ⌬sac1 strain, and found that a single copy of PH Osh2 now localized appreciably to the plasma membrane (Fig. 6B, compare with Fig. 1B). This result indicates that the ⌬sac1 mutation leads to an increase in PtdIns(4)P at the plasma membrane, which is likely, from the results above, to be synthesized by Stt4p. Therefore, our findings agree with previous results that the excess PtdIns(4)P that accumulates in ⌬sac1 cells is synthesized by Stt4p rather than by Pik1p (32).
We next examined the effect of the ⌬sac1 mutation on other PH domains that bind PtdIns(4)P. GFP-PH Osh1 did not target the enlarged pool of PtdIns(4)P at the plasma membrane, but remained largely punctate (Fig. 6C). This is further evidence that PH Osh1 does not target membranes solely because they are rich in PtdIns(4)P. We next examined the effect of ⌬sac1 on the distribution of other Golgi-targeted PH domains, using sequences from OSBP, FAPP1, and CERT (previously called Goodpasture's antigen-binding protein). Like PH Osh1 , these target the Golgi by interacting with both PtdIns(4)P and a second site (12). The distribution of GFP-PH OSBP and GFP-PH FAPP1 remained punctate, as seen for GFP-PH Osh1 (Fig. 6, D  and E). However, GFP-PH CERT was notably different, being targeted to the nuclear envelope and to peripheral patches, characteristic of the yeast ER (Fig. 6F). To check whether this targeting was caused by binding to PtdInsPs, we next expressed a mutated version of this construct, in which the critical basic residue for PtdInsP binding was mutated to an arginine. GFP-PH CERT (R43E) was entirely diffuse, with no ER or punctate localization (Fig. 6G). On close inspection, all three of PH Osh1 , PH OSBP , and PH FAPP1 showed much weaker targeting to the nuclear envelope and periphery (Fig. 6). These results suggest that some PtdIns(4)P accumulates on the ER in ⌬sac1 cells, and that this pool is specifically targeted by PH CERT , possibly because of a unique second site interaction. DISCUSSION PH domains are sequences usually 100 -120 amino acids in length that have a characteristic fold of a seven-stranded ␤

TABLE II
Analysis of the section of PH Osh1 that determines Golgi targeting GFP-tagged PH Osh1 constructs that varied only in the sequences in the fourth out of five segments of PH Osh1 (residues 61-82) were scored for their punctate localization. Introduction of the Osh2p sequence (construct 1) made 8 changes and rendered membrane targeting only weakly detectable (w). While the NH 2 -terminal 3 variant residues had little effect on targeting (construct 2), the COOH-terminal 5 variant residues were responsible for the majority of targeting (construct 3). Introduction of the first 4 of these 5 changes had no effect on targeting (construct 4), indicating an important role for the last variant residue: His 79 in PH Osh1 (Arg 80 in PH Osh2 ).  2. Separation of the site that determines Golgi localization from the site that binds PtdIns. Ribbon diagram of the structure of the PH domain from the kinase Akt (PH Akt ) together with Ins(1,3,4,5)-tetrakisphosphate ligand, the soluble head group of PtdIns(4,5)P 2 (red and light blue) drawn by CHIMERA software using known coordinates (46). Features indicated are: loop ␤1-␤2 (1) and loop ␤3-␤4 (2) that encompass the binding site of the ligand, leading to the predicted orientation of the domain with respect to membrane lipids, which have been drawn to approximate scale (PH domain diameter ϭ approximately 4 nm, Golgi membrane thickness ϭ 7.3 nm (47), phospholipid diameter 0.7 nm); and Thr 77 of PH Akt (dark blue, arrow), the residue equivalent to His 79 of PH Osh1 and Arg 80 of PH Osh2 , positioned on a side of the domain at one side away from the known ligand binding site. Of 14 solved PH domain structures, the predicted secondary structure of the core ␤ sheets and ␣ helix in PH Osh1 is most similar to PH Akt . The extra helix (asterisk) in loop ␤3-␤4 of PH Akt is a unique feature absent from PH Osh1/2 that is unlikely to affect the core structure.
sandwich closed off at one side by an amphipathic ␣ helix (28,29,33). PH domains have been found to interact with a variety of different ligands, using sites that are somewhat conserved for each purpose. The ␣ helix can interact with WD40 proteins including G␤␥ (34), and differing parts of the ␣ helix of PH domains in guanine nucleotide exchange factors bind the neighboring catalytic Dbl homology domain (35). Finally, the best characterized interaction of PH domains arises from the formation of a pocket by loops ␤1-␤2 and ␤3-␤4 to bind PtdIn-sPs. Lipid specificity is determined by a combination of basic residues lining the pocket and a generally positive electrostatic charge on this face of the domain. Whereas a few PH domains, in particular those that bind PtdIns(3,4,5)P 3 , show high affinity binding for just a single PtdInsP, most PH domains, including PH OSBP and related sequences, bind PtdInsPs relatively promiscuously (28, 36 -38). This lack of specificity need not preclude specific overall targeting if the domain also interacts with a second receptor (12).
We compared two highly related PH domains with different Golgi targeting. Both PH domains bound PtdInsPs with moderate strength, similar to the recent findings of Lemmon and co-workers (38). Indeed, of the 33 PH domains in the entire yeast genome, these two are among only seven that demonstrably bind PtdInsPs at all (38). Given a similar binding affinity for PtdInsPs, what determines the difference in targeting? This was mapped to the third residue of strand ␤7, which in PH Osh1 (tight binding) consists of RWHLKG, compared with RWRLKG in PH Osh2 (weak binding). The histidine is conserved in Osh1p homologues in humans (OSBP, ORP4, and ORP9), flies, worms, plants, and in other fungi (Candida albicans, Aspergillus nidulans but not Schizosaccharomyces pombe-serine, or Magnaporthe grisea and Neurospora crassa-threonine), whereas the arginine is unique to Osh2p in Saccharomyces. Different PH domain structures all indicate that this residue lies at the bottom of a shallow groove on a face of the domain adjacent to but not overlapping with the PtdInsP binding site (Fig. 2) (28,29,37). The finding that this histidine can specify Golgi localization identifies this side of PH domains as a site in this subfamily of PH domains that binds a second ligand. Given the location of the site away from the membrane, it would appear that the ligand is a protein, rather than a membrane lipid. Following our initial studies where we found that this second site depends on Arf1p in yeast (12), De Matteis and co-workers (21) have now demonstrated a direct interaction between both PH FAPP1 and PH OSBP with ARF1 in mammalian cells (the equivalent of yeast Arf1p). On this basis, one would therefore FIG. 4. Dual localization of PH Osh2 dimer. A, PH Osh2 was expressed as a tandemly repeated dimer of PH Osh2 extended ϳ30 amino acids at each end (total of 169 amino acids) bracketed by GFP (using pTL511). The dimer showed a dual localization: at the plasma membrane particularly enriched in small buds; and in numerous punctate structures often seen at the bud-neck in large budded cells, which is the site of secretion (arrows). The same dual localization was seen with another dimeric PH Osh2 construct with only a single GFP, and much narrower definition of PH Osh2 (106 residues, pTL512, data not shown). B, cells expressing GFP-PH Osh2 -dimer as in A, co-stained with the endocytic tracer FM4-64 (15 min uptake), which has entered late Golgi compartments. In the bottom panel, each image has been falsely colored: GFP in green, FM4-64 in magenta, with co-localization indicated in white. The very brightest punctae all showed good co-localization, as do a proportion of the less bright punctae (arrowhead). Using these settings, cells containing single markers showed complete separation of the fluorophores (data not shown). C, GFP-PH Osh2 dimer as in A, but both PH domains of the dimer carry mutations in two basic residues implicated in PtdInsP binding (K307E/R309E, pTL513). The construct was entirely cytoplasmic, indicating loss of both of the localizations of wild-type PH Osh2 .

FIG. 5. Different PtdIns 4-kinases synthesize the two pools of PtdIns(4)P.
All images are of the GFP-PH Osh2 dimer expressed (from pTL511) in the indicated mutant strains and the corresponding wildtype parental strains, grown to mid-log phase at the permissive temperature (25°C) and shifted to a strongly non-permissive temperature (39°C) for 15 min prior to imaging. A, inactivation of Pik1p completely inhibited the punctate localization of PH Osh2 . B, targeting to both sites was unaffected by inactivation of Mss4p, indicating no requirement for PtdIns(4,5)P 2 for localization. C, plasma membrane localization was lost selectively on inactivation of Stt4p, but not in the wild-type control. Plasma membrane targeting was also reduced in the stt4-4 mutant strain at a temperature permissive for growth (25°C, data not shown).
Although Golgi targeting can be achieved by combining two binding interactions, without a second site monomeric PH Osh2 does not target the Golgi. This reduced affinity for the Golgi may be crucial to targeting of full-length Osh2p, which has diverged in multiple ways from Osh1p (25). Because PH Osh2 interacts only with PtdInsPs, we have been able to develop a tool based on dimeric PH Osh2 that detects PtdIns(4)P and PtdIns(4,5)P 2 alone (Fig. 3). Binding to other PtdInsPs was shown to be irrelevant by the lack of a role for Vps34p. The dual localization of the GFP-PH Osh2 dimer in log-phase cells was consistent with previous studies of the two essential PtdIns 4-kinases (9), and their localizations (10,17). Whereas previous studies have shown that Pik1p produces PtdIns(4)P at the Golgi (12,13), this is the first demonstration that Stt4p synthesizes PtdIns(4)P on the plasma membrane. Because the Stt4-dependent pools of PtdIns(4)P and PtdIns(4,5)P 2 are of roughly equivalent size (11), it is difficult to determine which lipid is being detected by the GFP-PH Osh2 dimer, even using cells in which Mss4p can be inactivated.
If the GFP-PH Osh2 dimer is indeed a probe for 4-phosphorylated PtdIns, its distribution should change in response to changes in PtdInsP metabolism. We used cells in which the SAC1 gene was disrupted, because in the absence of this PtdInsP phosphatase, cells accumulate up to 10 times the normal levels of PtdIns(4)P, whereas PtdIns(4,5)P 2 decreases by 50% (16,32). Results with PH Osh2 in these cells showed that excess PtdIns(4)P accumulated mainly on the plasma membrane. However, other PH domains that bind PtdIns(4)P, such as those from Osh1p, OSBP, and FAPP1, remained largely localized to the Golgi in ⌬sac1 cells, similar to the recent findings of Lemmon and coworkers (38). This is further evidence that PH domains that target the Golgi as monomers localize using a second receptor in addition to PtdInsP. Interestingly, PH CERT behaved differently in cells carrying ⌬sac1 compared with the other sequences tested in targeting the ER in a PtdInsP-dependent manner. This suggests that PH CERT recognizes a second Golgi receptor that is relocated to the ER in a ⌬sac1 strain. Because this receptor differs from that recognized by PH Osh1p , PH OSBP , and PH FAPP1 , it may be that PH CERT does not bind ARF1. ER relocalization of a Golgi protein in ⌬sac1 cells has also been reported for the short yeast OSBP homologue Kes1p, which also binds PtdIns(4)P and PtdIns(4,5)P 2 (39). CERT was recently shown to transfer ceramide in mammalian cells from its site of synthesis (the ER) to the site of its conversion to complex sphingolipid (late Golgi) (19), a process that also occurs in yeast (40). Identical targeting by PH CERT and Kes1p suggests the possibility that they both access a subdomain of the late Golgi that is specialized for non-vesicular ceramide import.
It is clear from all studies of PH domains targeted to Golgi membranes that they bind PtdIns(4)P. However, these PH domains can also show fairly indiscriminate binding to a broad range of PtdInsPs (12,24,36,38), i.e. there is no intrinsic specificity for PtdIns(4)P. This is sometimes overlooked (41) because different techniques for measuring protein-lipid interaction yield differing results. Binding protein to lipids immobilized on nitrocellulose (so-called "Fat Western") originally showed that PH FAPP1 specifically binds PtdIns(4)P (20), however, this specificity was not seen with our liposome-binding technique. Although it is not yet known if other proteins that target PtdIns(4)P on the Golgi are equally promiscuous (22), it is tempting to speculate why such lack of specificity exists for PtdIns(4)P-binding proteins. PtdIns(4)P exists in multiple discrete pools, not only in yeast, but also in mammalian cells where multiple organelles can recruit PtdIns 4-kinases (4 -8, 22, 42). Thus, proteins that target PtdIns(4)P can only achieve a narrow distribution among the multiple pools of PtdIns(4)P by combining this interaction with others. Therefore, there may be very little evolutionary disadvantage to binding the combination of PtdIns(4)P and PtdIns(4,5)P 2 , and it is possible that a truly PtdIns(4)P-specific binding protein may not exist.
In summary, PH Osh2 appears to detect two independent pools of 4-phosphorylated PtdIns without the bias introduced by a second site that is otherwise found in PH domains that bind PtdIns(4)P and PtdIns(4,5)P 2 . The potential use of the probe is complicated by its interaction with two lipids. Now that the microheterogeneity of all intracellular organelles has become evident (43)(44)(45), it will become increasingly important to study how multiple targeting signals integrate to specify the overall distribution of a protein.
Acknowledgments-We acknowledge the support of Sean Munro, in whose laboratory this work was initiated, and who has given much useful advice. We also thank Scott Emr for providing yeast strains and communicating results prior to publication, and Anjon Audhya, and G, GFP-PH CERT * carrying the mutation R43E. The constructs used are indicated diagrammatically above each image (2-2, 2, 1, O, F, C, C*), and were expressed from pTL511, pTL342, pTL331, pTL332, pTL334, pTL333, and pTL514, respectively. Deletion of ⌬sac1 shifted dimeric PH Osh2 from punctate to plasma membrane targeting. Monomeric PH Osh2 showed enhanced targeting to the plasma membrane (compare with Fig. 1A). In comparison, PH Osh1 , PH OSBP , and PH FAPP1 targeted punctae predominantly in ⌬sac1 cells. In addition, these sequences showed minor targeting to the nuclear envelope (arrows) and cell periphery (arrowheads). PH CERT was unique in targeting the ER (nuclear envelope and peripheral patches). This was dependent on binding to PtdInsPs, as it was abolished by mutating the critical basic residue in PH CERT (Arg 43 ) to an acid (Glu).