A novel probe for phosphatidylinositol 4-phosphate reveals multiple pools beyond the Golgi

Eukaryotic organelles must maintain distinct protein and lipid compositions to preserve their unique functional identity, despite a process of vesicular transport that constantly exchanges the membrane between them (Roth, 2004). The polyphosphoinositides (PPIns) have emerged as a family of lipid molecules intimately associated with this process. PPIns are derivatives of the major structural phospholipid phosphatidylinositol (PtdIns) found in cytosolic membrane leaflets (Roth, 2004). The inositol head group can be phosphorylated at three distinct hydroxyls, generating three mono-, three bis-, and a single trisphosphorylated isomer. These PPIn isomers interact with hundreds of cellular proteins, regulating diverse processes at the membrane/cytosol interface, ranging from signal transduction to ion flux (Balla, 2013). The readily interchangeable phosphorylation status of the PtdIns head group by PPIn kinases and phosphatases enables restricted enrichment of these lipids on particular membrane compartments, even as bulk lipids move through vesicular transport (Roth, 2004).

PtdIns 4-phosphate (PtdIns4P) was originally thought be a simple intermediate of the functionally prolific plasma membrane (PM) PPIn, PtdIns 4,5-bisphosphate (PtdIns(4,5)P₂; Balla, 2013); more recently, it has been found to be a key functional component of Golgi membranes (Graham and Burd, 2011). Early work in the yeast Saccharomyces cerevisiae identified mutants with compromised PtdIns4P synthesis as having defects in the early secretory pathway at the level of the Golgi apparatus (Hama et al., 1999; Walch-Solimena and Novick, 1999). Since then, many PtdIns4P effectors have been discovered, all of which appear to function on Golgi membranes, either in vesicle budding (Wang et al., 2003; Godi et al., 2004), nonvesicular lipid transport (Tóth et al., 2006; D’Angelo et al., 2007), or maintenance of Golgi structure (Dippold et al., 2009). Thus, PtdIns4P coordinates a diverse array of protein functions whose common feature is localization at the Golgi apparatus.

Although these data point to a rather restricted function at the Golgi, recent evidence has unearthed a more elaborate and less clearly defined picture for PtdIns4P function. First, several recent studies have found that the largest pool of the lipid—more than half of it at least—resides in the PM (Hammond et al., 2009; Sarkes and Rameh, 2010; Nakatsu et al., 2012). Second, PtdIns4P synthesis seems to span several membrane compartments. In mammals, PtdIns 4-kinase (PI4K) activity is encoded by four genes: PI4KA and PI4KB (Sttp4 and Pik1, respectively, in S. cerevisiae; often referred to as type III enzymes after an antiquated classification of PtdIns kinase activities) as well as the evolutionarily distinct homologues PI4K2A and PI4K2B (with a single Lsb6 orthologue in S. cerevisiae; Brown and Auger, 2011; Balla, 2013). Of these, PI4KB and PI4K2A have been implicated in Golgi PtdIns4P synthesis (Godi et al., 1999; Hama et al., 1999; Walch-Solimena and Novick, 1999; Wang et al., 2003; Polevoy et al., 2009; Minogue et al., 2010). However, PI4K2A has also been shown to function in endosomal/lysosomal traffic (Salazar et al., 2005; Minogue et al., 2006; Burgess et al., 2012; Jović et al., 2012), as has its homologue PI4K2B (Balla et al., 2002). Furthermore, a pool of PtdIns4P is produced on autophagosomes to support PtdIns(4,5)P₂ synthesis required for autophagic lysosome reformation (Rong et al., 2012), and some PtdIns4P has been reported on a subset of lysosomal membranes under normal conditions, which is required for lysosomal homeostasis (Sridhar et al., 2013). Finally, PI4KA has been shown to be responsible for PtdIns4P synthesis at the PM (Balla et al., 2008b; Nakatsu et al., 2012; Tan et al., 2014).

Can disparate synthesis of PtdIns4P on multiple membrane compartments be reconciled with its recruitment of effector proteins exclusively to Golgi membranes? Answering this question would require detection of the free lipid molecules available to recruit effector proteins in living cells. This has been attempted using fluorescent protein fusions of PtdIns4P-binding pleckstrin homology (PH) domains from Golgi effector proteins, which invariably localize to the Golgi in a PtdIns4P-dependent manner (Balla et al., 2005), apparently supporting a Golgi-selective enrichment of the lipid. A major caveat of this approach is that PH domains often exhibit tertiary intermolecular interactions that bias the probes’ localization (Balla et al., 2000; Levine and Munro, 2002), i.e., PtdIns4P is not sufficient for membrane targeting. The PtdIns4P-binding Osh2p PH domain binds to the PM and Golgi in yeast (Roy and Levine, 2004) but localizes solely to the PM in mammalian cells, which could be interpreted as reflecting a relatively low concentration of available PtdIns4P at the Golgi. However, this probe also binds PM PtdIns(4,5)P₂ (Roy and Levine, 2004; Hammond et al., 2012). Immunocytochemical methods have detected PtdIns4P in both Golgi and PM (Hammond et al., 2009), but this technique is limited to fixed specimens and cannot answer the crucial question as to which pools of the lipid are available to recruit effector proteins.

Many bacterial pathogens have evolved secreted effector proteins with features that allow selective targeting of host cell membranes, which are then hijacked to support replication of the bacteria (Ham et al., 2011). The ability to bind or modify the PPIn head group is a common mechanism used by such proteins (Ham et al., 2011). We therefore sought to use the power of bacterial effector proteins as a tool to investigate PtdIns4P distribution in living cells. We focused on the SidM protein, secreted by the intracellular pathogen Legionella pneumophila. SidM is localized to L. pneumophila–containing vacuoles where its guanine nucleotide exchange factor and AMPylation activities stimulate Rab1 signaling, causing the recruitment of ER-derived materials (Machner and Isberg, 2007; Müller et al., 2010). L. pneumophila–containing vacuole recruitment is facilitated through binding to PtdIns4P by a unique PtdIns4P binding of SidM (P4M) domain (Brombacher et al., 2009). The P4M domain exhibits exquisite specificity and affinity for PtdIns4P in vitro (Brombacher et al., 2009; Schoebel et al., 2010) and is structurally distinct from the catalytic and Rab1-interacting activities (Schoebel et al., 2010; Zhu et al., 2010). The P4M domain from SidM is therefore an ideal candidate for an unbiased and sensitive biosensor for PtdIns4P in living cells.

Here, we show that PtdIns4P is necessary and sufficient for membrane targeting of P4M in living cells and that the probe detects readily accessible pools of the lipid in the Golgi, PM, and Rab7-positive endosomes/lysosomes. Each of the mammalian PI4Ks corresponds to at least one of these pools, and the PM and Golgi pools are accessible to the ER/Golgi-localized phosphatase Sac1.

We expressed a GFP-conjugated P4M domain consisting of residues 546–647 of L. pneumophila SidM (Brombacher et al., 2009) in COS-7 cells and viewed the distribution by spinning-disk confocal microscopy. The probe was enriched on several membrane compartments (Fig. 1 A). Most prominent was a compact perinuclear compartment consistent with the Golgi apparatus. There was also a clear accumulation of peripheral labeling encompassing the entire cell, likely corresponding to the PM. Less obvious as a result of their smaller size and poorer contrast against the cytosolic background were smaller vesicular compartments enriched in P4M. These became more apparent when the entire volume of the cell was viewed as a maximum intensity projection (Fig. 1 A, middle). Time-lapse imaging revealed that these vesicular compartments were highly motile relative to the largely stationary Golgi and PM (Fig. 1 A, right; and Video 1).

A Golgi distribution of a PtdIns4P-binding biosensor is to be expected given the known localization of previous PH domain–based probes (Balla et al., 2005); indeed, one such probe—namely the PH domain from FAPP1 fused to GFP—colocalized with an infra-RFP (iRFP)–tagged P4M construct exclusively at the Golgi (Fig. 1 B). iRFP-P4M also colocalized with the mCherry-tagged PtdIns(4,5)P₂-binding PH domain from PLCδ1 (Fig. 1 B), which localizes only to the PM (Várnai and Balla, 1998). As expected, no colocalization was seen between the two disparately localized PH domain probes (Fig. 1 B). To improve membrane avidity of our novel probe, we made a tandem fusion of two P4M domains with GFP. Despite improved binding to the PM, Golgi membranes appeared highly distorted and compacted by the tandem construct (Fig. 1 C), so we did not characterize this probe further.

How faithfully does the localization of P4M reflect cellular accumulation of PtdIns4P? In vitro, P4M exhibits a remarkably specific, monogamous, and high-affinity binding to PtdIns4P (Brombacher et al., 2009; Schoebel et al., 2010). Nevertheless, the cell is a far more complex and diverse molecular environment than can be recapitulated in vitro, raising the possibility of additional interactions that could bias the probe’s localization. We therefore set out to test whether PtdIns4P binding was necessary and sufficient to account for the localization of P4M in cells. We found that aldehyde fixation removed the membrane enrichment of P4M, precluding analysis of the probe’s localization with endogenous markers detected using immunocytochemistry. Difficulties in preserving the localization of the probe during such procedures were not entirely unexpected because we previously reported the need for varied fixation and permeabilization conditions to preserve different PtdIns4P-containing membranes for immunofluorescence localization (Hammond et al., 2009). Furthermore, it has been reported that low-affinity interactions are not preserved during aldehyde fixation (Schmiedeberg et al., 2009), which may be further confounded in the case of P4M by the reaction of aldehydes with the primary amines in the arginine and lysine side chains required for PtdIns4P binding (Schoebel et al., 2010). We therefore focused our efforts on techniques applicable to live cells, as described in the remaining Results sections.

To test whether binding to PtdIns4P was necessary for P4M targeting in cells, we used chemically induced recruitment of lipid-modifying enzymes to specific cellular compartments (Fili et al., 2006; Suh et al., 2006; Várnai et al., 2006). This system is based upon dimerization of the FK506 binding protein (FKBP) of 12 kD and the FKBP12-rapamycin–binding (FRB) domain from mammalian target of rapamycin induced by the cell-permeable immunosuppressant rapamycin (Brown et al., 1994). In our configuration, FRB is fused to a specific organelle resident motif, which recruits FKBP-conjugated PPIn kinases or phosphatases upon addition of rapamycin. This leads to rapid metabolism of the target PPIn specifically in the recruiting compartment (e.g., Fig. 2 A). This process can be followed by fluorescence microscopy in living cells by fusion of both constructs to fluorescent proteins. To deplete PtdIns4P, we used an mRFP- and FKBP-fused synthetic enzyme chimera Pseudojanin (PJ)-Sac (Hammond et al., 2012), which carries an active Sac domain from S. cerevisiae Sac1 that dephosphorylates PtdIns4P to PtdIns (Guo et al., 1999).

We began by using PJ-Sac to deplete PtdIns4P from the Golgi to test the specificity of P4M’s interaction with this organelle. In a previous study, we used FRB targeted to trans-Golgi membranes by TGN38 (Szentpetery et al., 2010). However, in recent experiments, we have found that recruitment of PPIn-modifying enzymes with TGN38 can lead to effects on lipids at the PM, consistent with the cycling of TGN38-containing membranes between the trans-Golgi network and the PM (Wakana et al., 2012). Instead, we used a construct using the C-terminal transmembrane domain of the cisternal Golgi-resident protein giantin, encoded by GOLGB1 (Komatsu et al., 2010). This construct produced a tight localization to the Golgi (Fig. 2, A and B) and efficiently recruited PJ-Sac upon addition of rapamycin (Fig. 2 A and Video 2). This was followed by release of GFP-P4M from the Golgi region (Fig. 2 A and Video 2). No significant change in P4M localization was observed after recruitment of PJ-Dead (Hammond et al., 2012), a chimera with an inactivating mutation in the Sac domain, indicating that release of P4M from the Golgi required catalytic activity of Sac1 (Fig. 2 A). We conclude that P4M’s localization to the Golgi is PtdIns4P dependent.

The finding that P4M recruitment to the Golgi is PtdIns4P specific is consistent with this lipid being the predominant PPIn at the Golgi. Conversion of PtdIns4P to PtdIns(4,5)P₂ has also been proposed at the Golgi (Godi et al., 1999), although the lipid is not detected there by the PH domain from PLCδ1 (Fig. 2 B; Várnai and Balla, 1998). Why not? Either Golgi PtdIns(4,5)P₂ levels are too low to be detected with PH-PLCδ1 or else tertiary interactions bias this probe’s detection of PtdIns(4,5)P₂ exclusively to the PM (Balla et al., 2000). To discriminate between these possibilities, we recruited an FKBP-conjugated PtdIns4P 5-kinase (PIP5K), which converts PtdIns4P to PtdIns(4,5)P₂, to the Golgi; this manipulation caused significant accumulation of PH-PLCδ1 at the Golgi (Fig. 2 B and Video 3) but not when an inactivating mutation was introduced into PIP5K (Fig. 2 B). Therefore, PH-PLCδ1 does have the ability to detect PtdIns(4,5)P₂ in the Golgi, and we conclude that its failure to do so under normal conditions results from levels of the lipid being below the detection limit of this probe. This is consistent with PtdIns4P being the predominant PPIn at the Golgi.

As discussed in the introduction, several PH domain probes for PtdIns4P appear biased toward Golgi localization. For example, the PH domain from FAPP1 binds to both PtdIns4P and the small GTPase Arf1 (Levine and Munro, 2002; Godi et al., 2004); as a result, inhibition of GDP exchange on Arf1 by Brefeldin A inhibits Golgi binding of this domain (Balla et al., 2005). Indeed, we found that Brefeldin A decreased (but did not ablate) Golgi localization of PH-FAPP1 (Fig. 3 and Video 4). Conversely, in the same cell, the binding of P4M to the Golgi was slightly increased (Fig. 3 and Video 4), perhaps because of a small increase in free PtdIns4P levels after release of PH-FAPP1. Surprisingly, Brefeldin A treatment caused little dispersion of the P4M domain or the remaining PH-FAPP1 domain, despite causing dispersal of a cis/medial-Golgi marker derived from Mannosidase II (Fig. 3, C and D), indicating that this was not an artifact of P4M overexpression. The remaining P4M and FAPP1-PH remaining in the Golgi region likely represent trans-Golgi membranes, which are also enriched in PtdIns4P (Godi et al., 2004; Weixel et al., 2005) and that do not disperse with Brefeldin A treatment (Lippincott-Schwartz et al., 1989; Reaves and Banting, 1992). Regardless, this result demonstrates that, unlike several PH domain PtdIns4P probes, P4M is unbiased by interactions with Arf1.

We next turned our attention to the specificity of P4M binding to the PM. Recruitment of PJ-Sac to the PM via a palmitoylated/myristoylated peptide from Lyn kinase caused the rapid release of P4M from the PM, without effect on the localization of the PtdIns(4,5)P₂ probe PH-PLCδ1 (Fig. 4 A and Video 5). This demonstrates that, unlike the previous PM PtdIns4P probe PH-Osh2p, P4M binds the PM in a PtdIns4P-dependent, but PtdIns(4,5)P₂-independent, manner. Despite the role of PtdIns4P in PtdIns(4,5)P₂ synthesis, reduction of PtdIns4P levels has no appreciable effect on the PtdIns(4,5)P₂ probe, which is entirely consistent with our previous study exploring this very phenomenon (Hammond et al., 2012). Conversely, recruitment of an active INPP5 domain from INPP5E, which converts PtdIns(4,5)P₂ into PtdIns4P, would be expected to increase the levels of PtdIns4P in the PM. Indeed, depletion of PtdIns(4,5)P₂ caused the release of PH-PLCδ1, and the accompanying increase in PtdIns4P was reflected by augmented PM recruitment of P4M (Fig. 4 B and Video 6).

Depletion of both lipids from the PM with PJ, a chimera carrying both active Sac and INPP5E domains (Hammond et al., 2012) caused release of both P4M and PH-PLCδ1 from the PM (Fig. 4 C), whereas inactivating mutations in both phosphatases (PJ-Dead) produced no changes in the localization of either probe (Fig. 4 D). These observations lead us to the conclusion that P4M’s interaction with PtdIns4P is both necessary and sufficient for its binding to the PM.

Previous studies have noted that the PH domain from oxysterol binding protein (OSBP), although predominately localized to the Golgi, can recognize PtdIns4P at the PM under conditions in which its concentration rises (Balla et al., 2005; Halaszovich et al., 2009), similar to our observations for P4M (Fig. 4 B). We therefore sought to compare the two domains. Given that PH-OSBP’s PM localization is barely detectable in confocal microscopy (Balla et al., 2005), we turned to the quantitatively more sensitive total internal reflection fluorescence (TIRF) microscopy. When observed in TIRF, depletion of PM PtdIns4P with PJ-Sac caused similar significant drops in the intensity of both P4M and PH-OSBP tagged with GFP (Fig. 4 E). Conversion of PM PtdIns(4,5)P₂ to PtdIns4P caused an ∼10% increase in recruitment of PH-OSBP (although this was not statistically significant), whereas P4M was elevated significantly more, by ∼30% (Fig. 4 F). Taking these observations together, P4M thus represents a superior probe for PM PtdIns4P that the previously available PH domain probes from OSBP and Osh2p.

Having established PtdIns4P-dependent binding to PM and Golgi membranes, it was important to identify the mysterious vesicular structures labeled by P4M, especially because these are not seen with other probes for PtdIns4P. Given PtdIns4P’s role in vesicular transport from the Golgi (Walch-Solimena and Novick, 1999; Wang et al., 2003; Godi et al., 2004), we initially suspected that these may represent secretory carriers. However, loading cells with LysoTracker green, which accumulates in acidic late endosomal and lysosomal compartments, revealed a surprisingly close association of P4M with motile, peripheral structures as well as more stationary objects closely apposed to the Golgi (Fig. 5 A and Video 7). Consistently, P4M-positive compartments also showed a tight overlap with Rab7 (Fig. 5 B), a marker of late endosomes and endosomes/lysosomes (Méresse et al., 1995). P4M showed some overlap with the early endosomal marker Rab5 (Bucci et al., 1992), although this was significantly less pronounced than with Rab7 (Fig. 5 B). In contrast, the EEA1 FYVE domain, a probe for PtdIns 3-phosphate (PtdIns3P) known to be tightly associated with Rab5-positive early endosomes (Gillooly et al., 2000; Fili et al., 2006), showed a tight colocalization with Rab5 that was much stronger than with Rab7 (Fig. 5 C). Despite some overlap of both probes with Rab5, we did not observe colocalization of the EEA1 FYVE domain with P4M, which labeled distinct vesicular structures (Fig. 6 A).

Does this endosomal/lysosomal P4M localization reveal a novel pool of PtdIns4P in this compartment? Careful dissection of the specificity of P4M binding was required to produce a satisfactory answer to this question because other PPIns, namely PtdIns3P and PtdIns 3,5-bisphosphate (PtdIns(3,5)P₂), have been reported on early endosomal and late endosomal/lysosomal compartments (Roth, 2004). Cotransfection of cells with the PtdIns3P reporter GFP-FYVE-EEA1 (Balla et al., 2000) revealed an exclusive localization compared with mCherry-P4M (Fig. 6 A and Video 8). Furthermore, addition of the PPIn 3-kinase inhibitor Wortmannin caused complete elimination of punctate FYVE domain labeling, with no effect on P4M distribution (Fig. 6 A and Video 8), as predicted by the lack of in vitro PtdIns3P binding by P4M (Brombacher et al., 2009; Schoebel et al., 2010). This observation also argues against P4M binding to PtdIns(3,5)P₂ because synthesis of this lipid, like PtdIns3P, depends on PPIn 3-kinase activity.

To address the PtdIns4P dependence of P4M recruitment to the endosomal/lysosomal compartment, we generated FRB-conjugated Rab7, which efficiently recruited PJ-Sac (Video 9) and caused the dissociation of P4M from the Rab7-positive membranes (Fig. 6 B and Video 9); in fact, almost all P4M-labeled vesicular structures were removed by this treatment, implying that Rab7-positive membranes are the major host compartment.

The S. cerevisiae Sac domain that we used in PJ-Sac has significant catalytic activity against PtdIns3P and PtdIns(3,5)P₂, both in vitro and in vivo (Guo et al., 1999). Therefore, as a control, we generated an FKBP-fused MTM1 (myotubularin 1) enzyme (Fili et al., 2006), which specifically hydrolyzes the 3-phosphate from PtdIns3P and PtdIns(3,5)P₂ but not other PPIns (Schaletzky et al., 2003). This enzyme did not produce an alteration in P4M localization at the Rab7-positive compartments (Fig. 6 C), firmly establishing the specific PtdIns4P dependence of P4M localization there.

As a further control for PJ-Sac specificity, we recruited both enzymes to PtdIns3P-rich early endosomes using an FRB fusion to Rab5 (Fig. 6 C). Whereas MTM1 caused a complete elimination of the PtdIns3P reporter GFP-FYVE-EEA1 from Rab5-positive membranes (Fig. 6 C), PJ-Sac had no significant effect (Fig. 6 C). Overall, these observations suggest that the catalytic preference of Sac for PtdIns4P is much greater than for PtdIns3P, at least in the configuration of PJ-Sac.

As a final test for the presence of PtdIns4P on Rab7-positive endosomal/lysosomal membranes, we reasoned that recruitment of the FKBP-PIP5K should generate PtdIns(4,5)P₂ there. Indeed, we saw recruitment of PH-PLCδ1 to Rab7-positive membranes after recruitment of the catalytically active, but not the inactive, PIP5K (Fig. 6 D and Video 10). Collectively, these observations firmly establish that PtdIns4P is generated in substantial quantities on late endosomal/lysosomal membranes.

The aforementioned experiments indicate that PtdIns4P is required to recruit P4M onto the membrane compartments labeled by it in live cells. However, for this to represent a definitive map of the cellular PtdIns4P distribution, we would need to confirm that PtdIns4P is sufficient for P4M recruitment. To test this, we sought a method to induce ectopic synthesis of PtdIns4P in cells. Infection by the plus-stranded RNA virus hepatitis C leads to replication of the virus on an ER-derived membranous web, to which the host cell’s PI4KA enzyme is recruited, producing ectopic PtdIns4P (Berger et al., 2011; Reiss et al., 2011). The interaction depends on viral NS5A (nonstructural protein 5A), although the protein is not sufficient to recruit and activate endogenous PI4KA (Reiss et al., 2013). Consistent with this observation, we found that overexpression of an mCherry-tagged NS5A did not produce alterations in P4M distribution. However, coexpression of an active PI4KA enzyme with NS5A caused substantial accumulation of P4M on peripheral weblike membrane structures that colocalized with NS5A (Fig. 7). No such ectopic P4M accumulation was observed when the catalytically inactive D1957A mutant of PI4KA was expressed with NS5A (Fig. 7). We conclude that PtdIns4P is sufficient to drive the cellular distribution of P4M, which therefore likely reports all the substantial pools of accessible PtdIns4P within cells.

If these multiple organelle pools of PtdIns4P represent the cell’s entire complement of this PPIn, does this correspond to the known localization of PtdIns4P metabolism? We sought to explore this through expression of the major PtdIns4P metabolic enzymes together with P4M. We first consider the PtdIns4P-synthesizing PI4Ks (Fig. 8 A): PI4KA has recently been shown to be recruited to the PM transiently and to rely on a multiprotein complex between the PM-anchored EFR3 proteins and the scaffold TTC7 (Baird et al., 2008; Nakatsu et al., 2012). Expression of these components in COS-7 cells together with P4M revealed a tight PM association of the palmitoylated EFR3B (as well as some internal membrane association), whereas TTC7B and PI4KA were mainly cytosolic (Fig. 8 A). Nonetheless, a small enrichment of both proteins was observed with EFR3, and PI4KA was the only enzyme to show any colocalization with P4M at the PM (Fig. 8 A). This is consistent with the dynamic recruitment of the kinase to the PM (Nakatsu et al., 2012) and its known role in PM PtdIns4P synthesis (Balla et al., 2008b).

More clearly defined was the localization of PI4KB, which, as expected (Godi et al., 1999; Hama et al., 1999; Walch-Solimena and Novick, 1999; Polevoy et al., 2009), localized exclusively to the Golgi, where it colocalized with P4M (Fig. 8 A). As for PI4K2A and PI4K2B, these exhibit a more scattered localization, with small vacuoles, peripheral vesicles, and a perinuclear accumulation that could correspond to either late endosomal/lysosomes or trans-Golgi membranes (Fig. 8 A). All of these distributions colocalized with a pool of membrane-bound P4M (Fig. 8 A) and are consistent with the reported endosomal/lysosomal and trans-Golgi localization of these enzymes (Balla et al., 2002; Wang et al., 2003; Salazar et al., 2005; Minogue et al., 2006; Minogue et al., 2010; Burgess et al., 2012; Jović et al., 2012). Altogether, these data reveal that all of the cellular PtdIns4P detected by P4M is accounted for by the localization of at least one of the PI4Ks, suggesting the enzymes each have an active role in PtdIns4P synthesis at distinct compartments.

Turning to the catabolic side of PtdIns4P metabolism, we focused our attention on the major PtdIns4P phosphatase Sac1 (encoded by SACM1L in humans). This enzyme localizes predominantly to the ER in mammalians cells (Nemoto et al., 2000; Rohde et al., 2003), where it is present at contact sites with both the PM (Stefan et al., 2011) and Golgi (Mesmin et al., 2013). It also traffics to the Golgi (Rohde et al., 2003) and can accumulate there during periods of cellular quiescence (Blagoveshchenskaya et al., 2008). We expressed an mCherry-tagged mouse Sac1 protein in COS-7 cells to investigate its possible colocalization with P4M. However, we discovered that overexpression of Sac1, despite its characteristic ER and Golgi localization, caused greatly reduced membrane localization of P4M compared with control cells expressing the catalytically inactive C389S mutant (Fig. 8 B), as assessed by the relative fluorescence intensity in membrane regions defined by Golgi, PM, or Rab7 marker proteins normalized to the total cellular fluorescence of the same cell (see Materials and methods). Localization of P4M to both the PM and Golgi compartments was significantly reduced, although the effect on Rab7-positive membranes was less clear (Fig. 8 B).

Is Golgi PtdIns4P broken down by ER-localized Sac1 at contact sites or by Golgi-localized Sac1? The Sac1–leucine-zipper alanine mutant cannot traffic to the Golgi (Blagoveshchenskaya et al., 2008) yet still degraded PM and Golgi PtdIns4P, if not to the same extent as the wild-type enzyme (Fig. 8 B). This implies that ER-localized Sac can access Golgi PtdIns4P pools as described recently (Mesmin et al., 2013), though much less efficiently than when Sac1 traffics into this organelle (Blagoveshchenskaya et al., 2008). The Sac1-K2A mutant accumulated in the Golgi (Rohde et al., 2003), although residual protein in the ER was probably responsible for depletion of PM PtdIns4P (Fig. 8 B). These observations support the recently described roles of ER- and Golgi-localized Sac1 in metabolizing both Golgi (Blagoveshchenskaya et al., 2008; Mesmin et al., 2013) and PM (Stefan et al., 2011) pools of this lipid.

In this paper, we have described a new PtdIns4P probe based on the P4M domain from L. pneumophila SidM. We showed that PtdIns4P is both necessary and sufficient to account for the localization of the probe in live cells. P4M reveals a more extensive localization of PtdIns4P than detected with previous probes, reporting both PM and Golgi pools, as well as a pool associated with late endosomes and/or lysosomes. Although SidM has an exceptionally high (∼10⁻⁸ M) affinity for PtdIns4P in vitro, the isolated P4M domain’s affinity is substantially reduced (Brombacher et al., 2009). This was a distinct advantage for a cellular probe because its affinity seems well matched to the concentration of PtdIns4P in cells. This is indicated by P4M’s ability to sense both increases and decreases in the lipid’s abundance at the PM (Fig. 4, A and B). P4M thus represents an excellent PtdIns4P biosensor for use in live cells and represents a substantial improvement over previously described probes (Roy and Levine, 2004; Balla et al., 2005, 2008a; Halaszovich et al., 2009; Hammond et al., 2009). We believe this tool will be invaluable for interrogating the emerging roles of PtdIns4P in cellular traffic and homeostasis at multiple cellular locations.

This study confirms the substantial accumulation of PtdIns4P at the PM (Hammond et al., 2009; Sarkes and Rameh, 2010; Nakatsu et al., 2012) and reports a pool on the Rab7-positive late endosomal/lysosomal network—as predicted from functional studies of PI4Ks (Salazar et al., 2005; Minogue et al., 2006; Burgess et al., 2012; Jović et al., 2012; Sridhar et al., 2013). What does this tell us about the overall function of PtdIns4P? Its presence at comparable concentrations, as indicated by binding of P4M, in at least three organelle membranes means that a specific PtdIns4P binding protein could not distinguish these compartments based on lipid binding alone. However, PtdIns4P binding may assist in targeting proteins to specific organelles through interaction with other bona fide organelle-selective molecules; examples may include small GTPases, integral membrane proteins, or membrane curvature (Carlton and Cullen, 2005).

Our data do seem to support exclusive localization of PtdIns3P at Rab5-positive early endosomal membranes (Gillooly et al., 2000; Roth, 2004), which appeared devoid of either PtdIns4P or PtdIns(4,5)P₂. A novel probe for PtdIns(3,5)P₂ has recently been reported, which confirms this lipid’s association with the endosomal/lysosomal network (Li et al., 2013). PtdIns(3,5)P₂ overlaps substantially with the early endosome–associated PtdIns3P pools (Li et al., 2013), which we reveal to be distinct from the PtdIns4P-containing endosomes (Fig. 6 A). Nonetheless, it will be intriguing to see whether late endosomal and lysosomal PtdIns(3,5)P₂ is present on the same compartments as PtdIns4P and what the functional relationship between these two lipid pools may be.

Our results also support the conclusion that most membrane PtdIns(4,5)P₂ is present at the PM (Várnai and Balla, 1998; Watt et al., 2002) because the PH-PLCδ1 probe localizes exclusively to this compartment but can detect PtdIns4P-containing membranes when ectopic PtdIns(4,5)P₂ synthesis is driven in them (Figs. 2 C and 6 D). However, PtdIns(4,5)P₂ is not a unique PPIn in the PM, which also contains levels of PtdIns4P far in excess of those required to support basal PtdIns(4,5)P₂ synthesis (Hammond et al., 2012; Nakatsu et al., 2012). Although PtdIns(4,5)P₂ specifically supports a plethora of PM functions (Balla, 2013), it appears to cooperate with PtdIns4P to supply a polyanionic lipid requirement (Hammond et al., 2012), which together with ubiquitous anionic lipids such as phosphatidylserine give the PM a uniquely strong negative electrostatic charge compared with other membranes (Yeung et al., 2008). This negative electrostatic surface is recognized by several proteins containing polybasic motifs, which are therefore specifically recruited to the PM (Heo et al., 2006). However, the role played by PPIn in this electrostatic mechanism appears unique to the PM because endosomal membranes’ weaker electrostatic potential seems to be satisfied by the presence of lipids such as phosphatidylserine (Yeung et al., 2008).

Drawing all of these observations together, it seems clear that PPIn metabolism is tightly restricted on specific membrane compartments. Multiple lipid kinases generate PtdIns4P at several cellular locations, suggesting a disparate function that is surely hinted at by the fact that, uniquely among PPIn kinases, PI4K activity has arisen twice by convergent evolution in the nonhomologous PI4K and PI4K2 families (Brown and Auger, 2011). What is the purpose of synthesizing the same PPIn on several organelle membranes? Recruitment of unique effector proteins to each compartment through tertiary, unique interactions is one possibility (Carlton and Cullen, 2005). However, in over a decade of research, no specific PtdIns4P effectors that function outside of the Golgi have been discovered (Graham and Burd, 2011).

An intriguing alternative is revealed by recent observations that ER-localized Sac1 can act on PtdIns4P generated in the PM (Stefan et al., 2011) or Golgi (Mesmin et al., 2013). Lipid transfer domains that function at these junctions may shuttle PtdIns4P back to the ER for hydrolysis by Sac1, releasing energy from a gradient generated by PtdIns4P synthesis in the PM or Golgi membrane. This energy can be harnessed to drive transport of other lipid cargo against their chemical gradients (Kim et al., 2013; Mesmin et al., 2013). Therefore, PtdIns4P functions as a key component driving nonvesicular lipid transport. We speculate that such a mechanism may also operate at ER–late endosome contact sites mediated by the Rab7-associated ORP1L proteins (Rocha et al., 2009). Intriguingly, a similar PPIn-driven lipid transfer mechanism has also been proposed to function in α-tocopherol transport to the PM using PtdIns(4,5)P₂ (Kono et al., 2013). We suspect that the PPIn system has not arisen in eukaryotes as a simple membrane identity mechanism defining organelle compartments. Instead, we propose that the established roles of the PPIn in nonvesicular lipid transport and the recruitment of peripheral effector proteins are manifestations of the same functional principle: that PPIns are a eukaryotic mechanism that couples the energy of ATP hydrolysis (via PPIn synthesis) to the acquisition and maintenance of nonequilibrium organelle protein and lipid compositions.

In conclusion, we have characterized a novel, highly selective probe for PtdIns4P in living cells that reveals a more widespread distribution than could be detected with any single previous probe. This will facilitate interrogation of PtdIns4P function at multiple cellular locations, including the Golgi, PM, and late endosome/lysosome.

All plasmids were constructed using the pEGFP-C1 or -N1 backbone (Takara Bio Inc.). When different fluors were used, EGFP was replaced with CFP, mTq2 (mTurquoise2; a gift from T. Gadella, University of Amsterdam, Amsterdam, Netherlands; Goedhart et al., 2012), mRFP (Campbell et al., 2002), mCherry (Shaner et al., 2004), or iRFP (Filonov et al., 2011).

The following cDNAs were amplified by PCR and subcloned into plasmids as follows: residues 546–647 from L. pneumophila SidM (available from GenBank under accession no. DQ845395), corresponding to the isolated P4M domain (Schoebel et al., 2010), were cloned into pEGFP-C1 (or spectral variants mCherry or iRFP) at BspEI and EcoRI sites. We also made a similar plasmid inserting the same P4M fragment at SalI and BamHI sites with a GGSASGLRS linker between GFP and the N terminus. This was used as a template to insert a second P4M insert at EcoRI and SalI sites with the same linker between GFP and the first P4M and a GGSAVDGGSASGLRS linker separating the tandem P4M domains. To generate FRB fused to canine Rab5 and Rab7 (a gift from R. Lodge, Institut de Recherches Cliniques de Montreal, Montreal, Quebec, Canada; Rojas et al., 2008), the entire coding regions were inserted at HindIII–KpnI sites of a modified piRFP-C1 vector containing the FRB domain (NCBI Nucleotide accession no. NM_004958; residues 2,021–2,113) flanked by GGAGA and GGSAGGSA linkers at the 5′ and 3′ ends, respectively, and inserted at BglII–HindIII sites. For pmCherry-C1-FKBP-MTM1, FKBP (NCBI Nucleotide accession no. NM_054014; residues 3–109) flanked by GAGGAARAAL and (SAGG)₅PRAQASNSA linkers at the 5′ and 3′ ends were inserted at NotI–SalI sites and MTM1(NCBI Nucleotide accession no. NM_000252) at SalI–BamHI. Hepatitis C NS5A (Budhu et al., 2007) was obtained from Addgene and inserted at HindIII–KpnI sites in pmCherry-N1. piRFP-N1-TTC7B (isolated from an EST that misses L53-A85 from exon 2; GenBank accession no. BQ426031) was inserted at NheI–SalI sites with a C-terminal GGSAGGSA linker with the iRFP. Constructs are available through Addgene. PI4KAv1 (NCBI Nucleotide accession no. NM_058004) was inserted at EcoRI–SalI sites of pmTq2-C1, and PI4KBv2 (NCBI Nucleotide accession no. NM_001198773) was inserted at XhoI–KpnI sites in pEGFP-C1. pEGFP-N1-EFR3B (NCBI Nucleotide accession no. NM_014971) was inserted at NheI–AgeI sites. Murine Sacm1l (NCBI Nucleotide accession no. NM_030692) was inserted at BglII–SalI sites in pmCherry-N1. Additional plasmids were obtained as follows: pEGFP-N1-PH-PLCδ1 (Várnai and Balla, 1998); pEGFP-N1-PH-FAPP1 and pEGFP-C1-PH-OSBP (Balla et al., 2005); PJ, its catalytic mutants, and pECFP-N1-Lyn₁₁-FRB (Hammond et al., 2012); pmRFP-FKBP-INPP5E (Várnai et al., 2006); pECFP-FRB-giantin^{3,140–3,269} (a gift from T. Inoue, Johns Hopkins University School of Medicine, Baltimore, MD; Komatsu et al., 2010) or a pmCherry variant; pECFP-C1-FKBP-PIP5K (Suh et al., 2006); pmCherry-C1-Rab5 and -Rab7 (Rojas et al., 2008); GFP-FYVE-EEA1 (Balla et al., 2000); pEGFP-N1-PI4K2A (Jović et al., 2012); pEGFP-N1-PI4K2B (Balla et al., 2002); and piRFP-C1-PH-PLCδ1 (a gift from P. De Camilli, Yale School of Medicine, New Haven, CT; Idevall-Hagren et al., 2012).

COS-7 cells (CRL-1651; ATCC) were maintained in high glucose DMEM supplemented with l-glutamine and sodium pyruvate, 10% fetal bovine serum, 100 U/ml penicillin, and 100 µg/ml streptomycin (Life Technologies) at 37°C in 5% CO₂ and a humidified atmosphere and passaged with TrpLE (Life Technologies). Cells were seeded in 29-mm dishes with 20-mm #1 cover glass bottoms (In Vitro Scientific) at appropriate densities so that cells reached 50–100% confluence on the day of imaging. 20–26 h before imaging, cells were transfected in 1 ml Opti-MEM precomplexed for 20 min at room temperature with 3 µg Lipofectamine 2000 (Life Technologies) and 1 µg total DNA; the relative amounts of each construct were empirically determined based on the relative expression of each construct combination. After 4 h, complexes and Opti-MEM were removed and replaced with antibiotic-free medium.

For imaging, cells were placed in 800 µl of phenol red–free DMEM containing Hepes and l-glutamine (Life Technologies) and mounted on the microscope stage of inverted microscopes (see details in the following paragraph). When stimulated with compounds as described in the figure legends, 200 µl of the compound was freshly diluted before use at 5× final concentration and added to the dishes by bath application. All imaging was conducted at room temperature (∼22°C).

Confocal imaging was performed on a laser-scanning confocal microscope (LSM 780; Carl Zeiss) operated with ZEN software (Carl Zeiss) using a 63×, 1.4 NA Plan Apochromatic objective lens. Respective laser lines for excitation and spectral detection windows for the fluorochromes were as follows: CFP and mTq2 (405 and 450–490 nm), GFP (488 and 508–543 nm), mRFP/mCherry (561 and 578–649 nm), and iRFP (633 and 650–758 nm). Combinations of all four fluors could be imaged on this system. To prevent cross talk, the spectrally well-resolved CFPs and iRFPs were excited and detected concurrently, whereas both GFPs and orange/RFPs were scanned sequentially. Spinning-disk confocal microscopy was performed on an inverted microscope (Eclipse Ti; Nikon) operated by Elements software (Nikon) and equipped with a 100×, 1.49 NA Plan Apochromatic objective lens, a 10,000-rpm spinning-disk scan head (CSU-X1; Yokogawa Corporation of America), and an electron-multiplying charge-coupled device camera (iXon DU897; Andor Technology). Excitation was with a fiber-coupled monolithic laser combiner (MLC-400; Agilent Technologies) equipped with 405-, 488-, 561-, and 640-nm lines. Appropriate GFP (500–550 nm), RFP (575–625 nm), or Cy5 (660–740 nm) emission filters (Chroma Technology Corp.) were used when sequentially imaging each fluorescent protein. A high speed piezo stage (City; Mad City Labs) was used to acquire z stacks. TIRF was performed on the same system using a second laser fiber output connected to a TIRF launch and a second camera port equipped with an electron-multiplying charge-coupled device camera (iXon DU897). Both microscope systems were equipped with reflected infrared light-driven focus-maintaining mechanisms that were used to ensure focus did not drift during acquisition or bath application of the agonist during time-lapse imaging. All data were acquired at 16-bit data depth. After linear scaling of the intensity, images were exported as 8-bit TIFFs for display in the figures.

Colocalization between two fluorescent signals can be difficult to visualize by overlaying the two channels, especially when the proteins may have different intensities or different levels of contrast between the cellular structure of interest and the cytosolic background. This method is also inherently nonquantitative. Instead, we have used a modified implementation of the normalized mean deviation product (nMDP), a method that allows the quantitative visualization of colocalization between two fluorescent signals (Jaskolski et al., 2005) using a custom-written macro in Fiji (Schindelin et al., 2012). We manually produced regions of interest (ROIs) encompassing the entire cell to be measured (we found that automatic outlining with Sobel’s kernels as originally proposed could not segment neighboring cells nor reliably pick out cells as a result of several contrasting regions in the fluorescent signal). The image was then normalized to the mean deviation of each pixel inside the ROI, i.e., the difference between maximum pixel intensity and the mean. After completing this operation for each channel, the nMDP was calculated as the pixel-by-pixel multiplication of the two images. Thus, regions in the two channels where intensities correlate will tend to produce larger, positive values, and regions where they differ produce smaller or more negative values. The overall index thus gives a statistically tractable measure of colocalization. For nMDP images shown in the paper, pixels in which both deviation products were below the mean, i.e., both negative (but producing a positive deviation product), were set to a value of 0 for clarity (i.e., black on the look-up table). Data were exported to Prism v6 (GraphPad Software), and all images were ranked by nMDP score. Images were selected for presentation that displayed clear morphological features and a high signal-to-noise ratio but that were close to the median score (and always fell within the interquartile range). Data failed a D’Agostino and Pearson omnibus normality test even after a natural logarithmic transformation, so we used the nonparametric Mann–Whitney U test for significant differences between control and experimental groups (Fig. 7).

Image intensity measurements were made in Fiji. We have previously used a PM index based upon the ratio of intensity of a fluorescent protein in a membrane compartment relative to the cytosol to produce quantitative and statistically tractable measurements from our imaging data (Hammond et al., 2012). However, the large number of compartments labeled by P4M, and the motility of these compartments, made defining fixed ROI in the cytoplasm problematic: vesicular structures continuously moved in and out of these zones. Therefore, after manually defining ROIs containing whole cells, we normalized pixel intensity to the mean value of the whole cell at each time point. This corrects for both photobleaching and changes in protein concentration in the cytosol as levels at particular membranes change. To define the membrane compartment for measurement, we generated a binary mask of the compartment based on markers that were usually the FRB-conjugated recruiters (hence values are expressed as I_Rab7 for intensity within the Rab7 mask, etc.). Masks were generated by an à trous wavelet decomposition at three length scales as previously described (Olivo-Marin, 2002) using a custom macro written for Fiji. In brief, this method smooths the raw image with Gaussian kernels of increasing pixel radii (i.e., length scales). These smoothed images are subtracted from the image smoothed at the preceding shorter length scale to produce detail images (the wavelet planes). The product of these three detail images was then calculated, and the modal pixel value (corresponding to featureless background) was subtracted before converting to a binary image. We found that that this approach could produce adequate, though imperfect, masks for each of the membrane compartments considered in this manuscript. We did not need to define a hard threshold for each wavelet as described by Olivo-Marin (2002), although the poorer signal to noise in images of the PM marker Lyn₁₁ meant that we subjected these images to a 3 × 3 mean filter before applying the decomposition.

These masks thus produced an output of the underlying pixel intensity relative to the whole cell fluorescence (so a value of 1 indicated no enrichment on the membrane relative to the whole cell). Numbers are usually quoted as the mean pixel ratio. The one exception is for the experiment in Fig. 6 A, in which images of GFP-FYVE and mCheryy-P4M were used to generate their own masks. Here, the total (integrated) pixel ratio is quoted because the masking objects themselves could disappear. For time-lapse data, images were normalized to the pixel ratio at the start (time 0) because the relative change in each cell is the paramount measurement, as opposed to the absolute value of the pixel ratio in each compartment, which varies between cells. However, for the data in Figs. 5 (B and C) and 8 B, the absolute pixel ratio value for each cell is reported because comparisons between cells are made. Data were exported, graphed, and analyzed in Prism v6. The pixel ratio data were transformed with the natural logarithm to approximate a normal distribution and subjected to parametric tests. These were either Student’s t test or two-way analysis of variance (ANOVA), with Bonferroni’s post test applied to significantly varying datasets. For time-lapse data, all comparisons are to the time point immediately before the addition of compound. P-values are quoted on the graphs. Where not indicated, P > 0.05.

Videos 1–10 show time-lapse confocal imaging of a 3D maximum intensity projection of GFP-P4M (Video 1), the effects of PJ-Sac recruitment on the Golgi localization of GFP-P4M (Video 2), FKBP-PIP5K recruitment on the Golgi localization of PH-GFP (Video 3), 5 µg/ml Brefeldin A on the localization of PH-FAPP1-GFP and mCherry-P4M (Video 4), PJ-Sac recruitment on the PM localization of GFP-P4M and iRFP-PH-PLCδ1 (Video 5), FKBP-INPP5E recruitment on the PM localization of GFP-P4M and iRFP-PH-PLCδ1 (Video 6), mCherry-P4M colocalization with LysoTracker green–labeled endosomes/lysosomes (Video 7), the effects of Wortmannin on the localization of GFP-FYVE and mCherry-P4M (Video 8), PJ-Sac recruitment on the late endosomal/lysosomal localization of GFP-P4M (Video 9), and FKBP-PIP5K recruitment on the late endosomal/lysosomal localization of GFP-PH-PLCδ1 (Video 10). Additional data are available in the JCB DataViewer at http://dx.doi.org/10.1083/jcb.201312072.dv.

We are grateful to Pietro De Camilli, Takanari Inoue, Theodorus Gadella, and Robert Lodge for the kind gift of the plasmids detailed in the Materials and methods. Confocal and TIRF microscopy was performed in the Microscopy and Imaging Core of the Eunice Kennedy Shriver National Institute of Child Health and Human Development with the kind assistance of Drs. Vincent Schram and James Russel. We also thank Robin Irvine, Joseph Albanesi, Julie Brill, and Marko Jović for insightful comments on the manuscript.

The work was supported by the intramural program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health.

The authors declare no competing financial interests.