Understanding phosphoinositides: rare, dynamic, and essential membrane phospholipids

Abstract

Polyphosphoinositides (PPIs) are essential phospholipids located in the cytoplasmic leaflet of eukaryotic cell membranes. Despite contributing only a small fraction to the bulk of cellular phospholipids, they make remarkable contributions to practically all aspects of a cell’s life and death. They do so by recruiting cytoplasmic proteins/effectors or by interacting with cytoplasmic domains of membrane proteins at the membrane–cytoplasm interface to organize and mold organelle identity. The present study summarizes aspects of our current understanding concerning the metabolism, manipulation, measurement, and intimate roles these lipids play in regulating membrane homeostasis and vital cell signaling reactions in health and disease.

Introduction

PPIs ( polyphosphoinositides) are reversibly phosphorylated derivatives of the membrane phospholipid phosphatidylinositol (PI, Figure 1A). They are low-abundance lipids of the cytoplasmic leaflet of all eukaryotic cellular membranes. PPIs are amphiphilic, meaning they have a polar inositol head group that faces toward the cytoplasm and non-polar hydrophobic fatty acid tails embedded in the lipid bilayer (Figure 1A). The inositol ring of PI can be phosphorylated by numerous cytoplasmic lipid kinases that add phosphates to the hydroxyl groups on inositol positions 3, 4, or 5, giving polyacidic lipids with a high negative charge [e.g. see PI(4)P synthesis from PI in Figure 1B]. Complementary, lipid phosphatases remove the phosphate. Such reversible phosphorylation gives rise to seven total PPI species (Figure 1C) with dissimilar localizations in the cell. Figure 2A shows a map of the preferred distribution of specific PPIs among the membranes of cellular organelles and the cell surface. We will refer to the PPIs as forming a ZIP code of signature lipids instructing cellular proteins where to bind or where to be active. Nevertheless, the concept of signature lipid should not be taken too strictly. For example, when we describe the Golgi as being a PI(4)P organelle (Figure 2A, blue membrane), this reflects a preponderance of the PI(4)P form but not that PI(4)P is the only species found there. Typically, at least a precursor of the signature lipid will also be present. In addition, cellular activities and signaling may alter the PPI composition of any membrane profoundly but transiently.

Figure 1. PPI biogenesis.

(A) Chemical structure of 1-stearoyl 2-arachidonoyl phosphatidylinositol (PI 38:4). Arrows pointing at the D3, D4, and D5 position on the myo-inositol head group indicate the three reversibly phosphorylatable hydroxyl groups. (B) Model of PI(4)P production from PI at the cytoplasmic (cyto.) interface of the plasma membrane (PM). (C) Diagram summarizing major PPI lipid kinase and phosphatase reaction pathways. Red arrows represent the PPI lipid kinases and blue arrows represent PPI lipid phosphatases. Red and blue labels are the gene name of enzymes capable of catalyzing each reaction. Gene names with question marks (?) and dashed reaction arrows represent enzymes with some ambiguity surrounding their ability to catalyze a specific reaction.

Figure 2. PPI metabolism at MCSs.

(A) Diagrammatic representation of the preferential distribution of PPIs in a eukaryotic cell. (B) Known proteins residing in ER– PM MCSs arranged in a hypothetical scenario. It remains to be fully tested whether each MCS contains this full standardized set of proteins or rather just a subset. For the dynamic equilibrium of PPI metabolism, (I) PI is transferred from the ER to the PM via TMEM24 dimers and this PI is then converted to PI(4)P by PM PI 4-kinases. (II) Under times of VAP-mediated Nir2 recruitment, PI would be transferred to the PM in exchange for phosphatidic acid (PA). (III) PA is subsequently converted to PI in a multi-step reaction. (IV) ORP5/8 countertransports PM PI(4)P to the ER in exchange for phosphatidylserine. Transferred PI (4)P is subsequently dephosphorylated to PI by the ER-resident phosphatase Sac1. (V) PM PI(4,5)P₂ supports PM ion channel function and is also the substrate of PLC. Receptor-mediated activation of PLC hydrolyzes PM PI(4,5)P₂ into IP₃ and DAG. IP₃ binds to IP₃ receptors on ER membranes and to initiate release of Ca²⁺ from the ER, while DAG recruits PKC. Excess PM DAG may be cleared from the PM to the ER via extended synaptotagmin 2 (E-Syt2). (C) Known proteins residing in ER-Golgi MCSs. (I) PI is transferred from the ER to Golgi via VAP-A-Nir2 interactions. (II) Golgi PI is the substrate for PI4KIIIβ which generates PI (4)P in an Arf-1-dependent manner. (III) OSBP tethers ER-Golgi membranes through FFAT-mediated interactions with VAP-A on ER membranes and PH domain binding of PI(4)P on Golgi membranes. The ORD (OSBP-related domain) domain of OSBP can bind and transport PI(4)P from Golgi to ER membranes and cholesterol (against its concentration gradient) to Golgi membranes. (IV) Transferred ER PI(4)P is subsequently dephosphorylated into PI by ER Sac1, supporting the steep PI(4)P gradient between the two membranes.

We first discuss PPI biogenesis, before defining the cardinal roles these lipids play in orchestrating cellular signaling cascades, and the devastating consequences that disorders of PPI metabolism have for human health.

Biogenesis of phosphoinositides

PI, the precursor of all phosphoinositides, is generally thought to be synthesized in the endoplasmic reticulum (ER) by conjugation of myo-inositol with CDP-DAG (diacylglycerol) by a PI synthase (PIS) enzyme (Figure 2B). PI produced in the ER is believed to be transported out to requiring membranes via traditional vesicular transport and by non-vesicular lipid transfer protein mechanisms at membrane contact sites (MCSs) (see section ‘Molecular architecture of phosphoinositide transport’). Additionally, one study identified highly mobile PI-synthase-containing vesicles that were postulated to be sites of PI synthesis [1]. These platforms originated from the ER, moved rapidly throughout the cytoplasm, and have been suggested to mediate the delivery of PI to other organelles such as the plasma membrane (PM) or its de novo synthesis [2] at non-ER cellular membranes. Despite the abundance of PI among mammalian phospholipids (∼10–20 mol%), we ironically have the least information regarding its subcellular distribution, as currently there are no reports of PI-specific biosensors. Nevertheless, some information can be gleaned from monophosphorylated PI(3)P and PI(4)P. The knowledge that PI can be converted either to PI(3)P on endosomal membranes or to PI(4)P at the trans-Golgi and PMs suggests that PI is available on these membranes for de novo synthesis of the designated monophosphorylated phosphoinositide. The synthesis, distribution, delivery, and metabolism of PI remains an area with significant questions still to be answered.

Phosphatidylinositol 4-phosphate

(PI(4)P) can be generated in mammalian cells via one of three pathways: (1) phosphorylation of phosphatidylinositol at the 4-position of the inositol ring (Figure 1B), (2) dephosphorylation of PI(4,5)P₂ by PI(4,5)P₂-5-phosphatases, or (3) dephosphorylation of PI(3,4)P₂ by PI(3,4)P₂-3-phosphatases (Figure 1C). A combination of cell fractionation, radiolabeling, fluorescent biosensors, and immunofluoresence investigations have confirmed that PI(4)P is found in significant amounts across several organelle compartments, especially the PM and the Golgi (Figure 2A). The PM pool of PI(4)P, which acts as a precursor for PM PI(4,5) P₂, appears to be generated mostly by the PI4KA enzyme [3], which is targeted via an evolutionarily conserved complex containing TTC7B, FAM126A and EFR3 [3] (Figure 1B). The majority of Golgi PI(4)P is localized to the trans-Golgi, where PI4KB along with PI4K2A and B are responsible for its synthesis via lipid modifications and Arf1-dependent recruitment, respectively (Figure 2C) [4–7]. Following its production, Golgi PI(4)P is tightly regulated by the ER lipid PI(4)P-4-phosphatase, Sac1 (Figure 2B,C). (Like many other names of PPI-metabolizing enzymes, the name Sac1 comes from yeast genetics.) Whether Sac1 dephosphorylates its substrate in the cis configuration [acting on PI(4)P in the ER] or the trans configuration [reaching out and acting on PI(4)P on another membrane], or both, has been a topic of discussion. Crystal structures suggest that the long flexible linker between the catalytic domain and the transmembrane domain could allow Sac1 to hydrolyze PI(4)P in either configuration [8]. Currently, there appears more evidence for Sac1 working in cis [9–11] to dephosphorylate PI(4)P on ER membranes following its transfer by OSBP at ER-Golgi membrane junctions (Figure 2C; see ‘ER-Golgi sites’).

Phosphatidylinositol-3-phosphate

(PI(3)P) is a key player in membrane dynamics and trafficking. Based on biosensor information, the majority of cellular PI(3)P is found on early endosomal autoantigen (EEA1)-positive vesicles, suggesting that this monophosphorylated PPI is most abundant in an early endosome compartment. In vitro, PI(3)P can be generated following the hydrolysis of PI(3,4)P₂ by INPP4A and INPP4B [12,13]. Cellular experiments further underscore the concept that INPP4A can produce PI(3)P with its overexpression rescuing the enlarged endosome morphology caused by PI(3)P deficiency [14]. Alternatively, phosphorylation of PI on the 3-position of the inositol ring by the activity of the class III PI 3-kinase (PIK3C3) [15], with additional contributions coming from the class II 3-kinase [16,17], produces PI(3)P. In humans, PIK3C3 seems to exist in at least two heterotetrametric complexes: complex I (containing PIK3C3, p150, Beclin1, and ATG14L) functioning in autophagy, and complex II (containing PIK3C3, p150, Beclin1, and UVRAG) participating in endocytic sorting [18,19]. The turnover of PI(3)P is regulated by myotubularin family phosphatases that dephosphorylate it to PI. Additionally, the PI(3)P 5-kinase, PIKfyve, can convert PI(3)P into PI(3,5)P₂ as endosomes mature into late endosomes/multivesicular bodies.

Phosphatidylinositol 5-phosphate

(PI(5)P) synthesis and cellular functions remain poorly understood. Similar to PI(3)P, it is present only in low amounts in mammalian cells. Once considered an intermediate in the synthesis of phosphatidylinositol bisphosphonates or trisphosphate, PI(5)P has been reported to be present at the PM and endomembranes, as well as the nucleus. At these diverse cellular locations it appears to have roles in Akt/mTOR signaling [20], apoptosis [21], and as a nuclear transducer of stress signaling [22], respectively. The biochemical route for PI(5)P generation in mammalian cells remains controversial. The most ‘direct’ route is generation by phosphorylation of PI by a PI 5-kinase (such as PIKfyve or Type I PI5K enzymes), with an ‘indirect’ route via dephosphorylation of PI(3,5)P₂ by myotubularin phosphatases [23] (Figure 1). Interestingly, during the early phases of an infection by Salmonella, PI(5)P is produced from PI(4,5)P₂ [24] via the actions of type I and II PI(4,5)P₂ 4-phosphatases [25,26]. In contrast, the type II P(5)P 4-kinase can transform PI(5)P into PI(4,5)P₂. In resting cells, it appears that the amount of PI(4,5)P₂ synthesized from PI(5)P is minor; however, it cannot be ruled out that local PM PI(5)P pools could serve as precursors for the generation of PM PI(4,5)P₂.

Phosphatidylinositol 4,5-bisphosphate

(PI(4,5)P₂), the most abundant phosphorylated PPI, is the signature phosphoinositide of the PM (Figure 2A). It is produced primarily through the sequential phosphorylation of PI by PI 4-kinases [27] to generate PI(4)P and type I PI(4)P 5-kinases to generate PI(4,5)P₂. The three type I PI 5-kinases (α, β, and γ) are localized to the PM and involved in the conversion of PI(4)P into PI(4,5)P₂ [28]. In human cells, the type I PIP5Kβ forms homo- and hetero-dimers with the PIP5Kγ that appear essential for enzymatic activity and localization to the PM [29]. Additionally, PI(4,5)P₂ is also produced through dephosphorylation of PI(3,4,5)P₃ by PI(3,4,5)P₃-3-phosphatases, such as PTEN, TPIPα, β, and γ [30]. PI(4,5)P₂ can be down-regulated at the PM via two main pathways; (1) receptor-mediated hydrolysis by various subtypes of phospholipase C (PLC), and (2) dephosphorylation by members of the PI(4,5)P₂-5-phosphatase family (SYNJ1, SYNJ2, OCRL, INPP5B, INPP5E, INPP5J, INPP5K: see Figure 1C). The importance of 5-phosphatase-mediated regulation of cellular PI(4,5)P₂ concentrations is highlighted by the broad spectrum of diseases and disorders, including various cancers, obesity, type 2 diabetes, neurodegenerative diseases, and rare genetic conditions, which result from mutations or alterations in the expression of these enzymes (see section ‘Defective phosphoinositide metabolism in disease’).

Phosphatidylinositol 3,4-bisphosphate

(PI(3,4)P₂) represents <0.1% of total PPI (Table 1) in human cells under basal conditions and mainly localizes to the PM and endocytic compartments (Figure 2A). The majority of PI(3,4)P₂ seems to be formed downstream of receptors that activate PI3K, consistent with its generation from PI(3,4,5)P₃ [31,32] by INPP5D (SHIP1) and INPPL1 (SHIP2) phosphatases (Figure 1C). An alternative pathway that could potentially increase PI(3,4)P₂ at the PM is phosphorylation of PI(4)P via class II PI3K lipid kinases [33]. In addition, the PPI 5-phosphatases INPP5D, INPPL1, OCRL1, INPP5B, as well as synaptojanins 1 and 2 can dephosphorylate PI(3,4,5)P₃ into PI(3,4)P₂ (Figure 1C), it remains to be fully understood if these enzymes participate in its generation in vivo. PI(3,4)P₂ is under a further layer of regulation through the metabolic phosphatase actions of INPP4A/B [34] and PTEN [35], that act on their substrates to generate PI(3)P and PI(4)P, respectively. Thus, the current view is that PI(3,4)P₂ is transiently elevated at the PM following stimulation of growth factor receptors and persists through and contributes towards endocytotic events, before being trafficked to more distal membrane compartments.

Table 1.

Abundance, location, measurement, and roles of PPI lipids in eukaryotic cells

Lipid	Abundance	Distribution	Assays for measurement	Cellular role
Phosphatidylinositol (a.k.a. PI, PtdIns)	∼80 mol% of total cellular PPIs	Abundant in ER but potentially all membranes	TLC, HPLC-MS/MS, [3H] or [32P] radiolabeling to equilibrium	Precursor for PI(3)P and PI(4)P
Phosphatidylinositol 4-phosphate (a.k.a. PI (4)P, PtdIns4P, PI4P)	∼2–5 mol% of total cellular PPI	PM, endosomes, trans-Golgi	TLC, HPLC-MS/MS, [3H] or [32P] radiolabeling to equilibrium, immunofluorescence, fluorescent biosensor (P4M)	Precursor for PI(4,5) P2, recruitment of clathrin adaptors, lipid binding to OSBP, ORP, CERT, FAPP2.
Phosphatidylinositol 3-phosphate (a.k.a. PI (3)P, PtdIns3P, PI3P)	∼0.2–0.5 mol % of total cellular PPI	Early endosomes	TLC, [3H] or [32P] radiolabeling to equilibrium, fluorescent biosensors (FYVE)	Involved in endocytic vesicle fusion and trafficking, autophagy
Phosphatidylinositol 5-phosphate (a.k.a. PI (5)P, PtdIns5P, PI5P)	∼0.01 mol% of total cellular PPI	PM, endosomes, nuclear envelope	TLC, [3H] or [32P] radiolabeling to equilibrium	Stress signal, apoptosis, Akt/ mTOR signal pathway
Phosphatidylinositol 4,5-bisphosphate (a.k. a. PI(4,5)P2, PtdIns4,5P2, PI4,5P2, ‘PIP2’)	2–5 mol% of total PPI	PM, recycling endosomes, lysosomes	TLC, HPLC-MS/MS, [3H] or [32P] radiolabeling to equilibrium, immunofluorescence, genetically encoded fluorescent biosensor (PH domain from PLCδ1 for PM labeling), ion channel currents.	Intimately involved in key PM events, including regulation of endo- and exocytosis, phagocytosis, cell motility, signal transduction, and ion channel function
Phosphatidylinositol 3,4-bisphosphate (a.k. a. PI(3,4)P2, PtdIns3,4P2, PI3,4P2)	<0.1mol% of total PPI	PM, early endosomes	TLC, HPLC-MS/MS, [3H] or [32P] radiolabeling to equilibrium, immunofluorescence, genetically encoded fluorescent biosensor (TAPP1 PH domain)	PI3K/Akt signaling pathway, clathrin-coated vesicle maturation, endocytic trafficking
Phosphatidylinositol 3,5-bisphosphate (a.k. a. PI(3,5)P2, PtdIns3,5P2, PI3,5P2)	<∼2 mol% of total PPI	Late endosomes and Lysosomes	TLC, HPLC-MS/MS, [3H] or [32P] radiolabeling to equilibrium, immunofluorescence, genetically encoded fluorescent biosensor (Fyve)	Protein sorting at the late endosomes/ multivesicular bodies
Phosphatidylinositol 3,4,5 trisphosphate (a. k.a. PI(3,4,5)P3, PtdIns3,4,5P3, PI3,4,5P3, PIP3)	<0.05% of total PPI	PM, some endocytic compartments	Thin layer chromatography (TLC), HPLC-MS/MS, [3H] or [32P] radiolabeling to equilibrium, immunofluorescence, genetically encoded fluorescent biosensor (AKT, BTK biosensors).	Cell proliferation and cell survival, cytoskeleton dynamics, cell motility, membrane trafficking and apoptosis

Abbreviations: ER, endoplasmic reticulum; HPLC, High-pressure liquid chromatography; MS, mass spectrometry; PM, plasma membrane; PPI, Polyphosphoinositides; TLC, Thin layer chromatography.

Phosphatidylinositol 3,5-bisphosphate

(PI(3,5)P₂) is the signature PPI of late endosomal membranes. It regulates endosomal fission and fusion to maintain endomembrane homeostasis and exchange of membrane cargo. To date, there have been no reports of a PPI 4-phosphatase hydrolyzing PI(3,4,5)P₃ to generate PI(3,5)P₂ in cells. Therefore, the only currently known route for PI(3,5)P₂ production in humans is through synthesis catalyzed by the PI(3)P 5-kinase, PIKfyve [36] (Figure 1C). As in yeast cells, VAC14 serves as a platform regulating PI(3,5)P₂ synthesis by interacting directly with PIKfyve, FIG4/SAC3, and VAC7 to fine tune the regulation of PI(3,5)P₂ levels [37,38]. At endosomes, production of PI(3,5)P₂ induces the release of cortactin from the endosomal, branched actin network via direct interaction between the actin filament-binding region of cortactin and PI(3,5)P₂. This regulation is important for membrane trafficking since actin cytoskeleton dynamics regulate membrane curvature and transport of vesicles [39]. Finally, the importance of regulated PI(3,5)P₂ abundance for cellular processes is highlighted by the embryonic lethality of PIKfyve^−/− knock-out mice [40].

Phosphatidylinositol 3,4,5 trisphosphate

(PI(3,4,5)P₃) represents <0.05% of total PPI in human cells, and in quiescent cells it is almost undetectable. The class I PI3Ks, of which there are four isoforms in mammalian cells (α, β, δ, and γ), can be activated by a variety of cell-surface receptors to increase PI(3,4,5)P₃ levels rapidly and transiently up to 100-fold in the inner leaflet of the PM [41,42]. The major reason that this minor lipid abundance is so tightly regulated is that it plays a central role in key signaling pathways: cell proliferation, migration, growth, survival, and cancer. For example, the pleckstrin homology (PH) domains of proteins such as Akt/PKB, PDK1, and Btk1 interact electrostatically with PI(3,4,5)P₃. This means that receptor-mediated elevations in PI(3,4,5)P₃ recruit these kinases to the PM to shape cellular signaling cascades. In many cell types, the production of PI(3,4,5)P₃ and subsequent recruitment of PH domain-containing proteins to the cytoplasmic leaflet results in local actin polymerization, important for cell migration and division. The many downstream consequences of PI(3,4,5)P₃ production highlight the oncogenic potential of such a powerful anabolic signal and the essential need for regulated metabolism of this minor lipid. A major regulator is the PTEN phosphatase, which catalyzes the dephosphorylation of PI(3,4,5)P₃ on the 3-position to regenerate PI(4,5)P₂. PTEN has been characterized extensively as a tumor suppressor [30], and mutations in the PTEN gene are found in many cancers [43]. Indeed, loss of PTEN phosphatase activity, coupled with the very common mutations in the PIK3CA isoform, makes enhanced PI3K signaling the most common oncogenic event [44].

Molecular architecture of phosphoinositide transport

We have been discussing enzymatic pathways and molecular mechanisms underlying the generation and destruction of each PPI species in a membrane-specific way. This is an oversimplification. Indeed, cellular membranes communicate extensively with one another via two distinct cellular processes: by traditional membrane trafficking and by lipid exchange at MCSs. PPIs co-ordinate and regulate intracellular trafficking by acting as molecular beacons and effectors to co-ordinate and recruit important trafficking machinery. Furthermore, the continuous membrane flow between organelle compartments mediated by vesicular traffic, coupled with precise spatial and temporal localization of PPI-metabolizing enzymes, is an essential process that ensures the heterogeneous distribution of PPIs throughout eukaryotic cells. Underscoring the symbiotic relationship between trafficking, PPI enzyme localization, and PPI distribution, local changes in PPI levels mediate fine-tuning of key cellular events, such as vesicular budding and membrane fusion, whereas gross manipulation of PPI abundance at the Golgi [45–47], lysosomes [48], or endosomes [49] results in disruption of trafficking organelles and downstream reorganization of the PPI zip code. It is clear that co-ordinated vesicular trafficking is required for normal PPI distribution.

The second class of transport mechanisms used to transport PPIs are MCSs. These sites are platforms of information transfer akin to synapses where two intracellular membranes come into close proximity to one another (∼15 nm) and allow lipids (among other things) to be exchanged. The concept of close proximity between two organelle membranes without membrane fusion is important. Unlike traditional vesicular trafficking that transports cargo (lipids and protein) from one organelle to another via membrane fusion, the exchange of lipids at MCSs occurs via protein-mediated transport or exchange (Figure 2B,C). The lipid transfer proteins form a protected conduit — a bridge — for lipid transfer. Thus, organelle identity and architecture are preserved as molecules are transferred. Despite great heterogeneity in the molecular elements defining each MCS, it is striking that the majority of these MCSs complexes contain and transport several PPIs and catalyze their metabolism.

ER–PM contact sites

The best characterized MCS in terms of PPI metabolism is that between the ER and the peripheral PM (Figure 2B). First characterized in muscle cells in the 1950s, an essential role for ER–PM contact sites is the regulation of cytoplasmic calcium during excitation–contraction coupling in striated muscle cells [50–52]. More recently, it has become increasingly evident that ER–PM MCSs also provide a rapid non-vesicular transport pathway for PPIs, phosphatidylserine, and sterols to be transferred and modified at the ER and PM.

Transfer of PI at MCSs:

The precursor for all PPIs is PI. PI can be transferred to the PM from the ER via the ER-resident membrane protein TMEM24 [53] (Figure 2A). The TMEM24-mediated transfer of PI is enabled by C2-domain interactions with the PM and by a lipid transport module of the synaptotagmin-like, mitochondrial, and lipid-binding protein (SMP) family, which binds glycerolipids, including PI. The binding of ER TMEM24 ‘in trans’ with the PM occurs in a phosphorylation- and Ca²⁺ -dependent manner, allowing the protein to disconnect transiently from the PM as the Ca²⁺ concentration spikes and then reassociate upon dephosphorylation [53]. TMEM24 is thought to be important in neuro-endocrine cells for regulated insulin secretion [53]. Another mechanism of PI transfer at ER–PM MCSs involves the phosphoinositide transfer proteins (PITPs), Nir2 and Nir3 (Figure 2B). Following Gq-coupled receptor stimulation, PM PI(4,5)P₂ is hydrolyzed and Nir2 and Nir3, are recruited to ER–PM MCSs (Figure 2B) [54–57]. This recruitment is triggered through the combined actions of phosphatidic acid and DAG [54,56] and mediated through an interaction with the FFAT motif on the ER vesicle-associated membrane protein (VAMP)-associated proteins A and B (VAP-A and VAP-B) [58] (Figure 2B). The stimulated recruitment of Nir2 and Nir3 at ER–PM junctions delivers PI to the PM as a precursor to replenish PM PPIs, while concurrently returning phosphatidic acid to the ER for future PI regeneration [56].

Transfer of PI(4)P at ER–PM MCS:

There are two major pools of PI(4)P within cells, in the PM and in the Golgi. The majority of PM PI(4)P appears to be generated by PI4KIIIα and its associated factors (Figure 1B,C). Recently, two integral ER membrane proteins, oxysterol-binding protein (OSBP)-related protein 5 (ORP5) and 8 (ORP8), have been identified as additional key regulators of PM PI(4)P [59,60]. ORP5 and 8 tether ER–PM MCSs through the action of a hydrophobic tail sequence that is anchored in the ER membrane and a PH domain that interacts with PM PI(4)P and PI(4,5)P₂ [59]. These ORP junctions facilitate the countertransport of phosphatidylserine (PS) for PI(4)P. Transferred PI(4)P is then dephosphorylated to PI by the PI(4) P-4-phosphatase Sac1 on ER membranes. The major distinction between ORP5 and 8 is their binding affinities towards PI(4)P and PI(4,5)P₂, with ORP5 capable of engaging PM PPIs under resting conditions, whereas ORP8 requires higher levels of each substrate. These differences in PPI affinity present two modes of action for ORP5 and 8. Mode one involves ORP5 binding and transferring PI(4)P to the ER under resting levels of PI(4) P. Here, decreases in PM PI(4)P or PI(4,5)P₂ will uncouple ORP5 from the PM to aid replenishment of each PPI pool. Mode two involves ORP8, which binds PI(4)P and PI(4,5)P₂ under periods of excess PI(4)P and PI (4,5)P₂ production. Here, increases in PM PI(4)P and PI(4,5)P₂ would enhance PM ORP8 binding to facilitate increased transfer of PI(4)P to the ER, reducing PM PI(4)P and PM PI(4,5)P₂. In both scenarios, ORP5 and 8 proteins serve as a rheostat regulating PM PI(4)P levels.

ER-PM contact sites are under a further layer of regulation from the extended synaptotagmin (E-Syt) family of proteins [61]. These three (E-Syt 1, 2, 3) ER-resident proteins interact with PM PI(4,5)P₂, via their C2 domains, to tether ER-PM membrane contact sites. In the context of PPI metabolism, E-Syt1 and E-Syt2 appear important for the extraction and transfer of DAG from the PM to the ER during periods of G_q coupled receptor activation (Figure 2B) [61]. Given this role, it seems likely that E-Syt’s act in cooperation with the aforementioned Nir2 pathway to stabilize ER-PM contact sites and control the accumulation of PI(4,5)P₂ metabolites, while providing precursor elements for future PI synthesis.

ER-Golgi contact sites

Transfer of PI(4)P at ER-Golgi MCSs:

At ER-Golgi MCSs three lipid transfer proteins with FFAT domains bind ER-localized VAP proteins: CERT, Nir2, and OSBP. OSBP1 contains an ORD and PH domain to bind and transfer both sterols and PI(4)P, respectively. Under conditions of high Golgi PI(4)P, paired with the presence of VAP-A and -B on ER membranes, OSBP is recruited and tethers ER-Golgi junctions (Figure 2C). Once positioned at ER-Golgi MCSs the ‘OSBP cycle’ proceeds wherein cholesterol on ER membranes is exchanged for PI (4)P on trans-Golgi membranes [11,62]. The energy required for this cycle is supported by the steep PI(4)P gradient between Golgi and ER. There is continuous dephosphorylation of PI(4)P by the ER-localized Sac1 and continuous production of PI(4)P at the TGN by PI4-kinases, notably PI4KIIIβ located close to OSBP in an Arf1-dependent manner. Accordingly, in the OSBP cycle, PI(4)P serves as a lipid form of free energy to support the unidirectional transport of cholesterol from the ER (site of its synthesis) to the Golgi, against its concentration gradient. In this sense, OSBP function is analogous to that of the Na⁺/Ca²⁺ exchanger that uses the energy stored in the electrochemical gradient of Na⁺ to drive Ca²⁺ extrusion out of the cell against its concentration gradient.

ER–endosome contact sites

Recent studies have revealed multiple MCSs between the ER and endosomes [63–66]. Two of them have a PPI as a central component to promote microtubule-dependent endosomal transport [67] or retromer- and WASH-dependent budding from endosomes [68]. For PPI-dependent, ER–endosome-mediated transport, the ER protein protrudin forms contact sites with late endosomes through electrostatic interactions with PI(3)P in a Rab7-GTP-dependent manner. This coincidence binding on late endosome membranes promotes MCS formation and enables hand-over of kinesin-1 from protrudin to the adaptor protein FYCO1 on late endosomes. Consequently, late endosomes are transported via plus-end-directed transport along microtubules into growing neurites to promote protrusion formation and neurite outgrowth.

Contacts between the ER and endosomes also influence vesicle budding through changes in lipid composition that affect actin dynamics. These MCSs are formed by VAPA/B binding to the retromer subunit SNX2 and to OSBP, which binds PI(4)P on the endosomal membrane via its PH domain. In WT cells, transient accumulation of PI(4)P on endosomes via type II PI4Ks is coupled to a transient burst of WASH-dependent actin nucleation to facilitate retromer function [68]. In cells that lack VAP, excessive and persistent PI(4)P accumulation on endosomal membranes causes disruption of retromer-dependent budding and disruption of trafficking between endosomes and the Golgi complex.

It is worth noting that ER–endosome contact sites containing either protrudin or VAP-retromer expand the role of MCS from stages of lipid and calcium transfer to hubs of protein docking and sorting that control critical cellular events.

Tools for exploring phosphoinositide pools

Part of the reason that we know so much about the biogenesis and distribution of these rare lipids is development of new tools. Indeed, outstanding cell biologists who excel in logic, biochemistry, genetics, microscopy, and molecular biology have pioneered development of biosensors for detecting PPIs and other tools that alter PPI abundance. Using these complementary approaches allows investigators to visualize the PPI species in question and to manipulate that species to determine its potential function. We now discuss techniques and strategies for the detection and manipulation of these lipids. Such outstanding tools, expanded upon below, have been instrumental in revealing the important cellular roles of PPIs.

Detecting the PPIs

Historically, the most practical approach to monitor pools of cellular PPIs and their dynamics is via classical biochemical approaches that depend on chemical analysis of lipid extracts from biological samples. These methodologies measure the aggregate PPI content of populations of cells rather than the content of a specific cellular membrane or organelle. Although it is possible to assay subcellular fractions, the important phosphorylation state of PPIs may change during procedures lasting many minutes, requiring fastidious attention to slowing the actions of endogenous lipid phosphatases and kinases. The chemical methods often need >100 000 cells, and they are static, with each assay resolving only one time point. One of the first methods was equilibrating preincubation of cells with [³²P]phosphate or [³H]inositol for several days, acid extraction of lipids into an organic phase, and reading the counts in the various peaks resolved by thin-layer or other chromatography. In such labeling experiments, it is possible with different elution solvents to discriminate most of the major PPI classes and to determine their relative amounts. With briefer incubations with [³²P]phosphate instead, it is possible to observe fluxes, the rates of synthesis of PPIs. A more elaborate method without labeling, separates the isolated glyceryl-headgroups on anion exchange HPLC with detection by conductivity [69]. Becoming more common now is ultra-high-pressure liquid chromatography coupled to tandem mass spectrometry (HPLC-MS/MS). This still-evolving method can resolve large numbers of lipid species, and permits lipidomics. For PPI lipids, the negative phosphate groups in the sample need to be methylated to make them neutral [70]. The analysis of ionized fragments yields the atomic mass, readily making possible a determination of the number of phosphates, the total chain length of attached fatty acid side chains, and their degree of unsaturation [71–73]. With care for sensitivity, more than a dozen different fatty acid side-chain combinations are readily resolved for each major PPI class. In the pools of PI and PPIs of native mammalian cells the dominant (∼70%) fatty acid composition is 1-stearoyl 2-arachidonoyl (symbolized 38:4 for 38 carbons and 4 double bonds; Figure 1A), but in cultured mammalian cells the arachidonoyl content is lower [71–73]. The fatty acid content is roughly the same in all classes of PPI in a single sample as if they form a dynamic and interconverting metabolic pool. Methods are being developed to distinguish some of the regio-isomers in a PPI family [e.g. separating PI(3,4)P₂ from PI(3,5)P₂] [73], but this is still not yet routine. Mass spectrometry can be calibrated with internal standards to yield the absolute numbers of molecules of each lipid per cell.

Another static method for visualizing distribution and abundance of PPIs is immunofluorescence. Commercial antibodies have been raised against PI(4)P and PI(4,5)P₂. They are used in conjunction with cell fixation and permeabilization protocols tailored to preserve PPI lipids and specific membranes of interest. The cells are stained with primary and secondary antibodies to give fluorescence localization under confocal microscopy [74]. Such results are qualitative, but they have two important advantages over the other static methods: The observations are made on single cells, and they provide subcellular localization at the light-microscope level.

Genetically encoded biosensors are versatile tools that allow dynamic optical, real-time monitoring of PPIs in single cells [75]. They take advantage of the lipid-binding modules that many cytoplasmic proteins have evolved to interact with specific membrane PPIs [76]. Recently, several excellent reviews carefully detail methods used to determine the lipid binding specificity of individual biosensors, the limitations of these methods, and the challenges when interpreting the resulting data [75,77–79]. To illustrate the general utility in using biosensors, we provide the example of a binding domain from phospholipase Cδ1 (PLCδ1). PLCδ1 has a PH domain that selectivity recognizes PI(4,5)P₂. Fusion proteins formed between the 120-residue PH domain sequence and fluorescent proteins serve as genetically expressible fluorescent labels that indicate where PI(4,5) P₂ is in cells [80,81]. In normal cells, the fluorescent diffusible PH-PLCδ1 probes localize to the PM, virtually the only membrane with PI(4,5)P₂ (Figure 2), but if the lipid is depleted by receptor-activated PLC, the probes move quickly into the cytoplasm, having lost their binding partners. They return to the membrane as PI(4,5)P₂ is resynthesized from PI. Probe movements can be monitored as translocation in confocal or total internal reflection microscopy or by changes in fluorescence resonance energy transfer between the probe and another suitable fluorophore. A large repertoire of such lipid-binding modules has been developed as PPI probes for live-cell experiments [79]. Considerations when using such biosensors include how selective they are for a particular PPI species and that some biosensors may not report every membrane containing their target lipid but rather act like coincidence detectors that require other cofactors such as Arf-GTPases to recognize a membrane [76]. Illustrating these two points, PH-PLCδ1 domains can bind IP₃ as well as PI(4,5)P₂, and a more selective choice for monitoring PI(4,5)P₂ might be the Tubby domain, whereas the PH domain from OSH1 requires the coincidence of PI4P and Arf1 for targeting to trans-Golgi network membranes. Despite these caveats and the fact they do not report absolute amounts of lipids, these genetically expressible probes have great advantages of versatility, speed, low cost, and continuous real-time monitoring in living cells. They are readily detected in single cells or cell populations. They do require the use of a fluorescence microscope and expression of proteins in the cells.

Manipulating PPIs

Many good tools exist for manipulating the pools of PPIs during chronic or acute experiments in living cells. Most involve manipulation of enzymes of PPI metabolism: the lipid kinases, the lipid phosphatases, and the phospholipases. These enzymes have been knocked out and knocked down genetically, they have been overexpressed by transfection, they have been activated by receptors or by voltage, they have been blocked pharmacologically, and they have been recruited from cytoplasm to membranes by chemical- or light-mediated dimerization. Although much basic and important PPI biology was discovered through classical genetics and molecular biology in yeast and then mice, we focus here on other cell biological approaches.

Receptor-activated lipid perturbation has been limited to two families of enzymes, PLC (activated by Gq-coupled receptors or growth-factor receptors) and PI-3-kinases (also activated by growth-factor receptors) (see section ‘Physiological roles of phosphoinositides’). Receptor-linked lipid phosphatases seem uncommon. Nevertheless, PI(4,5)P₂-5-phosphatases and PI(4)P-4-phosphatases have been engineered to make several powerful genetically expressed tools. The idea is to attach the enzyme catalytic domain without regulatory or localization sequences to a platform that can be recruited to the PM or to an intracellular membrane. The recruitment needs another genetically expressed partner, the anchor, already attached to the chosen target membrane. The platform with its enzyme is engineered to be dispersed in the cytoplasm where the enzyme does not encounter its substrate. Heterodimerization of the platform with the anchor then is achieved chemically as with a rapamycin dimerization system [82,83] or by light-activated dimerization systems [84,85] translocating the enzyme to the desired target. Translocatable platforms combining two enzymes, e.g. both PI(4,5) P₂-5-phosphatase and PI(4)P-4-phosphatase together [86], or just PI(4)P-5-kinase or a 3-kinase [82] also have been useful. These tools when directed to a known membrane serve to localize subcellular PPI pools and to document the effects of perturbing one pool on another [86–88]. These tools are convenient and easily used. The translocation and the actions of the enzymes on specific PPI pools are completed in 10–20 s. An even faster depletion of PI(4,5)P₂ can be accomplished with voltage-sensing phosphatases (VSPs) that, depending on the animal species of origin, have different ranges of voltage sensitivity [89,90]. The principal enzymatic activity is a lipid 5-phosphatase, but the enzyme also removes 3-phosphates at a lower rate [91,92]. This tool requires patch-clamp electrophysiology to make depolarizing pulses. Since any applied depolarization is limited to the PM, it acts very specifically on PM PPIs. With zebrafish voltage-sensing phosphatase (DrVSP), a plasma membrane depolarization to +110 mV can dephosphorylate the PI(4,5)P₂ pool to PI(4)P in <1 s [93]. The main advantages of the method are speed, and the possibility of spatial and temporal control of the amount of PI (4,5)P₂ depletion.

Several other tools are convenient for PPI manipulation. The simplest is a pharmacological block of lipid kinases or phosphatases. The two most widely used are wortmannin, which blocks lipid 3-kinases at <10 nM and some 4-kinases at >100 nM concentrations [94,95], and phenylarsine oxide, which reacts with vicinal cysteines and blocks 4-kinases as well as tyrosine kinases [81]. Other, more specific inhibitors of PPI-metabolizing enzymes are being developed, including inhibitors against PI4KA [96], PI4KB [97], PIP5KA [98], and PIP5KC [99]. For the class 1 PI3Ks, there is a large and expanding body of data establishing the target selectivity of numerous inhibitors. These efforts were made specifically to improve on older molecules such as wortmannin that have many off-target effects. Illustrative examples of relatively selective drugs include Idelalisib, the class I PI3K delta selective inhibitor approved for B cell malignancies [100], or pan-class I inhibitors such as Pictilisib/GDC0941 [101] or Buparlisib/BKM120 [102] currently in clinical studies for the treatment of patients with advanced solid tumors and breast cancer, respectively. Despite the growing list of compounds that can alter PPI species it would be valuable to develop others. Another approach requires exposing the cytoplasmic face of the PM as is done in inside-out patch clamping. While recording ion channel function, the effective membrane PPI concentration then can be reduced by bathing the cytoplasmic face with PPI-chelating polycations such as polylysine or polyamines or with anti-PPI antibodies. PPIs can be restored by perfusing the membrane patch with water-soluble dioctanoyl-PPI analogs such as diC₈-PI(4,5)P₂. For ion channel studies, the use of diC₈ compounds has been very successful to estimate relative affinities from concentration–response experiments [103,104]. Recently, it has also become possible to apply PPIs to intact cells from the outside as membrane-permeable acetoxymethyl esters and even as membrane-permeable, light-releasable caged PPIs [105].

A tool of quite a different kind is mathematical modeling. Especially, when working with living cells, trying to dissect out and predict the importance of a particular enzyme of lipid metabolism is complicated by the ongoing actions of many other enzymes and by the dynamics of traffic and exchange reactions between organelles. Qualitative discussions often fail to recognize the emergent properties of such background integrated metabolism. Realistic modeling that includes many enzymes greatly helps in interpreting the activity and consequences of stimulating or inhibiting one of them [92,93,106–110].

Physiological roles of phosphoinositides

Many functions for phosphoinositides have been revealed during the last 50 years, in parallel with growing understanding of their chemical structures and metabolism. Virtually all functions fall loosely under the heading of signaling. Here, we emphasize signaling by the phosphoinositides themselves, but first we mention briefly two important examples where a phosphoinositide is the precursor of a signal. One early major discovery was realization in the 1970s and early 1980s that PI(4,5)P₂ is the principal substrate of receptor-activated PLC. This enzyme cleaves PI(4,5)P₂ to yield two potent second messengers, inositol trisphosphate IP₃ and DAG. In this signaling cascade, the IP₃ releases calcium from intracellular stores, and the DAG activates several proteins, notably protein kinase C (PKC) [111,112]. Thus, many hormones and modulatory neurotransmitters acting through Gq-coupled receptors or through some tyrosine-kinase receptors can initiate intracellular calcium signaling and protein phosphorylation by PKC. Virtually every animal cell receives regulatory inputs acting through this important pathway. Another major discovery in the late 1980s was lipid 3-kinases acting on several phosphoinositides that, among other reactions, convert PM PI(4,5)P₂ to PI(3,4,5)P₃. This very low abundance and transient phosphoinositide recruits protein kinases (Akt, BTK, etc.) and downstream signaling cascades to the membrane. Thus, insulin and many growth hormones that signal through tyrosine-kinase receptors to a PPI 3-kinase initiate metabolic changes, cell proliferation, and growth programs. Defects in such signaling may lead to cell transformation and are common in cancers. We refer the reader to reviews on PI 3-kinase signaling [44,113].

Hilgemann hypothesis for intrinsic membrane proteins

As we have seen, organellar membranes and the PM present a phosphoinositide ZIP code unique to that compartment. After discovering the first examples of an ion channel and of a transporter that each required PI(4,5) P₂ for function [114], Donald Hilgemann proposed ‘thus ion channels and transporters that are activated by PIP₂ will probably be inactive during the processing and passage through the secretory pathway’ [115]. Such a membrane protein that is synthesized in the ER, traffics through the Golgi, and is sent in vesicles to the PM would remain silent until it arrives ‘home’ at the PI(4,5)P₂-containing PM, and if it is subsequently recycled by internalization, it would be silenced again. We call this the Hilgemann hypothesis. By generalization, there could be other membrane proteins that need to be silenced at the PM and are active only at another particular compartment. That could be accomplished by requiring a different phosphoinositide for function. Indeed, two ion channels of endolysosomal compartments (TRPML1 and TPC1) require the endolysosomal signature lipid PI(3,5)P₂ and are not activated or are even inhibited by PI(4,5)P₂ [116,117]. Finally, there are ion channels whose PPI preference is less strict; they can be activated by more than one phosphoinositide [118]. This might permit them to perform their functions in several specific organellar membranes. In general, Hilgemann’s hypothesis still needs broader testing, but it seems to us to offer a powerful explanation for why many intrinsic membrane proteins have evolved strong preferences for specific phosphoinositides and a neat cell biological principle for functional compartmentalization of membrane proteins.

Regulation of peripheral membrane proteins

Membranes isolated by cell fractionation include proteins that are loosely bound in addition to the transmembrane proteins and proteins that are firmly anchored by lipid modification. These are the peripheral membrane proteins that may be associated by electrostatic bonds to membrane components. Frequently at least one of their interaction partners is a charged lipid head group, especially PPIs. Peripheral proteins can be released by high salt solutions or by low pH. Polybasic domains of the protein associate reversibly with the acidic phosphates of the head group and dissociate to find new partners as the membrane lipids change during membrane trafficking and signaling. In this way, a long list of cytoplasmic proteins including, for example, PLC, cPLA2, pleckstrin, spectrin, ankyrin, AP-2, mSOS1, Rab35, Rho, Arf6, ARNO, and CYTH, transiently associate with the PM through PH domains and other lipid-interaction modules [76]. The emphasis is that this association is dynamic and regulated in part by the PPI phosphorylation pattern on the membrane. This invites the important question as to how one PPI species might regulate many different effector proteins independantly? One general hypothesis is that local diffusional barriers structurally organize or concentrate specific PPI pools, essentially creating segregated PPI signaling domains (reviewed [119]).

Dynamic regulation of intrinsic membrane proteins by Gq-coupled receptors

Brown and Adams [120] discovered a potassium current of sympathetic neurons that was suppressed by muscarinic agonists and by other GPCR agonists. These agonists were subsequently shown to activate Gq and PLC. The current originally called M-current to reflect its muscarinic regulation was eventually shown to be carried by members of the KCNQ (K_V7) ion channel family. Since these K⁺ channels are partially open at rest, they dampen the electrical excitability of the neurons. Receptor-induced closure leads to heightened electrical activity, behavioral arousal in central circuits, contraction of smooth muscle, and vasoconstriction [121–124]. More than 20 years after the discovery of M-current, closure of the channels was finally found to be due to receptor-mediated depletion of PI(4,5)P₂ [125,126]. Such current suppression by PLC can be mimicked by stimulating a depolarization-activated voltage-sensing lipid phosphatase [93,127] or light-activated optogenetic lipid 5-phosphatase [85], enzymes that deplete PI(4,5)P₂ by dephosphorylation to PI(4)P.

The hypothesis that KCNQ channels require PI(4,5)P₂ to function was initially unexpected. Although cleavage of PI(4,5)P₂ was the well-known enzymatic activity of PLC, it had not been fully appreciated that PI(4,5)P₂ is at such low concentration and is synthesized sufficiently slowly that it could be used up by stimulating endogenous PLC. Subsequently, it was shown by HPLC and by tandem HPLC-mass spectrometry that activating transfected or endogenous Gq-coupled receptors in cell lines and in native cells can deplete nearly the entire cellular pool (∼90%) of PI(4,5)P₂ with a time constant as short as 7–10 s. The depletion was perfectly synchronous with the translocation to the cytoplasm of PLCδ1 PH domains and with the loss of KCNQ channel current [72,128,129]. The total cellular PI(4)P pool fell by more than 70% in tens of seconds as well, reflecting the tight metabolic coupling of cellular PI(4,5)P₂ and PI(4)P pools despite their partial separation in space. In sympathetic neurons and in expression systems, the same receptor-mediated PI(4,5)P₂ depletion also decreased current in two types of Ca_V1 (L-type) and in two types of Ca_V2 (P/Q and N-type) voltage-gated Ca²⁺ channels [130–133]. In these cases, channel gating was modulated, but 40–70% of the original peak current still remained as if PI(4,5)P₂ was regulatory but not absolutely essential for Ca²⁺ channels to open. Such receptor-mediated Ca²⁺ channel modulation would decrease the release of neurotransmitter at synapses. Since the Ca²⁺ currents could also be down-regulated by activating voltage-sensing or light-activated lipid 5-phosphatases that generate PI(4)P, we conclude that for both KCNQ channels and Ca_V channels monophosphorylated PI(4)P is not a good substitute for PI(4,5)P₂.

How general is such rapid depletion of lipid by receptors and consequent regulation of membrane protein function? Only a few studies have shown depletion using direct chemical measurements of PI lipids in native cells [72,128,134], but indirect evidence comes from studies of ion channels. Suppression of KCNQ and Ca_V channel currents occurs in a broad variety of neurons and smooth muscles via a wide range of different endogenous PLC-coupled receptors, including M₁ and M₃ muscarinic, tachykinin, bradykinin, endothelin, vasopressin, 5-HT₂, AT₁ angiotensin II, α₁-adrenergic, H₁ histamine, P2Y purinergic, and GnRH [123,135–140]. When an endogenous PLC-coupled GPCR is present at only low density, it may still evoke sufficient production of IP₃ and DAG for signaling but might not activate PLC strongly enough to deplete PI(4,5)P₂.This is the case for purinergic and bradykinin receptors in sympathetic ganglia [141,142] and endogenous purinergic receptors in tsA cell lines [108,109]. Nevertheless, many native cells are equipped with a high enough density of PLC-coupled GPCRs and of PLC to experience quick and extensive receptor-mediated depletion of PI(4,5)P₂.

PI(4,5)P₂-sensitive intrinsic membrane proteins

What membrane proteins are sensitive to PI(4,5)P₂? The first two to be recognized were a K_ATP channel and a Na⁺–Ca²⁺ exchanger [114], two multi-pass transmembrane proteins that move ions across the membrane. After two more decades, 80 ion channels and 6 transporters have known PI(4,5)P₂ requirements [143–145]. For channels, today’s list includes nearly every TRP channel, the five KCNQ channels, four Ca_V channels, all of the GIRK channels, many other inward rectifier K⁺ channels, and numerous others. For transporters, the list includes notably the dopamine and the serotonin transporters [146].

Three questions come immediately to mind that we now proceed to answer: (i) Why are so many channels and transporters sensitive to PPIs? (ii) Why not other kinds of proteins? (iii) How do these proteins interact with the lipids and how does this regulate their activities? The first two questions might be easy to speculate about. We suggest that the list of channels and transporters is so long because all of them are regulated by the PI ZIP codes they encounter during membrane traffic — the Hilgemann hypothesis. This would prevent them from leaking or transporting ions across the wrong compartment membranes, reserving that job for other channels and transporters functionally adapted for that compartment. Alternatively, perhaps the list is long to allow many channels and transporters to be regulated co-ordinately by PLC-coupled receptors. Such a model becomes more reasonable when one considers that not every cell has the full gamut of PI(4,5)P₂-sensitive channels and transporters, thus precise spatial distribution of select PLC-coupled receptors, channels, and transporters could allow local changes in electrical activity in specialized cells like neurons. The second question asks why the list contains few transmembrane enzymes, receptors, and other proteins. We suggest here that these proteins present a greater challenge for the experimentalist to study, and quite probably with time many receptors and intrinsic enzymes will appear on the list. Supporting such a statement, recent evidence suggests G-protein-coupled receptors interact with PPIs to stabilize their active state and enhance selectivity of G-protein coupling [147], while β-arrestin binding to PPIs seems required for its trafficking to clathrin-coated pits [148]. Channels have been relatively easy to study because the patch-clamp assay gives an instantaneous before–during–after readout as the lipids are manipulated in living cells by the many tools available. These experiments take only some seconds to perform. For other classes of membrane proteins, we lack rapid read-outs in living cells. An alternative explanation of the paucity of other proteins on the list might be that cellular physiology is not profoundly perturbed if these proteins continue to be functional as they traffic among different compartments or perhaps they are shut off adequately by other factors like low vesicular pH.

We turn now to the mechanism of these interactions. Channels differ in their apparent lipid affinity, in their PPI lipid specificity, and in the effect of the lipid on them. The KCNQ channel exemplifies low-affinity binding, which means that in normal cells the KCNQ channels are not fully saturated with lipid and respond readily to decreases in PI(4,5)P₂. The lifetime of the KCNQ channel-lipid complex is < 0.5 s, perhaps much less [93]. The effects of lipid can be described as a large increase in channel open probability, and, with a very low level of the lipid, the channel may be shut off entirely. Ca_V channels also seem to form low-affinity complexes with PI(4,5)P₂. The lifetime of the complex is <100 ms, but standard means of depletion of PI(4,5)P₂ still leave 40–70% of the current [133]. At another extreme, Kir2.1 inwardly rectifying K⁺ channels are nearly saturated with PI(4,5)P₂. They form a high-affinity complex with the lipid that requires extensive lipid depletion to dissociate and might not be much affected by typical physiological receptor-mediated PI(4,5)P₂ reductions [103,149,150]. KCNQ channels are PI(4,5)P₂ preferring but can be activated by other PPI lipids and related compounds (not IP₃) presented to an excised membrane patch [104]. Even more promiscuously, GIRK1/ GIRK4 channels need, but do not distinguish between, PI(3,4)P₂, PI(3,5)P₂, PI(4,5)P₂, or PI(3,4,5)P₃ [103].

Typical neuronal KCNQ channels are heterotetramers of KCNQ2 and KCNQ3 subunits. In single-channel recordings, careful titration with increasing concentrations of dioctanoyl PI(4,5)P₂ reveals a two-component activation curve for the KCNQ2/KCNQ3 heteromers. The higher affinity component attributed to the KCNQ3 subunits has a midpoint at 1.3 µM and the lower affinity attributed to KCNQ2 subunits, at 76 µM [104]. These apparent affinities correspond well with the affinities seen with the two types of homomeric channels. Such experiments show that the same channel tetramer can bind PI(4,5)P₂ at more than one site. A simple presumption is that each of the four protein subunits binds one lipid molecule, and the functional results imply that a channel can open when only two sites are occupied, although with a lower open probability than with four.

Putative binding sites for PI(4,5)P₂ on many ion channels have been identified by a combination of mutagenesis, molecular modeling, and docking [151]. Invariably they contain numerous arginine and lysine residues as basic counterions for the five negative charges on the lipid. More definitively, three cryo-EM structures at atomic resolution show PI(4,5)P₂ bound to K⁺ channels: a Kir2.2 channel [152] (Figure 3A), a GIRK2 channel [153] (Figure 3B), a K_ATP channel complex [154], and another pair of structures compare an apo-TPC1 channel with a PI(3,5)P₂-bound TPC1 channel [155] (Figure 3C). Each tetrameric channel has four bound lipid molecules and the dimeric TPC1 channel has only two. The lipids are oriented with respect to the membrane as one might expect for phospholipids in a membrane with lipid tails extending away from the cytoplasm and the head group lying at the level of the membrane–cytoplasm interface. However, the lipid is mostly encased by hydrophobic regions (blue area, ‘surface charge’, Figure 3) of the pore-forming protein, and the head group phosphates are interacting in a cup with arginines, lysines, a histidine, glutamine, and asparagine of the cytoplasmic domain of the channel protein. Particularly in the Kir2.2 structure (Figure 3A), one might say that the encased lipid acts like a rivet securing the cytoplasmic domain to the membrane domain of the channel in a functional configuration [152]. In each example, the basic residues of the binding site come from well-separated sequences along the main carbon chain that are folded to bring them together. There is no consensus fold such as is found in PH domains. We can expect further direct structure determination for many other channels in the future.

Figure 3. Structural basis of PIP₂ interactions with several ion channels.

In each case, the lipid is the diC₈ version of a PIP₂. (A) Top view: schematic representation of Kir2.2 subunits before and after PI(4,5)P₂ binding. Side view: X-ray crystal structures of the Kir2.2 homotetrameric channel in presence of PI(4,5)P₂ (gray, PDB: 3SPI [152]). Each channel subunit is in complex with a single PI(4,5)P₂ molecule represented as spheres and colored according to atom type: carbon, gray; phosphorous, orange; and oxygen, red. Surface charge: electrostatic map of Kir2.2 homotetrameric channel structure (blue is positive; red negative; white neutral). The PI(4,5)P₂ binding site comprises numerous basic residues (blue) that interact electrostatically with the negatively charged phosphates of PI(4,5)P₂. Zoom: PI(4,5)P₂ binding site. Note that positively charged residues in the protein structure are closely apposed to and predicted to interact with the negatively charged phosphate groups at the D4 and D5 positions of PI(4,5)P₂. (B) Top view: schematic representation of Kir3.2 (GIRK2) subunits before and after PI(4,5)P₂ binding. Side view: X-ray crystal structure of the Kir3.2 homotetrameric channel with accompanying Gβγ subunits (gray, PDB: 3SYA [153]). Surface charge: electrostatic map of Kir3.2 homotetrameric channel structure. Zoom: PI (4,5)P₂ binding site. (C) Top view: schematic representation of TPC1 subunits before and after PI(3,5)P₂ binding. Side view: cryo-EM structure of the endolysosomal TPC1 homodimeric channel in the PI(3,5)P₂-bound state (gray, PDB: 6C9A [155]). Note that similar to Kir2.1 and Kir3.2, one PI(3,5)P₂ molecule binds to each TPC1 subunit. Surface charge: electrostatic map of TPC1 homodimeric channel structure. Zoom: PI(3,5)P₂ binding site.

Other physiological roles for the PPIs

We now briefly mention other important roles of PPIs in exo-/endocytosis, tethering, fusion, cell migration, cell division, and autophagy. During calcium-regulated synaptic vesicle release, PI(4,5)P₂ is required to attract the C2B domain of vesicular synaptotagmin to the PM active-zone for docking and fusion [156], then to switch on the clathrin adapter protein AP2 to form clathrin-coated pits [157], and later to attract dynamin for pinching off the endocytosing vesicle and synaptojanin for dephosphorylating the PPI for its next roles and releasing AP-2 [158]. In cell migration, the chemoattractant sets up a PPI polarity by activating PI3K at the leading edge to produce PI(3,4,5)P₃ and by localizing the 3-phosphatase PTEN to the trailing edge favoring PI (4,5)P₂ in the rear [159,160]. A similar scheme might apply to the leading edge of the myelinating Schwann cell as it forms myelin by advancing spirally to wrap around an axon [161]. In cell division, PPIs are implicated in spindle orientation, mitotic cell shape, and bridge stability (see [162] for review), with PI(3,4,5)P₃ levels during pro- and metaphase important for spindle positioning, PI(4,5)P₂ accumulation during telophase essential for coordinating assembly of the contractile actomyosin ring [163], and PI(3)P participating in abscission during cytokinesis [164]. For autophagy, PI(3)P is the major PPI controlling cellular homeostasis. During the initial phases of autophagosome formation, PI(3)P is produced by the VPS34 kinase complex [165,166]. By recruiting different proteins having PI(3)P-specific binding domains, such as FYVE domains (DFCP1, PIKFYVE), PX domains (Phospholipase D1), or WD40 repeats (WIPI1), PI(3)P promotes autophagosome nucleation and expansion, as well as cargo recruitment and autophagosome maturation.

Defective phosphoinositide metabolism in disease

The spatial and temporal distribution of PPIs is established through the complex activities of ∼50 PPI-metabolizing enzymes (34 phosphatases and 20 kinases) [167]. The dynamic equilibrium of their combined activities results in the PI ZIP code map presented in Figure 1C. Adding to this complexity, a recent proteomics study identified over 400 PPI-interacting proteins that serve to influence membrane identity and cellular function [168]. Given the sheer number of proteins involved in the PPI-interactome, it is not surprising that mutations in the enzymes in Figure 1C result in a wide range of devastating and heterogeneous disorders. Here, we focus on a few, more common monogenetic disorders; for a deeper review of more than 20 different diseases, see Staiano et al. [167].

Oculocerebrorenal syndrome of Lowe/Lowe’s syndrome/OCRL

This X-linked (almost exclusively found in males) recessive disorder primarily affects the brain, eyes, and kidneys. Patients typically present clinically first with congenital cataracts and other visual impairments, followed by cognitive and behavorial impairments. Behavioral problems and seizures have also been reported in children with this rare (∼1 : 500 000 frequency) condition. Genetically, this disease is caused by a loss of function mutation in the OCRL gene, which encodes a PI(4,5)P₂ 5-phosphatase. In health, the enzyme is predominantly localized to endocytic compartments, Golgi, and PM where its role is to dephosphorylate PI(4,5)P₂ to produce PI(4)P. Intuitively, loss of OCRL catalytic activity leads to accumulation of PI(4,5)P₂ on endosomal membranes, which is subsequently responsible for cytoskeletal abnormalities [169]. Recently, some patients diagnosed with proximal tubule dysfunction in another X-linked genetic disorder called Dent disease were also found to have mutations in the OCRL gene. More commomly, Dent disease results from mutations in a gene that encodes a chloride/proton exchanger, thus patients presenting with OCRL mutations have been defined as suffering from Dent-2 disease.

Charcot–Marie–Tooth neuropathy

Charcot–Marie–Tooth disease (CMT) is one of the most common inherited neurological disorders, affecting ∼1 in 2500 people in the U.S.A. CMT is a hereditary motor and sensory neuropathy that affects peripheral nerves. Although levels of many different proteins are abnormal in different forms of CMT disease, all the mutations affect the normal structure and function of either the peripheral nerve axon or the myelin sheath. CMT affects both motor and sensory nerves and can be caused by mutations in PPI phosphatases (myotubularin family members, FIG4, and synaptojanin1, SYNJ1) and PPI kinases (PIP5Ks).

Ciliopathies

Joubert and MORM syndrome are ciliopathies that share a dysfunction of primary cilia. Joubert syndrome is a rare autosomal recessive disease whose diagnostic hallmark is a unique cerebellar and brainstem malformation and cerebellar ataxia. It can be caused by mutations in 21 different genes [170]. One of these is the catalytic domain of INPP5E, a 5-phosphatase that dephosphorylates PI(3,4,5)P₃ and PI(4,5)P₂ into PI(3,4)P₂ and PI(4) P, respectively, aiding in the development and stability of the primary cilium [171]. Similar to Joubert syndrome, the related MORM ciliopathy also results from a defect in INPP5E function. Interestingly, the mutation in this disorder results in a truncated protein rather than a catalytically inactive one; it retains PPI phosphatase activity but loses the ability to be correctly targeted to the ciliary axoneme [172].

Alzheimer’s and Parkinson’s

This short list does not underscore sufficiently the importance of PPIs for human health. As noted by Balla ‘with some efforts every human disease can be linked to altered inositol lipid metabolism’ [173]. Indeed, PPIs are responsible not only for the genetic disorders (some rare) mentioned above, but also are implicated in more common disorders such as cancer, obesity, diabetes, Alzheimer’s, and Parkinson’s. In the case of Alzheimer’s disease (AD), there is much evidence linking alterations in PPI levels to disease progression. This includes numerous reports of alterations in PPI abundance and metabolism in diseased brains [174–178] and the correlation between cellular PI(4,5)P₂ levels and the abundance of 42-residue amyloid β (Aβ) peptide in familial AD-associated presenilin mutations [179]. Furthermore, binding of Aβ to critical PPI-metabolizing enzymes, such as synaptojanin [176] and PI4K2A [180], significantly alters catalytic activities. Finally, highlighting the potential role for PPIs in controlling amyloid protein processing, genetic polymorphisms or mutations in genes encoding PPI-metabolizing enzymes such as INPP5D and SYNJ1 are risk factors for late-onset AD [181,182].

For Parkinson’s disease, a homozygous missense R258Q mutation within the Sac domain of synaptojanin 1 has been identified in patients with an autosomal recessive early-onset Parkinson’s. This mutation is designated PARK20. A murine model carrying this mutation develops neurological manifestations similar to those of human patients, with synapses of these mice displaying endocytic defects and accumulation of clathrin-coated intermediates, reinforcing the Sac domain’s importance in endocytosis and establishing a link between dysfunction in early endocytic traffic and Parkinson’s disease [183].

Future outlook and concluding remarks

This article has discussed some of the broader issues relating to PPI metabolism in health and disease. It is clear there is significant information still to determine. The rapidly expanding toolkit, now including super-resolution imaging modalities and mass-spectrometry, to interrogate PPI metabolism with nanoscale and quantitative precision promises greater understanding of how PPIs can simultaneously regulate many biological processes. Concerning PPI manipulation, gene editing (CRISPR/Cas9) promises to have a revolutionary impact on the field. For example, the generation of endogenously tagged proteins at physiological levels offers the unique opportunity to identify the preferred patterning and dynamics of PPI-enzymes. Some of the broader questions that will hopefully inspire future research endeavors include: How can we visualize these low-abundance lipids directly, without the need for binding-domains or effectors? What are the spatial and temporal characteristics of PI metabolism? For MCSs, is there heterogeneity in terms of proteins or transfer activities at the same membrane-membrane interface? If so, what governs such signaling diversity? Perhaps most importantly, how can we leverage our current knowledge of PPIs and the extensive coverage of the cellular PPI-interactome to influence the onset or progression of human disease? This latter question may be farthest from our grasp in part because of the relative lack of information on PPI metabolism in native primary cells compared with expression systems, such as HEK cells, and the paucity of specific pharmacological reagents that interfere with the generation and destruction of these lipids. Despite this long list of information still to determine, the past 80 years of PPI research, especially the explosion within the last twenty years impressively underscores how essential these minor lipids are for the fundamental regulation of all aspects of membrane dynamics in eukaryotic cells.

Abbreviations

AD

Alzheimer’s disease

Ca_V

voltage-gated Ca²⁺ channel

CMT

Charcot–Marie–Tooth

DAG

diacylglycerol

diC₈-PIP₂

a dioctanoyl version of the lipid

ER

endoplasmic reticulum

E-Syt2

extended synaptotagmin 2

Kir

inwardly rectifying K⁺ channel

K_V

voltage-gated K⁺ channel

MCS

membrane contact site

ORD

OSBP-related domain

ORP

OSBP-related protein

OSBP

oxysterol-binding protein

PH

pleckstrin homology

PI

phosphatidylinositol

PKC

protein kinase C

PLC

phospholipase C

PM

plasma membrane

PPI

polyphosphoinositide