Morton Medal Lecture

The ubiquitous phosphoinositides

P.J. Parker


There are now known to exist seven phosphoinositides all derived through various metabolic routes from the parent lipid phosphatidylinositol. With one additional metabolite, diacylglycerol, these represent a rich resource of bioactive lipids responsible for recruiting protein effectors and marking membrane compartments. The metabolic map of this pathway and the nature of the binding partner interactions are reviewed.

  • phosphoinositide
  • protein domain
  • second messenger

Phosphoinositide (PI) metabolism

Since the pioneering work of Lowell and Mabel Hokin in the 1950s (see [1]), the inositol lipids increasingly have held the attention of researchers unravelling the complexities of cellular responses to internal and external cues. The ‘cornerstone’ that cemented the foundations of this seemingly ever-expanding field was the elucidation of proximal targets of hormonally induced PtdIns(4,5)P2 (phosphatidylinositol 4,5-bisphosphate) metabolism, i.e. the definition of DAG (diacylglycerol) and Ins(1,4,5)P3 (inositol 1,4,5-trisphosphate) as second messengers (Figure 1). In the case of DAG, Yatsutomi Nishizuka and colleagues [2] determined that the then-newly-discovered kinase activity, coined ‘protein kinase C’ [PKC; following alphabetically after protein kinase A (cAMP-dependent protein kinase) and the then protein kinase B (phosphorylase kinase); it was certainly not appreciated then that this alphabet would not cope with the protein kinase superfamily], was reversibly activated by DAG presented in the context of acidic phospholipids and Ca2+, and that this could be recapitulated in intact cells through the use of membrane-permeant short-chain DAGs (e.g. [3]). Furthermore, it was found that tumour promoters of the phorbol ester class were also able to activate PKC, mimicking the effect of DAG in vitro [4]. In parallel with these developments, Michael Berridge and colleagues [5] demonstrated that Ins(1,4,5)P3 triggered the release of intracellular Ca2+ from cellular stores in a selective manner, causing the characteristic elevation of cytosolic Ca2+. The excitement engendered by these seminal contributions contributed enormously to the still-rolling juggernaut of PI research.

Figure 1 The classical PI second messenger pathway

The synthesis and hydrolysis of PtdIns(4,5)P2 is illustrated, alongside the two proximal targets initially identified for the two second messenger products.

In the context of this classical pathway of PI-based signalling, the last 20 years or so has led to the identification of multiple enzymes involved in the synthesis of PtdIns(4,5)P2, including the identification of PtdIns5P and the non-classical PtdIns5P→PtdIns(4,5)P2 pathway [6]. Similarly, multiple classes of PI phosphatases have been identified, some with multiple functions (see below), as well as families of PI-specific phospholipases C (PI-PLC) responsible for phosphodiesterase cleavage of PI (where not phosphorylated at the 3-hydroxy position); in particular, PtdIns(4,5)P2. Linked in with these studies have come insights into the layers of control coupling various receptor classes to these metabolic processes. The PI-PLCβ class operates downstream of Gαq and Gβγ families, whilst the PI-PLCγ family are targeted to tyrosine kinase pathways through their SH2 domains (see [7], and references therein).

A key addition to the metabolism of the PIs came with the insight that the Type I PI-kinase (operationally defined, based upon elution from anion-exchange columns) does not catalyse the expected conversion of PtdIns into PtdIns4P, but, rather, its conversion into PtdIns3P [8]. This triggered a frenzy of activity in what has become known as the P3-kinase pathway, an area of research enormously facilitated by the availability of two inhibitors of this class of enzymes, namely LY294002 [9] and wortmannin [10]. Cloning, expression and characterization of the four classes of mammalian PI 3-kinases (PI3Ks) also went hand in hand with the development of insights into their regulation; in particular, through tyrosine-kinase-linked receptors and also through heterotrimeric G-protein-linked receptors (see [11]).

The last piece of the PI jigsaw, as we currently picture it, came with the identification of PtdIns(3,5)P2 [12]. The enzyme responsible for its synthesis in mammals was identified as p235/PIKfyve [13]; unusually for this complex pattern of interconverted lipids, there appears to be only one gene responsible for PtdIns(3,5)P2 production in mammals [14]. The addition of this lipid to the fold brings the total number of PIs to seven, and with DAG these make up a collection of eight metabolically inter-related ‘bioactive’ lipids (Figure 2). Whereas DAG itself can be metabolized further to phosphatidic acid by DAG kinases (reviewed in [15]) as well as by lipases, this commentary will be restricted to the PIs and the immediate metabolite DAG.

Figure 2 The metabolic relationships of known PIs

All the different phosphorylated species identified to date are shown. The three different classes of PI3K and the three PIP-kinase (PIPKin) classes are also shown, indicating where they act.

Lipid effectors

For all the PIs and also for DAG there are specific binding proteins. These interactions are conferred by modular domains, which are found in multiple proteins and protein families (Figure 3). It is not intended to cover these domains in a comprehensive manner, but rather to illustrate effectors for each domain class.

Figure 3 The PI-binding domains

Associated with the metabolic map of interconverted PIs, binding domains that display specificity for distinct PI species are indicated. For PtdIns5P/WD40, the ‘?’ implies an unknown functional significance.

DAG-binding properties are conferred by many, but not all, C1 domains (see [16]). This innocuous nomenclature derives from the first conserved domain identified in the linear sequences of the classical PKC family (α, βI, βII and γ) [17]. Notably for these proteins, the C1 domains are repeated such that the current nomenclature refers to C1A and C1B domains [18]. Both contribute to varying extents in different family members to DAG binding, and indeed to binding phorbol esters and related tumour-promoting compounds. It is important to note that not all C1 domains bind DAG. For example, the single C1 domain present in PKCζ does not bind DAG, although the structural basis for this remains elusive. Structural analysis of an individual C1 domain reveals a structure centred upon the co-ordination of two Zn2+ atoms by the conserved HXnCX2CX13CX2CXn,HXn,C motif (where ‘X’ represents any amino acid) [19]. The critical element in the operation of this domain is the formation of a contiguous hydrophobic surface once ligand has bound. This leads to the prediction (consistent with behaviour) that binding occurs at the membrane, and that the lipid effector is held there through these strong hydrophobic interactions. In the case of PKC, it is predicted that conformational change associated with this membrane interaction relieves the autoinhibition conferred by a basic pseudosubstrate site [20]; this is compounded further by the interaction of this inhibitory, basic sequence with the proximal acidic membrane lipids, effectively sequestering the autoinhibitory sequence [21].

Many PIs are bound by PH (pleckstrin homology) domains (see [22]). These domains comprise two opposed β-sheets capped at one end by an α-helix. This structure is related to those of the PTB (phosphotyrosine-binding) and PDZ domains, but with distinct backbone topologies and binding behaviour. For the PH domains, PI interaction occurs in a pocket formed at the face opposite the α-helix. The structures of a number of PH domains have been solved, including some with soluble inositol phosphates bound {e.g. spectrin PH–inositol(4,5)P2 [24]}. (Note that although inositol phosphates have been employed experimentally as surrogate PI head groups, at appropriate concentrations these may also influence specific PH-domain-containing proteins by displacing them from their PI binding partner [23]).

Structural solutions have provided consensus motifs for interactions with particular phosphate groups in the inositol lipid head-group. Such analyses have also provided a rationale for the sometimes non-exclusive pattern of binding specificity. For example, the PKB/Akt PH domain empirically was found to bind both PtdIns(3,4,5)P3 and PtdIns(3,4)P2; the structural analysis of this binding behaviour reveals that the 5-phosphate sits in an aqueous environment not influencing (positively or negatively) binding [25]. Functionally, PI-binding PH domains, as do other lipid-binding domains, confer membrane recruitment potential to proteins. Whilst this contribution to compartmentation is a key element in their function, it is likely that some lipid interactions also confer conformational change on the recruited protein. The PH domain of PKB/Akt undergoes a conformational change on interaction with PIs [25]. By contrast, the PH domain of PtdIns–PLCδ1 seems to behave as an enzyme-linked tether, conferring scooting activity on the phospholipase [26], albeit with perhaps other associated properties [27].

The FYVE domain derives its name from Fab1p, YOTB, Vac1 and EEA1 (early endosome antigen 1) proteins, which all have one such domain [28,29]. Like the C1 domains, the FYVE domain is stabilized by the presence of two Zn2+ atoms co-ordinated in a cross-braced fashion [30] by a highly conserved group of eight cysteine residues (or, in some instances, seven cysteine residues plus one histidine). Many FYVE-domain-containing proteins are known to be involved in decision processes associated with membrane traffic, obvious examples of which include genes identified in genetic screens for defective traffic or those shown retrospectively to induce such defects on disruption (e.g. the Saccharomyces cerevisiae proteins Fab1p, Vps27p and Vac1p; see [31]). These compartment-specific processes are of course served well by appropriate lipid-binding devices. Crystallographic analysis of the ligand-bound EEA1 FYVE domain indicates an interaction amplified by dimerization and probably stabilized through multivalent interactions [32].

The ENTH (epsin N-terminal homology) domain has been shown to bind either PtdIns(4,5)P2 or PtdIns(3,5)P2. For both epsin and AP180, the ENTH domain displays a preference for PtdIns(4,5)P2 over other PIs; structures have been derived for domain–head-group interactions [33,34]. The predicted binding pocket comprises three distinct regions of the domain, and binding is associated with a substantial change in conformation of site 1; this unstructured N-terminal region appears to become ordered [34]. As one might predict for these two proteins, the ENTH domains confer PtdIns(4,5)P2-dependence on endocytosis, as evidenced by studies combining structural and mutational analyses (reviewed in [35]). By contrast, the ENTH domain of ent3p binds PtdIns(3,5)P2 and is required for entry into the lumen of the multivesicular body [36].

PX domains derive their name from the phagocyte oxidase (PhoX) complex required for neutrophil killing of phagocytosed pathogens and associated with the human disease chronic granulomatous disease [37]. Two components of this multimeric complex retain PX domains (p40phox and p47phox) involved in binding PtdIns3P and PtdIns(3,4)P2 (see [38]). The SNX (sorting nexin) family [39] of proteins also retain PX domains that are generally selective for PtdIns3P. These proteins are involved in endosomal sorting processes, probably as part of larger complexes, as evidenced by the yeast SNX protein Vps5p, which is a component of the retromer complex directing endosome-to-Golgi traffic (see [40]).

Genetic screening for PtdIns(3,5)P2 effectors has identified a class of WD40 domain proteins as PI-binding proteins. The original protein identified in yeast, Svp1p (also identified as AUT10/ATG18/CVT18), binds PtdIns(3,5)P2 with high affinity (low nM) and high specificity [41]. By contrast, the related mammalian WIPI-49 protein binds PtdIns3P, PtdIns(3,5)P2 and PtdIns5P [42]. In yeast, genetic analysis demonstrates a role for Svp1p in membrane recycling from the vacuole [41]. In Homo sapiens, there are at least four gene family members, and to date functional analysis of only WIPI-49 has been performed, revealing its association with the mannose-6-phosphate receptor recycling pathway [42].

The FERM domain earns its name from its presence in the 4.1 (, ezrin, radixin and moesin family of proteins that are responsible for the dynamic linking of actin to membranes (reviewed in [43]). The FERM domain is conserved at the N-terminus of these proteins and binds PtdIns(4,5)P2. Binding has a critical role in releasing N-terminal/C-terminal interactions, releasing the latter to interact with actin and the former to bind additional protein partners [44].

In addition to these well-defined PI-binding domains, there are a number of proteins that will interact with PIs through short amino acid motifs. These typically comprise polybasic regions, and interaction can have profound effects upon conformation, as exemplified by the structural ‘opening’ of the focal-complex-associated protein vinculin [45]. In the yeast protein ETF1, involved in the cytoplasm-to-vacuole sorting pathway, the short basic motif KKPAKK confers binding to PtdIns3P [46]. It is anticipated that this interaction determines the PI3K(vps34) involvement in this pathway.

Combinatorial inputs

Despite the ability of proteins to interact with PIs, it is evident that the prolonged association of proteins with particular membrane compartments is effected via two or more membrane contacts. This might involve multiple lipid interactions and/or protein-plus-lipid interactions. It is also the case that specificity is exerted through such combinatorial interactions. Examples of these mechanisms and strategies are briefly discussed here.

The classical PKC isotypes (cPKC α, β and γ) provide a well-established paradigm for multiple lipid interactions. These proteins are allosterically activated by DAG via their C1 domains. However, their ability to ‘survey’ membrane DAG content is determined (at least in part) by cytosolic Ca2+, which will promote acidic phospholipid interaction of cPKCs via their C2 domains (see [47]). The evidence is consistent with a C2-domain-dependent conformational change that is then permissive for DAG–C1-domain binding. Irrespective of these topological details, it is clearly the case that the combined C1 and C2 domain lipid interactions can reversibly anchor cPKC to membranes. In the case of cPKCs there is an additional lipid interaction afforded by the binding of the released polybasic pseudosubstrate site to acidic phospholipids [48]. Such interactions, though non-selective, are reminiscent of the polybasic motifs implicated in binding PIs (see above). Therefore the overall regulated behaviour of these proteins reflects multiple membrane interactions driven by recognition of lipids with distinctive degrees of specificity and with the combinatorial element of these interactions contributing to the specificity of the compartmentalization.

There are upwards of 35 FYVE-domain-containing proteins in the database (see, of which a number have been shown to bind PtdIns3P, as noted above. However, these proteins show distinct, albeit sometimes overlapping, patterns of compartmentation. Thus, in many cases, other interactions work in combination with PtdIns3P binding to determine localization (this is not obligatory, since for the SMAD-anchor protein SARA, the FYVE domain is sufficient for localization [49]). For the FYVE-domain-containing proteins EEA1 and Rabaptin4, localization is determined by PI binding in combination with interactions with members of the small GTPase Rab family; EEA1 binds GTP-Rab5 [32], whereas Rabaptin binds GTP-Rab4 [50]. These Rab proteins are themselves membrane-anchored through isoprenylation at their C-termini, and themselves define compartments involved in endocytic traffic (Rab5) and local recycling from endosomes (Rab4). Thus these FYVE-domain proteins confer specific PI3K-dependent events on the distinct Rab5 and Rab4 compartments (discussed in [31]).


The general acceptance that all PIs are ‘bioactive’, in the sense that each has the ability to facilitate the recruitment of one or more classes of proteins to membranes, creates quite a metabolic problem for cells. Twenty years ago, the route to production of PtdIns(4,5)P2 in cells was the sequential generation of PtdIns4P, then PtdIns(4,5)P2, via the action of appropriate lipid kinases. However, it transpires that cells can in effect arrive at PtdIns(4,5)P2 by three distinct mechanisms (Figure 2). This flexibility probably provides opportunities to locally generate PtdIns(4,5)P2 without undesired side products. Catabolically, eukaryotes have also evolved some interesting phosphatases that act to convert PIs. In particular, synaptojanin (1 and 2) both comprise a 5-phosphatase domain juxtaposed to a Sac domain (3- and 4-phosphatase activity; reviewed in [51]). These proteins thus have the capacity to convert PtdIns(4,5)P2 into PtdIns by sequential dephosphorylation, and hence the potential to insulate the cell from what might be the undesired accumulation of phosphorylated intermediates. Other inositol lipid 5-phosphatases (see [51]) do not contain Sac domains, and as such can be exploited by the cell to generate specific intermediates and hence distinctly ‘mark’ an evolving membrane compartment or compartment subdomain.

These considerations of PIs and, in particular, their metabolism make it increasingly important that we develop tools to be able to analyse the presence of distinct lipid species in particular compartments, as well as their dynamic changes. Mass spectrometry (MS) of cellular populations promises much (e.g. [52]) and the ability to select specific compartments when combined with MS is likely to prove powerful. Similarly, genetically encoded devices for the detection of specific lipids have been instructive (recently reviewed in [53]), and approaches to the induction of local alterations in membrane lipids will be invaluable also. In compiling a precise understanding of the role of PIs, the challenge will be to manipulate and monitor the distribution of multiple PIs within cells, at high resolution and over significant periods of time.

Morton Medal Lecture Delivered at the SECC, Glasgow, on 19 July 2004 Peter Parker

Abbreviations: DAG, diacylglycerol; EEA1, early endosome antigen 1; ENTH, epsin N-terminal homology; PH, pleckstrin homology; PK(A/B/C), protein kinases A, B and C respectively; cPKC, classical PKC; PI, phosphoinositide; PI3K, PI 3-kinase; PI-PLC, PI-specific phospholipases C


View Abstract