Annual Review of Biochemistry

Vol. 67:481-507 (Volume publication date July 1998)


Phosphatidylinositol, a component of eukaryotic cell membranes, is unique among phospholipids in that its head group can be phosphorylated at multiple free hydroxyls. Several phosphorylated derivatives of phosphatidylinositol, collectively termed phosphoinositides, have been identified in eukaryotic cells from yeast to mammals. Phosphoinositides are involved in the regulation of diverse cellular processes, including proliferation, survival, cytoskeletal organization, vesicle trafficking, glucose transport, and platelet function. The enzymes that phosphorylate phosphatidylinositol and its derivatives are termed phosphoinositide kinases. Recent advances have challenged previous hypotheses about the substrate selectivity of different phosphoinositide kinase families. Here we re-examine the pathways of phosphoinositide synthesis and the enzymes involved.


Although phosphatidylinositol (PtdIns) represents only a small percentage of total cellular phospholipids, it plays a crucial role in signal transduction as the precursor of several second-messenger molecules. The inositol head group contains five free hydroxyls with the potential to become phosphorylated (Figure 1a). Thus, numerous derivatives of PtdIns1could exist in cells, each with a unique function. To date, the following phosphoinositides have been identified in cells: PtdIns-3-phosphate (hereafter termed PtdIns-3-P), PtdIns-4-phosphate (PtdIns-4-P), PtdIns-5-phosphate (PtdIns-5-P), PtdIns-3,4-bisphosphate (PtdIns-3,4-P2), PtdIns-3,5-bisphosphate (PtdIns-3,5-P2), PtdIns-4,5-bis- phosphate (PtdIns-4,5-P2), and PtdIns-3,4,5-trisphosphate (PtdIns-3,4,5-P3) (Figure 1b). Three general functions of these lipids could be imagined: (a) to serve as phospholipase substrates for the generation of soluble inositol phosphate (and membrane-associated diacylglycerol) second messengers; (b) to interact directly with intracellular proteins, affecting their location and/or activity; (c) to alter local membrane topology by electrostatic interactions. The first two functions are well established for certain phosphoinositides, and some data provide evidence for the third role. The specific cellular functions of the individual phosphoinositides have recently been reviewed in detail (1, 2, 3, 4). Our intention here is to present our current understanding of the pathways of synthesis of each phosphorylated lipid and how different steps are regulated. Although some phosphoinositides can be generated by lipid-specific phosphatases acting upon more highly phosphorylated forms, the lipid kinases are the focus of this review.

Many phosphoinositide kinases were described initially as enzymatic activities capable of transferring a phosphate to a particular position on the inositol ring of PtdIns or one or more of its phosphorylated derivatives. Studies of these purified enzymes led to the categorization of PtdIns or phosphoinositide kinases into three general families: phosphoinositide 3-kinases (PI3Ks), PtdIns 4-kinases (PtdIns4Ks), and PtdIns-P (PIP) kinases (PIP5Ks). Research in this area has been accelerated by the cloning of genes encoding several of these enzyme classes. Members of each phosphoinositide kinase family have been identified in yeast and other lower eukaryotes; each shares substantial protein-sequence homology with its mammalian counterparts. This evolutionary conservation underscores the importance of these enzymes in the physiology of all eukaryotic cells. Sequence homology generally supports the separate classification of PI3Ks, PtdIns4Ks, and PIP kinases (for reviews of phosphoinositide kinase phylogeny, see References 5, 6, 7). However, studies of the recombinant enzymes have revealed that certain phosphoinositide kinases have different or broader activities than realized previously.

Each section that follows begins with a review of the molecular biology of a particular phosphoinositide kinase family and concludes with a discussion of the enzymology and regulation of the family members. For some enzymes we suggest new nomenclature that reflects our current understanding of substrate selectivity. We conclude with an overview of phosphoinositide synthesis pathways, using the proposed nomenclature where appropriate. The reader is referred to several recent reviews that detail the cellular functions of the enzymes and their lipid products (1, 2, 3, 4, 8, 9).


PI3Ks have been studied intensively since the discovery of a PI3K activity associated with two viral oncoproteins: polyoma middle T (mT) antigen and pp60v-src (10). Subsequent work has confirmed a role for PI3Ks and their products not only in growth regulation but also in various other cellular responses. Interest in these enzymes has further increased following recent findings that PI3K activation prevents cell death (9), that PI3K is a retrovirus-encoded oncogene (11), and that PI3K mutations increase lifespan in Caenorhabditis elegans (12). We refer to the different PI3K classes using nomenclature proposed recently by Domin & Waterfield (7).

Class I PI3Ks


Class I PI3Ks are heterodimeric proteins, each of which consists of a catalytic subunit of 110–120 kDa and an associated regulatory subunit. Three mammalian PI3Ks sharing 42–58% amino acid sequence identity have been cloned and designated p110α, p110β, and p110δ (13, 14, 15). Each of these proteins contains an N-terminal region that interacts with regulatory subunits, a domain that binds to the small G protein ras, a “PIK domain” homologous to a region found in other phosphoinositide kinases, and a C-terminal catalytic domain (Figure 2). p110-related genes have been cloned from a range of eukaryotes including C. elegans, Drosophila melanogaster, and Dictylostelium discoideum (12, 16, 17). Together these gene products are termed class IA PI3Ks.

The regulatory subunits that associate with class IA PI3Ks are often called p85 proteins, based on the molecular weight of the first two isoforms to be purified (18) and cloned (19, 20, 21): p85α and p85β. p85 proteins do not possess any known enzymatic activity but are composed of several domains with homology to those found in other signaling proteins. These domains are termed modular because they can be separated functionally and spatially from the rest of the protein in which they reside. p85α and p85β contain an N-terminal src-homology 3 (SH3) domain, two or three proline-rich segments, a region of homology to GTPase-activating proteins for the rho family of small G proteins (rho-GAPs), and two src-homology 2 (SH2) domains (Figure 2). The function of each of these modules is discussed below. Between the two SH2 domains is a region that is necessary and sufficient for interaction with the N terminus of p110 catalytic subunits. Termed the inter-SH2 domain, this region contains sequences that are predicted to form α-helices that fold into a coiled-coil structure.

The p85α gene has several splice variants, two of which encode the smaller proteins p50α and p55α (22, 23, 24, 25). These proteins have unique N termini of 6 and 35 amino acids, respectively, and share the C terminus of p85α, including the second proline-rich motif, the SH2 domains, and the inter-SH2 domain (Figure 2). A third gene, p55γ, encodes a protein with similar overall structure to p55α (26). There is no evidence to date that different regulatory subunit isoforms pair preferentially with different p110 isoforms (15). However, some data suggest that different p85 subunits may associate with different subsets of intracellular proteins (27, 28, 29). Drosophila contain a p85-related protein of 60 kDa with conserved SH2 domains and an inter-SH2 domain with probable coiled-coil structure (30).

A protein with 36% identity to p110α was cloned and designated p110γ (31, 32, 33). p110γ contains a PIK domain, a kinase domain, and a ras-binding domain but diverges from class IA PI3Ks at its N terminus (Figure 2) and does not interact with p85 proteins. It has therefore been designated a class IB PI3K. p101, a putative regulatory subunit for p110γ, possesses no recognizable homology to other proteins (33). The regions of interaction between p101 and p110γ have not been mapped.


In vitro, all the class I PI3Ks are able to phosphorylate PtdIns, PtdIns-4-P, or PtdIns-4,5-P2 on the free 3-position (Table 1). Class IA PI3Ks also phosphorylate PtdIns-5-P in vitro (34). However, agonists that stimulate these enzymes in vivo cause increases mainly in the cellular levels of PtdIns-3,4-P2 and PtdIns-3,4,5-P3. In addition, kinetic studies in 32P-labeled cells suggest that PtdIns-3,4-P2 may be produced in part by the action of a 5-phosphatase on PtdIns-3,4,5-P3 (35). Thus, the class I PI3Ks may be selective for PtdIns-4,5-P2 in vivo. Of note, the inter-SH2 domain of p85α binds to PtdIns-4-P and PtdIns-4,5-P2 in vitro (36); this property may provide a mechanism for presenting these substrates to the catalytic subunit or to concentrate the enzyme at membranes rich in these lipids.


Properties of mammalian phosphoinositide kinases

Class I PI3Ks possess intrinsic protein kinase activity that is inseparable from their lipid kinase activity (15, 37, 38, 39). In fact, all phosphoinositide kinases contain several key residues conserved in the catalytic domains of classical protein kinases. The major substrates of the protein kinase activity of a class I PI3K are serine residues within the catalytic subunit itself and/or its associated regulatory subunit. p110α phosphorylates p85α at serine 608; interestingly, this phosphorylation results in down-regulation of the lipid kinase activity of p110α (37, 38). p110δ prefers autophosphorylation to intersubunit phosphorylation, but this autophosphorylation similarly down-regulates enzyme activity (15). p110γ also autophosphorylates, but without demonstrably affecting enzyme activity (39). p85/p110 complexes can phosphorylate insulin receptor substrate-1 (IRS-1) in vitro and possibly in vivo (40, 41); however, other protein substrates for class I PI3Ks in vivo have not been established.

The fungal metabolite wortmannin is a potent inhibitor of the lipid (and protein) kinase activities of class I PI3Ks. The 50% inhibitory concentration (IC50) values for inhibition of the isolated enzymes are all in the range of 1–10 nM (Table 1). Similar concentrations are required to inhibit p85/p110 PI3K in vivo, as judged by effects on the activity of the enzyme immunoprecipitated from cells treated with the drug. Wortmannin irreversibly inhibits p110α by reacting covalently with lysine-802 (42), a residue required for catalytic activity that is conserved in all phosphoinositide kinases (and in protein kinases). A second pharmacological PI3K inhibitor, Ly294002, is a reversible inhibitor of class I enzymes with IC50 values of approximately 1 μM (43). Wortmannin and Ly294002 have been used extensively to study the physiological role of class I PI3Ks in various cellular responses. However, some of these studies should be interpreted with caution owing to the emerging evidence that at somewhat higher concentrations these compounds inhibit other signaling enzymes, including PtdIns4Kβ (44) and the related protein kinases TOR (45) and DNA-PK (46).

Class IA PI3Ks are regulated by interaction with the small G protein ras. Many extracellular stimuli activate ras by increasing the ratio of bound GTP to GDP. In turn, ras-GTP interacts with a number of downstream “effectors.” p110α has been established as a ras effector: Its activity is increased in vitro and in vivo by ras-GTP, dominant negative forms of ras can interfere with 3′-phosphorylated phosphoinositide production, and ras effector domain mutants that fail to interact with PI3K are defective in certain ras-dependent cellular responses (47, 48, 49, 50).

The activities and subcellular locations of Class IA PI3K catalytic subunits are also regulated by p85 proteins. p85 and its relatives are sometimes referred to as adapter subunits because they possess several modular domains with the capacity to interact with other signaling proteins. The SH2 domains of p85 have been studied in detail. Like SH2 domains in other signaling proteins, they bind selectively to phosphotyrosyl (pTyr) residues within specific sequence contexts. In all known p85 proteins, both the N-terminal SH2 (N-SH2) and C-terminal SH2 (C-SH2) domains bind preferentially to polypeptides containing a p-Tyr-X1-X2-Met motif (51). A second methionine or valine at the X1 position increases binding affinity, particularly for the N-SH2 domain (51). A crystal structure of the N-SH2 domain bound to a phosphopeptide explains the binding selectivity (52). A solution structure of the C-SH2 domain has also been determined by nuclear magnetic resonance (53). Synthetic peptides containing tandem pTyr-Met-X-Met (pYMXM) motifs separated by a spacer region bind with high affinity to p85 proteins and, importantly, increase the catalytic activity of the associated p110 subunits two- to threefold in vitro (54, 55).

Many stimuli trigger phosphorylation of YMXM motifs, which recruits p85-p110 complexes and thereby enhances PI3K activity (discussed in detail in 8). Most of the agonists that activate p110 via p85/pYMXM interactions also activate ras, itself a p110 activator, as discussed above. For example, polyoma mT has a YMXM sequence to interact with PI3K adapter proteins and other tyrosine residues that mediate interactions leading to activation of ras. Variant polyoma mT proteins that fail to activate ras are unable to increase production of cellular 3′-phosphorylated phosphoinositides, even if the YMXM motif is intact (56, 57). Many YMXM-containing proteins are membrane associated, as is ras. Therefore, recruitment of p85/p110 complexes not only increases catalytic activity but also brings the PI3K from the cytoplasm to the membrane, where its substrates and a potential activator (ras) reside.

Interestingly, the SH2 domains of p85 proteins also bind to phosphoinositides in vitro and exhibit marked selectivity for PtdIns-3,4,5-P3 (58). Lipid binding to the SH2 domains competes with the binding of pTyr-containing peptide, which suggests that the production of PtdIns-3,4,5-P3 by activated PI3K causes dissociation of the p85/phosphopeptide complex from its pYMXM docking sites. This model is supported by the finding that wortmannin treatment stabilizes p85/pTyr interactions in cells (58). It is also possible that lipid binding regulates enzyme activity allosterically or by influencing its membrane attachment.

SH3 domains are known to interact with proline-rich sequences with the consensus motif ϕ–P-p-ϕ-P, where P is an invariant proline, p is a weakly conserved proline, and ϕ is an aliphatic amino acid (59). This motif forms a left-handed type II polyproline helix that fits into a hydrophobic platform in the SH3 domain (60). Solution structures have been determined for the p85α SH3 domain alone or bound to a high-affinity peptide ligand (61, 62). A GST fusion protein of p85α SH3 selected several candidate partner proteins from bovine brain extracts, including the GTPase dynamin (61). In addition, the p85α-SH3 domain may interact with proline-rich motifs within the p85 protein itself (63). Intramolecular association of these two modules may prevent either of them from finding other, higher-affinity partners in unstimulated cells.

The proline-rich motifs of p85α mediate binding to SH3 domains of src family kinases including src itself, lck, lyn, and fyn (63, 64, 65, 66, 67, 68, 69). The SH3 domain of the cytoplasmic tyrosine kinase abl also associates with p85α (63). For binding to lyn and fyn, the N-terminal proline-rich motif of p85α is more effective than the C-terminal motif (69). The proline-rich regions of p85β diverge considerably in sequence from the analogous regions of p85α. In addition, p85β possesses a third PPXP motif between the C-SH2 domain and the C terminus. Thus, p85β may select a different set of SH3-containing proteins in vivo. The p50α, p55α, and p55γ regulatory subunits contain only a single polyproline motif and lack the N-terminal motif that is selective for binding src.

The binding of protein kinases of the src family via their SH3 domains to the proline-rich motifs of p85 does not depend on any posttranslational modification. This fact suggests that the p85-kinase interaction is regulated by a different mechanism. The crystal structures of src and hck in their inactive states shows that the endogenous SH3 domain makes intramolecular with a cryptic polyproline-like helix formed by the spacer sequence between the SH2 and kinase domains (70, 71). Thus, intermolecular association of src family kinases and p85 proteins may first require release of intramolecular s within each protein. Association of p85α with the phosphoprotein cbl is also consistent with this type of model. Binding of pTyr residues on cbl to the SH2 domains of p85α appears to expose the SH3 domain to allow high-affinity interactions with proline-rich regions of cbl (72).

The rho-GAP homology region—originally termed the breakpoint cluster region (bcr)-homology domain—of p85α lacks GTPase-promoting activity in vitro toward the small G proteins rho, rac, and cdc42. The crystal structure of the p85α rho-GAP domain is similar to bona fide rho-GAPs but lacks five conserved residues likely to be important for catalysis (73, 74, 75). Nevertheless, the p85α rho-GAP domain binds to rac and cdc42 in vitro in a GTP-dependent manner (76, 77). It is possible that rac and/or cdc42 regulates p85/p110 complexes or helps locate these complexes at specific regions of the cell (or vice versa). Because the rho-GAP homology region of p85α lies between its SH3 domain and its second proline-rich motif in the primary structure, binding of a small G protein may either disrupt or enhance the intramolecular interactions discussed above. The rho-GAP region of p85β is only 42% identical to the corresponding domain of p85α, and it similarly lacks residues thought to be important for GTPase-promoting activity (20, 73).

p110γ differs from other class I enzymes because of its ability to be directly activated by βγ subunits of heterotrimeric G proteins (31, 32, 78, 79). Many G protein–coupled serpentine receptors increase the levels of various 3′-phosphorylated PtdIns species upon ligand binding. Recent evidence has shown that these increases are mediated by p110γ (78). One group found that Gβγ subunits could activate p110γ directly (32), whereas another reported that the associated p101 subunit binds Gβγ and is required for significant activation (33). Interestingly, a p85/p110-type PI3K activity was also shown to be activated by Gβγ subunits, but only in the presence of pTyr peptides that bind to the SH2 domains of the adapter subunit (79, 80). p110γ/p101 activity is not affected by phosphotyrosyl peptides. p110γ also contains a ras-binding domain and associates with ras in vitro (80A). However, a role for ras in p110γ activation has not been demonstrated.

Class II PI3Ks


Class II PI3Ks are large (170–210 kDa) proteins that contain a PIK domain and a catalytic domain 45–50% similar to class I PI3Ks (Figure 2). Class II PI3K genes have been cloned from humans and mice as well as from D. melanogaster, D. discoideum, and C. elegans (17, 81, 82, 83, 84). Each of these proteins contains a C-terminal region with homology to C2 domains; indeed, the class II PI3Ks have been termed PI3KC2 (used herein) or cpk (for C2-containing phosphoinositide kinase). Other proteins with C2 domains include certain protein kinase C (PKC) isoforms and the synaptic vesicle membrane protein synaptotagmin. C2 domains have been implicated in the Ca2+-dependent binding of proteins to lipid vesicles. However, certain Asp residues important for Ca2+ binding are absent in the C2 domains of class II PI3Ks. The Drosophila PI3KC2 was found to bind acidic phospholipids in a Ca2+-independent manner (81). Mammalian PI3KC2α and the novel human PI3KC2β have an additional sequence motif termed the PX domain (see Figure 2) (S Volinia, personal communication). Homologous domains have been noted in several signaling proteins, including the NADPH oxidase-associated proteins phox-40 and phox-47. The function of PX domains is unknown.


In vitro, class II PI3Ks preferentially phosphorylate PtdIns and PtdIns-4-P (Table 1), although human PI3KC2α phosphorylates PtdIns-4,5-P2 in the presence of phosphatidylserine (84). The enzymes from Drosophila, mouse, and human differ significantly in sensitivity to inhibition by wortmannin (IC50 values of 5, 50, and 450 nM, respectively) (81, 83, 84) (Table 1). It has not yet been determined which lipids are produced by class II PI3Ks in vivo and how their activities are regulated. If these enzymes make PtdIns-3,4-P2 in vivo, they are not likely to be active in resting cells where this lipid is undetectable. It is possible that class II PI3Ks contribute to the buildup of PtdIns-3,4-P2 observed in stimulated cells. There is evidence for rapid and transient tyrosine phosphorylation of PI3KC2α and -β following mitogen stimulation (82; S Volinia, personal communication), but the effect on enzyme activity is unknown.

Class III PI3Ks


The prototype class III PI3K was first identified in yeast in a screen for mutants conditionally defective in vacuolar protein sorting (85). The corresponding gene, VPS34, was cloned and found to be essential for accurate transport of newly synthesized proteins from the Golgi to the vacuole, an organelle similar to the lysosome of higher eukaryotes. The lipid kinase activity of this yeast protein was appreciated only after cloning of the p110 subunit of mammalian class I PI3Ks revealed that the proteins shared extensive sequence homology (86) (Figure 2). Bona fide VPS34 homologues have now been cloned from humans, Dictyostelium, and Drosophila (17, 87, 88). Each of the proteins also possesses a PIK domain (Figure 2).


Class III PI3Ks phosphorylate only PtdIns (Table 1); therefore they should be called PtdIns 3-kinases to differentiate them from the phosphoinositide 3-kinases with broader substrate specificity. Although the yeast Vps34 is relatively insensitive to wortmannin (IC50 ∼2.5 μM), human and Drosophila Vps34 homologues are inhibited with IC50 values of 2–10 nM (Table 1). In addition, wortmannin treatment lowers the level of PtdIns-3-P in platelets (89, 90).

The yeast protein Vps34 associates with another protein, Vps15, that is also required for vesicle sorting (91). Vps15 is a seri