There are eight members of the phosphoinositide family of phospholipids in eukaryotes; PI, PI3P, PI4P, PI5P, PI(4,5)P2, PI(3,4)P2, PI(3,5)P2 and PI(3,4,5)P3. Receptor activation of Class I PI3Ks stimulates the phosphorylation of PI(4,5)P2 to form PI(3,4,5)P3. PI(3,4,5)P3 is an important messenger molecule that is part of a complex signalling network controlling cell growth and division. PI(3,4,5)P3 can be dephosphorylated by both 3- and 5-phosphatases, producing PI(4,5)P2 and PI(3,4)P2, respectively. There is now strong evidence that PI(3,4)P2 generated by this route does not merely represent another pathway for removal of PI(3,4,5)P3, but can act as a signalling molecule in its own right, regulating macropinocytosis, fast endophilin-mediated endocytosis (FEME), membrane ruffling, lamellipodia and invadopodia. PI(3,4)P2 can also be synthesized directly from PI4P by Class II PI3Ks and this is important for the maturation of clathrin-coated pits [clathrin-mediated endocytosis (CME)] and signalling in early endosomes. Thus PI(3,4)P2 is emerging as an important signalling molecule involved in the coordination of several specific membrane and cytoskeletal responses. Further, its inappropriate accumulation contributes to pathology caused by mutations in genes encoding enzymes responsible for its degradation, e.g. Inpp4B.
Pathways for the formation of PI(3,4)P2
The currently accepted pathways for synthesis of PI(3,4)P2 are shown in Figure 1. Activation of Class I PI3Ks by a wide variety of GPCRs and RTKs leads to the formation of PI(3,4,5)P3 . This is accompanied by formation of PI(3,4)P2 [2–4]. The relative magnitude and kinetics of PI(3,4,5)P3 and PI(3,4)P2 production appear to vary widely according to the receptor and cell context but the formation of PI(3,4)P2 is usually lagged and slower than accumulation of PI(3,4,5)P3. These kinetics, together with the preference of Class I PI3Ks for PI(4,5)P2in vitro (under conditions of physiological substrate presentation) and the discovery of active PI(3,4,5)P3 5-phosphatases, suggested this PI(3,4)P2 was created by subsequent dephosphorylation of PI(3,4,5)P3 . The tissue/cell type which shows the greatest relative accumulation of PI(3,4)P2 is thrombin-stimulated mammalian platelets and, here, the genetic deletion of the 5-phosphatase SHIP1 dramatically reduces the accumulation of PI(3,4)P2, confirming that PI(3,4,5)P3 hydrolysis is the major route for PI(3,4)P2 formation under these conditions [6,7]. There have also been several recent studies using fluorescent reporters of PI(3,4)P2 accumulation which have indicated dephosphorylation of PI(3,4,5)P3 by 5-phosphatses is required for formation of pools of spatially localized PI(3,4)P2 on early endocytic structures, dorsal ruffles and lamellipodia (see below). However, the extent to which these pools contribute to the total PI(3,4)P2 synthesized downstream of Class I PI3K activation (e.g. downstream of RTKs responding to growth factors) is unclear and it remains theoretically possible that significant amounts of PI(3,4)P2 might be formed by the action of Class I PI3Ks acting on PI4P directly.
Pathways of PI(3,4)P2 formation
Recent studies have indicated that PI(3,4)P2 can also be formed by the direct phosphorylation of PI4P catalysed by the Class II PI3K, PI3KC2α, during clathrin-mediated endocytosis (CME)  (see below). Class II PI3Ks are also thought to phosphorylate PI to form PI3P, but the extent and factors that determine which reaction is catalysed by these enzymes are still largely unclear .
PI(3,4)P2 is thought to be metabolized largely by the 4-phosphatases Inpp4A and Inpp4B, to form PI3P . The best evidence for the involvement of Inpp4A/B is during endocytosis and the extent to which these enzymes carry the flux through PI(3,4)P2 within a receptor-stimulated Class I PI3K/PI(3,4,5)P3 5-phosphatase pathway is uncertain. Further, the somewhat paradoxical lack of PI3P accumulation that is often seen to accompany PI(3,4)P2 accumulation under these conditions might suggest that PI(3,4)P2 can also be de-phosphorylated by 3-phosphatases, perhaps by members of the phosphatase and tensin homolog (PTEN) family, e.g. voltage-sensitive phosphatase or Pten itself [10,11].
It is clear from the above that despite a large body of research, many key aspects of the enzymology defining the pathways for PI(3,4)P2 metabolism are still uncertain. One of the barriers to progress is that most quantitative approaches to measuring PI(3,4,5)P3 and PI(3,4)P2 still rely on radio-labelling and HPLC, which are laborious, require large quantities of radioactive precursors (3H-inositol or 32P-Pi) and a compatible biological system (usually cells in culture). A further, more profound barrier to understanding is that many of the relevant enzymes show varying and overlapping degrees of substrate selectivity when assayed in vitro, e.g. several 5-phosphatases can remove the 5-phosphate from both PI(4,5)P2 and PI(3,4,5)P3 [9,12]. This is probably partly because constraints which restrict their activity in cells are difficult to re-construct in vitro (e.g. membrane environment/access to substrate) and partly because the pathways have evolved some redundancy, creating signalling networks that are robust, yet retain flexibility and sensitivity. This means it is usually difficult to infer the physiological function of a route of PI(3,4)P2 formation simply by manipulating the levels of a single enzyme.
Direct effectors of PI(3,4)P2
The receptor-stimulated synthesis of PI(3,4,5)P3 represents a central signal transduction pathway that co-ordinates several important cell responses, such as the regulation of cell growth and movement . The principle by which PI(3,4,5)P3 acts as a signal in this network is by the binding of its phosphorylated head-group to conserved protein domains in several effector proteins, the best studied of which are a sub-group of PH domains [13–15]. The binding of PI(3,4,5)P3 to these effectors regulates their localization and/or activation in areas of the plasma membrane in which this phospholipid is synthesized. Some of these PH-domain-containing effectors bind to PI(3,4,5)P3 with substantial selectivity over all other phosphoinositides (e.g. Btk, Grp-1). Most PI(3,4,5)P3 effectors however can bind both PI(3,4,5)P3 and PI(3,4)P2 with similar relative affinities, but bind other phosphoinositides with a range of lower affinities. In some cases this preference over other phosphoinositides is so large that they are clearly evolved to respond to rises in PI(3,4,5)P3 and PI(3,4)P2 concentrations after Class I PI3K activation (e.g. Pdk1, Akt1/2, Arap3, Prex1/2, Dapp1) but, in other cases, this selectivity is not so clear, suggesting they may be recruited by several different phosphoinositides. There is only one family of proteins, Tapp1/2, which has clearly been demonstrated to possess a PH domain with substantial selectivity for PI(3,4)P2 over all other phosphoinositides .
PI3P is an accepted regulatory molecule controlling the localization and function of several effectors in the endo-lysosomal system . PI3P can be recognized by effectors with FYVE or PX domains which can bind the head group of this lipid with high selectivity over all other phosphoinositides (e.g. EEA1 and p40Phox, respectively). In most scenarios where the role of PI3P is best understood, the route for PI3P formation is via Vps34-containing complexes that direct a Class III PI3K activity to the relevant membrane location (e.g. complex I to autophagosomes, complex II to endosomes). Several recent studies however have suggested some PX-domain-containing effectors possess more promiscuous-binding selectivities and can be recruited and regulated by PI(3,4)P2 in specific plasma membrane or endocytic locations [e.g. Tks5, sorting nexin 9 (Snx9)] [8,18,19]. The PI(3,4)P2 generated in these scenarios has been suggested to arrive via dephosphorylation of PI(3,4,5)P3 or via Class II PI3K (see below).
All direct effectors of phosphoinositides possess other upstream regulatory interactions that co-operate with phosphoinositide binding to regulate the context and scale of their interaction with downstream elements of the signalling network (e.g. GTPases, protein kinases or, adaptor proteins). In some scenarios, the level of phosphoinositide-binding specificity, combined with large changes in concentration of the relevant phosphoinositide ligand, point clearly to a major regulatory impact. In other scenarios, the role of phosphoinositide binding is less clear, and may be best viewed as either facultative (if not driven by changes in Iigand concentration) or pleiotropic (if regulation can occur with more than one ligand/context). However, the observation that some effectors have evolved preferential affinity towards PI(3,4)P2 does suggest that this lipid must carry some molecule-specific signalling information.
Positive and negative roles for PI(3,4)P2 in Class I PI3K signalling
The levels of PI(3,4,5)P3 which accumulate after receptor activation of Class I PI3Ks are controlled by the action of both 3- and 5-phosphatases . The role of Pten as a major PI(3,4,5)P3 3-phosphatase is clear and is accepted to be a major reason why loss of Pten is a highly prevalent driving mutation in many different human cancers, i.e. Pten is a tumour suppressor which limits the pro-growth activity of the Class I PI3K signalling pathway . The role of 5-phosphatases in controlling the activity of Class I PI3K signalling is much less clear and probably reflects the context-dependency with which PI(3,4)P2 can act as a positive signal in this pathway [20,22]. The discovery that the PI(3,4)P2 4-phosphatase Inpp4B is also a major tumour suppressor, sometimes acting alongside the deletion of Pten, is interpreted to indicate that rises in PI(3,4)P2 levels can be pro-growth [23,24] (although Inpp4B has also been suggested to act as a PI(3,4,5)P3 phosphatase in some tissues ). In the present study, PI(3,4)P2 is usually imagined to act alongside PI(3,4,5)P3 in the regulation of joint PI(3,4,5)P3/PI(3,4)P2 effectors (such as Akt1/2), but recent data suggests it may actually represent a qualitatively distinct signal, by regulating a spatially discrete pool of Akt2 in endosomes .
However, there are other circumstances where 5-phosphatases acting on PI(3,4,5)P3 appear to clearly down-regulate Class I PI3K signalling. There is evidence that recruitment of SHIP1 to ITIM motifs in inhibitory FcγRIIB-receptors acts to limit Class I PI3K signalling through antibody and antigen receptors in B-cells and mast cells . There is evidence that SHIP2 acts to limit insulin-stimulated glucose uptake in muscle  and very recently, INPP5J has been suggested to be a tumour suppressor in breast cancer . The general assumption here is that removal of PI(3,4,5)P3 by the 5-phosphatase down-regulates the pathway by inhibiting PI(3,4,5)P3-dependent signalling. Equally plausible however, is the idea that PI(3,4)P2 can act in a negative feedback loop. Strong evidence in favour of this concept comes from studying the Tapp proteins. Tapp1 and 2 are the only proteins characterized thus far with clear specificity for binding PI(3,4)P2 over all other phosphoinositides. Mice have been created with point mutations in the C-terminal PH domain of these two proteins which do not affect expression but destroy PI(3,4)P2 binding. These mice exhibit enhanced insulin-stimulated phosphorylation of Akt in heart and muscle  and enhanced antigen-stimulated phosphorylation of Akt in B-cells , suggesting Tapp proteins act to supress activation of Class I PI3K signalling. Tapp1/2 can interact with several proteins via their PDZ domains (e.g. PtpL1, Mupp1, syntrophins)  but the molecular mechanisms by which they might exert these effects are still unclear. Nonetheless, Tapp1/2-mediated feedback inhibition of Class I PI3K signalling remains the single most convincing, specific, signalling effect of PI(3,4)P2.
Thus, the available evidence suggests it is too simplistic to characterize a general role for PI(3,4)P2 in Class I PI3K signalling; how and where the pathway is being activated, the involvement of effectors which can or cannot discriminate between PI(3,4,5)P3 compared with PI(3,4)P2 and also the magnitude and spatiotemporal kinetics of PI(3,4,5)P3 compared with PI(3,4)P2 accumulation all appear to play a role in defining PI(3,4)P2 function. Some further, specific examples of where PI(3,4)P2 has been proposed to play an important regulatory role are discussed below.
Roles for PI(3,4)P2 in different routes of endocytosis
CME is a major route for the directed internalization of specific cargos at the cell surface, including the recycling of many different growth factor receptors. It is a complex process in which PI(4,5)P2 plays an important role in initiating the AP2-adaptor-cargo-dependent formation of a clathrin-coated pit and involves the co-ordinated activity of >30 different types of proteins . The Class II PI3K, PI3KC2α, is recruited on to the cytosolic face of the growing clathrin-caged structure, where its activity is required for efficient maturation and scission from the plasma membrane . A major role for PI3KC2α in this process appears to be the synthesis of PI(3,4)P2 and the subsequent recruitment of the SH3-PX-BAR-domain-containing scaffold protein, Snx9; the PX domain is thought to mediate PI(3,4)P2-dependent recruitment and the BAR domain stabilizes highly curved or tubulated membrane structures . The SH3 domain of Snx9 binds to specific polyproline motifs in target proteins, including dynamin and WASPs. These interactions are thought to direct Snx9 to areas of high curvature around the neck region of the maturing vesicle, promoting the recruitment and activation of dynamin and the WASP-Arp2/3-dependent construction of an associated actin filament network, which together allow efficient constriction and scission of the clathrin-coated vesicle from the plasma membrane  (Figure 2).
Roles for PI(3,4)P2 in endocytosis
Macropinocytosis is an actin-driven invagination of the plasma membrane that is usually imagined to coincide with the ‘folding back’ of circular dorsal membrane ruffles. Macropinocytosis creates large (>0.2 μm) fluid-filled endocytic structures that sample the extracellular environment. They can be important for fluid phase uptake of nutrients and both dorsal ruffling and increased macropinocytosis are common responses to growth factor stimulation. Membrane ruffling is widely observed to be dependent on Class I PI3K signalling and the regulation of the cortical actin cytoskeleton by PI(3,4,5)P3-regulated GEFs/GAPs for Rho and Arf family GTPases . In addition, recent studies have now implicated the stepwise de-phosphorylation of Class I PI3K-generated PI(3,4,5)P3 to PI(3,4)P2, and then to PI3P, in the efficient formation and closure of macropinosomes  (Figure 2). There is evidence that dorsal ruffling is regulated by the PI(3,4)P2 effector TAPP1, possibly via its interaction with syntrophins . There is also evidence that PI(3,4)P2 plays a similar role to that described in CME, recruiting PX-BAR-containing sorting nexins (Snx5, 9 or 18) and promoting effective maturation and scission of the macropinosome from the plasma membrane .
Fast endophilin-mediated endocytosis (FEME) is a very recently described clathrin-independent route of endocytosis that allows cargo-specific internalization of certain receptors, e.g. some RTKs, GPCRs and the IL2-R . FEME is regulated by growth factor stimulation and dependent on Class I PI3K activation. There is good evidence that SHIP1/2-dependent production of PI(3,4)P2 is required for the recruitment of lamellipodin (Lpd) and its binding partner endophilin . Endophilin is a BAR-domain-containing protein that recruits dynamin and initiates actin filament rearrangement through WASP or Ena/Vasp interactions, thereby promoting scission from the plasma membrane [38,39] (Figure 2). Thus PI(3,4)P2 appears to play remarkably analogous roles in CME, macropinocytosis and FEME (Figure 2).
A number of studies have pointed to a clear demarcation in the roles for different phosphoinositides in distinct membrane compartments [13,40]. Thus PI(4,5)P2 and PI(3,4,5)P3 are thought to play their major roles in the inner leaflet of the plasma membrane and during early events shaping the formation of endocytic structures. Indeed the removal of these two lipids is thought to represent a critical timing point in the maturation of these structures. The generation of PI(3,4)P2 though 5-phosphatase-mediated removal of PI(3,4,5)P3 or via PI3KC2-mediated phosphorylation of PI4P appears to be induced or can persist later in the lifetime of endocytic vesicles, playing a role in their scission from the plasma membrane (see above) or in their function on entering the early endocytic compartment (Figure 2). In this later respect, PI(3,4)P2 generated in endosomes, via PI3KC2γ  or PI(3,4,5)P3 hydrolysis, has been proposed to represent a sustained and spatially-restricted signal for APPL1/2-supported Akt2 activation [26,42]. In this context, 4-phosphatase catalysed removal of PI(3,4)P2 is proposed to restrict growth. However, in another context, Inpp4B-mediated conversion of PI(3,4)P2 to PI3P has been shown to regulate PI3P-dependent activation of another AGC family kinase, Sgk3, promoting growth and invasive migration . Thus Inpp4B can be viewed as supporting or supressing tumour growth, perhaps explaining its inconsistent role as a tumour suppressor.
Roles for PI(3,4)P2 in the formation of invadopodia, lamellipodia and membrane ruffles
A collection of recent studies has further implicated PI(3,4)P2 in the regulation of the actin network in specific cellular compartments that share some common features (Figure 3). Growth factor-stimulated membrane ruffling and lamellipodia formation has been shown to depend on membrane recruitment of Lpd, where it clusters and tethers members of the Ena/Vasp family to actin filaments, antagonising capping proteins and promoting profilin/actin-ATP delivery to the growing (barbed) ends of filaments [44–46] (Figure 3). PI(3,4,5)P3-generated PI(3,4)P2 has been proposed to play a role in the direct membrane recruitment of Lpd, via binding to its PH domain , and PI(3,4)P2 availability itself may be buffered by binding to profiling . Ena/Vasp proteins are also important in regulating the growth of actin filaments during neurite outgrowth, axon guidance and have recently been implicated in mediating transient interactions with talin–integrin complexes .
Roles for PI(3,4)P2 in the regulation of lamellipodia, ruffling and invadopodia
PI(3,4)P2-dependent recruitment of Lpd can also contribute to GTP–Rac-mediated regulation of the SCAR/WAVE complex and the Arp2/3-dependent nucleation of branched actin filaments  (Figure 3). SCAR/WAVE regulation of the actin cytoskeleton is important for the efficient formation of protrusive lamellipodia, cell migration and wound closure .
Ship2/Snj2-generated PI(3,4)P2 has also been proposed to drive PX-domain-mediated membrane recruitment of Tks5, which promotes nucleation of multi-protein complexes, including N-WASP and Arp2/3, leading to the formation of podosomes and invadopodia [18,51,52]. Podosomes are large actin-driven processes at the base of the cell responsible for the degradation of extracellular matrix. Invadopodia are analogous structures in cancer cells that are more motile and rapidly turned over than podosomes and which support invasion and metastasis. Importantly, Tks5 is an obligate component of invadopodia .
Thus PI(3,4,5)P3 5-phosphatase mediated formation of PI(3,4)P2 has been proposed to play a role in the Class I PI3K-mediated regulation of the actin cytoskeleton that underpins several cell structures [32,53]. A core feature of PI(3,4)P2 involvement appears to be aiding the recruitment of PX- or PH-domain-containing scaffolding proteins that mediate WASP/SCAR or Ena/Vasp-regulated branching and elongation of actin filaments.
In principle, formation of PI(3,4)P2 could be viewed as an alternative pathway for removal of PI(3,4,5)P3, an alternative route to formation of PI3P or the formation of a regulatory molecule in its own right [22,32]. There are now good examples for all three of these scenarios. Thus, the 5-phosphatase SHIP1 acts to supress antigen-receptor signalling in B-cells, Inpp4B-generated PI3P is important in signalling to Sgk3 and PI(3,4)P2 controls the maturation of endocytic structures. There are also some remarkably common elements in several of the proposed regulatory roles for PI(3,4)P2, particularly in the direct or indirect recruitment of proteins that co-ordinate actin filament rearrangement and membrane deformation. In most cases, the lack of clarity in the substrate-specificity of enzymes making or degrading PI(3,4)P2, or in the phosphoinositide-binding selectivity of potential effectors, particularly when assayed in vitro, mean defining the precise significance of PI(3,4)P2 binding in many of these processes remains uncertain. However, this apparent lack of specificity may be mitigated by a strong spatial dimension to the PI(3,4)P2 signal, for example coincident signalling in specific endocytic structures or patches of membrane ruffling. Clearly, there is much that we still do not understand about PI(3,4)P2 as a signalling molecule, but making progress in this area will be vital to understanding the contribution that PI(3,4)P2 makes to pathology, particularly where over activation of Class PI3K signalling drives cancer  or in other diseases driven by malfunction of phosphoinositide metabolizing enzymes .
We thank Mouhannad Malek and Anna Kielkowska for helpful reading of the manuscript and Veronique Juvin for drawing the figures.
This work was supported by the Biotechnology and Biological Sciences Research Council, U.K. [grant number BB/J004456/1].
Signalling 2015: Cellular Functions of Phosphoinositides and Inositol Phosphates: Held at Robinson College, University of Cambridge, Cambridge, U.K., 1–4 September 2015.