Phosphatidylinositol Phosphate Kinases, a Multifaceted Family of Signaling Enzymes*

The importance of phosphoinositides as lipid signaling molecules in eucaryotic cells was first recognized by Lowell and Mabel Hokin in the 1950s (who also discovered the enzyme activities that phosphorylate phosphatidylinositol (PI)) (1–5). Since those early years, PI signaling pathways have expanded both in importance and complexity. The classical pathway transforms PI to PI-4,5-P2 by the successive actions of PI 4-kinases and PI-4-P 5-kinases. PI4,5-P2 is the precursor for second messengers and also acts directly to modify effectors, for example actin-binding proteins (6–9). Significant roles for other phosphoinositide lipid products in signaling, combined with recently identified lipid kinase activities, are illuminating the many mechanisms by which cells use lipid messengers (10–13). This review will focus on the phosphatidylinositolphosphate kinase (PIPK) family, which has the ability to synthesize all known PIP2 isomers and PIP3.

The regions of sequence homology are found within the C terminus. The identity between PIPKs can be as low as 27% (17). However, the regions of sequence identity are clustered, reminiscent of that between diverse protein kinases and phosphatidylinositol 3and 4-kinases (20,24). We now know these conserved domains represent the catalytic core of the kinases. The kinase domain, except in Fab1p homologs, is separated by an insert region. Recently, PIPKs in Arabidopsis thaliana have been characterized and based on their sequence appear to form a unique subfamily of enzymes (65). In Fig. 1, the subfamilies of PIPK have been summarized and the regions of identity in the kinase domains aligned and shown as a similarity plot.
Although statistically there is no homology with other kinases, there are similarities in the conserved sequence motifs. For example, a glycine-rich motif (GXSGS) in the homologs resembles the phosphate-binding loop of protein kinases and other ATP-binding proteins (24,31). It is also similar to GTP-binding sequences, which is consistent with the ability of PIPKs to use both ATP and GTP as phosphate donors (15,32). An invariant lysine (IIK sequence) that is C-terminal of the glycine-rich region resembles the conserved lysine found in protein kinases that binds the ␣-phosphate of ATP (24,25).
Because of their sequence identity to PIPKIs, PIPKII isoforms were thought to have similar substrate specificity. It therefore came as a surprise when the preferred substrate of PIPKII was identified as PI-5-P rather than PI-4-P (34). The product of PI-5-P phosphorylation remains PI-4,5-P 2 (17,(33)(34)(35). The earlier assumption that PIPKIIs are PI-4-P 5-kinase appears to be because of putative contamination of PI-4-P preparations with PI-5-P and the difficulty to distinguish PI-5-P from PI-4-P chromatographically. PIPKII isoforms also use PI-3-P as a substrate, and the product of this reaction is PI-3,4-P 2 . Therefore, in vitro the PIPKIIs are predominantly 4-kinases and appear to be less promiscuous in their substrate usage than PIPKIs. In summary, the in vitro substrate preference of the PIPKII isoforms is PI-5-P Ͼ PI-3-P Ͼ PI-4-P, 2 whereas the efficacy of substrate usage for characterized PIPKI isoforms is PI-4-P Ͼ PI-3-P Ͼ PI-3,4-P 2 Ͼ PI-5-P ϭ PI (14 -18, 33-38). The spectrum of messengers generated and substrates used by specific kinases in vivo is presumably tightly regulated. Nevertheless, these activities position PIPKs as potential participants in the generation of most known polyphosphoinositide signaling molecules. These kinase activities are remarkable and demonstrate the widest specificity of any family of signal-generating enzymes.
An indication that the in vitro activities of PIPKs may reflect their in vivo functions is the finding that PI-5-P and PI-3,5-P 2 have been recently identified in living cells (34,36). However, in both cases, the route of synthesis of these novel PIs remains to be elucidated. In vitro data show that certain PIPKI isoforms synthesize PI-5-P from PI (35). Whether PIPKI activity accounts for the low level of PI-5-P detected in NIH-3T3 fibroblasts or whether an as yet unidentified kinase or phosphatase is involved in its gener-* This minireview will be reprinted in the 1999 Minireview Compendium, which will be available in December, 1999. This is the third article of five in "A Thematic Series on Kinases (36) elegantly demonstrated that PI-3,5-P 2 is generated in vivo in both yeast and mammalian cells in response to osmotic stress. Yeast Mss4p is reported to produce only PI-4,5-P 2 and PI-3,4-P 2 (37,38). Thus, Fab1p, the only other PIPK homolog in yeast, is likely responsible for PI-3,5-P 2 production (see below).

PIP Kinase Structure
Recently, the structure of the first PIPK, PIPKII␤, has been solved (39). The tertiary structure of PIPKII␤ consists of two identical subunits that interact at the N terminus (Fig. 2). This structure is consistent with the oligomeric state of the protein in solution (14,15,39). At the dimerization site a combination of ␤ sheets and ␣ helices participates to form a clasp-like interface. The two ␤ sheets at the dimer interface form a flat surface area that extends beyond the two ␤ sheets to the C terminus of the PIPKII␤ (Fig. 2, top). Thus, although the subunits are globular the dimer is an elongated flat disc-like structure. One face of this disc is highly basic, containing 10 lysine, 4 histidine, and 4 arginine residues, with a net positive charge of ϩ10 (39). The charge and the flatness suggest that this region functions as an interface for membrane association. To illustrate this point the PIPKII␤ structure was modeled onto a phosphatidylcholine bilayer structure (Fig. 2, bottom). This demonstrated that the PIPKII␤ interface region contacts the bilayer through the phospholipid headgroups without penetration into hydrophobic membrane regions, suggesting that only electrostatic interactions are involved in docking. The reliability of this model is supported by orientation of the catalytic sites at the membrane interface.
Although the PIPKs are not statistically homologous to protein kinases, the structure does contain a region of roughly 80 amino acid residues in each subunit of the dimer that could be superimposed on the ATP binding and catalytic residues of protein kinase structures (39). Three catalytic site residues are absolutely conserved for PIPKs and have counterparts in protein kinases, namely Lys-150 (in the IIK sequence), Asp-278 (MDYSL), and Asp-369 (IID) of PIPKII␤ ( Fig. 1 and Ref. 36). The corresponding residues in protein kinase A (PKA) are Lys-72, Asp-166, and Asp-184 (40). The conserved lysine in PKA coordinates with the ␣-phosphate of ATP whereas Asp-184 binds Mg 2ϩ or Mn 2ϩ ions (40). Asp-166 appears to have a role in general base catalysis. Mutation of Lys-72 in protein kinases destroys activity, and this is also the case when Lys-150 is mutated in PIPKII␤. 2 Asp-166 in PKA is in the "RDLK" motif conserved in protein kinases. The PIPKs contain a similar conserved sequence, the "DLKGS" motif, that had been suggested as a counterpart (25). Although mutation of Asp-216 in the DLKGS destroys activity, 2 the structure of PIPKII␤ demonstrated that these two motifs are not equivalent. The counterpart of the PKA Asp-166, in the RDLK, is Asp-278 in the PIPKII␤ structure. The analog of the Asp-184 in protein kinases is Asp-369 in the PIPKs.
PIPKs have a glycine-rich loop or G-loop with the consensus sequence of GXSGS. This region corresponds topologically to the GXGXXG loop of protein kinases. In both kinase types this loop aligns consecutive peptide residues so that their amide groups interact with the phosphate triester backbone of ATP/GTP acting as a "phosphate anchor." The modeling of ATP into the binding pocket of PIPKII␤ showed that the adenine moiety fits into a conserved hydrophobic pocket in each monomer. The phosphate triester backbone is positioned to interact with the G-loop and Lys-150, and the ␥-phosphate is pointed toward the membrane interface. The specificity of PIPKII␤ was analyzed by modeling PI-5-P and ATP together into the structure. In this model, the 5-phosphate interacts with a cluster of partially conserved basic side chains in a shallow binding pocket. The openness of the putative catalytic site suggests that the phosphoinositol head group can freely rotate such that PI-3-P or PI-5-P could occupy the same binding site, consistent with the specificity of the kinases. In both cases, the 4-hydroxyl of inositol retains its position in line with the ␥-phosphate. A disordered loop in the structure (residues 373-391) was referred to as the "activation loop" because it topologically corresponds to the activation loop in protein kinases and possibly PI kinases (20,39,41). This loop is anchored on each side of the substrate binding site and thus has the potential to fold around the substrate and modulate both kinase specificity and activity. Taken together, the flatness of the membrane binding interface and the proximity of the substrate binding site to this interface indicate that the PIPKII dimer was elegantly designed to phosphorylate substrates at the membrane interface.

Cellular Functions and Signaling Mechanisms
Many of the phosphoinositide signaling enzymes directly associate with membrane receptors; the most characterized interactions are with tyrosine kinase receptors (6 -13). These signaling enzymes include isoforms of PI 3-kinase and phospholipase C. They associate at least transiently with receptors, and their activities are tightly coupled to agonist activation of receptors. Phospholipase C family members utilize PI-4,5-P 2 for second messenger production. Upon agonist stimulation, PI-4,5-P 2 is also observed to be a substrate for the agonist-activated PI 3-kinases (11), producing PI-3,4,5-P 3 . PI-3,4-P 2 can then be subsequently produced by the activity of 5-phosphatase (42). Polyphosphoinositide 5-phosphatases also associate with receptors (see the next minireview in this series Within this same theme, there is also evidence that PIPKs directly associate with receptors. These include the p55 tumor necrosis factor (TNF) receptor and the epidermal growth factor (EGF) receptor (27,43). In the case of the p55 TNF receptor, only PIPKII␤ associates with the receptor; the highly homologous PIPKII␣ does not (27). The EGF receptor is reported to associate with both PI 4-kinase and PIP 5-kinase activities, although the kinase isoforms have not been defined (43). Interestingly, both the TNF and EGF receptors interact with PIPKs through their juxtamembrane domains within sequences of ϳ20 residues extending from the membrane interface into the cytoplasm. The site on PIPKII␤ required for association is within the N-terminal 100 amino acids close to the dimerization interface. The PIPKII␤ association with receptors may facilitate dimer formation and specificity of membrane assembly. The functional significance of these interactions remains to be demonstrated, but these signaling enzymes are likely to contribute to second messenger production or regulation of receptor function. In platelets, there is also evidence for stimulation of a PI-3-P 4-kinase activity by thrombin and protein kinase C (44,45). PIP-KII␣ is present in platelets and as the only currently identified PI-3-P 4-kinase in platelets is a candidate for that PI-3-P 4-kinase activity (46,47). Finally, there is evidence that the newest PIPKII isoform, the PIPKII␥, can be phosphorylated in response to mitogenic stimulation (30).

Secretion/Vesicle Trafficking
Phosphoinositide signaling has been implicated in secretion and vesicular trafficking and has been reviewed recently (48). Fab1p was the first PIPK homolog to have an implied role in vesicular trafficking (17,25). Mutation or deletion of Fab1p results in formation of aploid and binucleate cells and a defect in vacuole function and morphology (25). The vacuole defect is manifested by enlarged vacuoles that fill the cytoplasm; other phenotypes appear to be secondary. Recently, a link has developed between FAB1 and VPS34. VPS34 encodes the only yeast PI 3-kinase, and it is required for protein sorting to the vacuole (49). When yeast were osmotically stressed the synthesis of PI-3,5-P 2 was stimulated, and this was dependent upon Vps34p function (36). Thus, PI-3-P generated by Vps34p is likely a substrate for a PIPK, and the most likely candidate is Fab1p. Fab1p mutants show defects in vacuolar function and may share some of the phenotypes of Vps34p mutants. Thus, Fab1p may form a functional alliance with Vps34p to synthesize PI-3,5-P 2 (see the second minireview in this series (49)).
A role for PIPK and PI-4,5-P 2 has been reported for catecholamine secretion (50). The process that required PIPK activity is the ATP-dependent priming step. This step is required for the subsequent Ca 2ϩ -dependent triggering of secretion. The PIPKs required for ATP priming were specific for PIPKI enzymes. In light of the specificity of the PIPKs, this in retrospect may not be surprising. The enzymes that are implicated in priming, PIPKI␣, -␤, and/or -␥, have at least one property in common, stimulation by phosphatidic acid (PA) (16,28,29). Phospholipase D (PLD), which synthesizes PA, has also been implicated in vesicular trafficking and possibly secretion (51). In fact, mammalian PLDs require PI-4,5-P 2 for robust enzymatic activity (52). Hence, the PIPKI enzymes may be functionally regulated by PA, and the PIP 2 produced may stimulate PLD activity. This would result in a positive stimulatory loop that would generate both PA and PIP 2 at specific sites within cells. In the case of Ca 2ϩ -stimulated secretion, the PIP 2 and PA synthesis may be spatially segregated to the secretory vesicle. In other signaling pathways, the synthesis of PA and PIP 2 could be spatially and temporally regulated to modulate other functions such as actin assembly.

Actin Assembly
For two decades evidence has accumulated that PI-4,5-P 2 can modulate the activity of proteins which regulate the assembly or disassembly of actin filaments in vitro (53). Recently, it was reported that overexpression of PIPKI isoforms in COS cells induced dramatic reorganization of actin in vivo (54). However, surprisingly a mutant of the PIPKI which lacked activity also induced actin reorganization (29). As a result, there remains some doubt as to whether or not PIP 2 generation is required for PIPKI modulation of actin assembly in vivo. Nevertheless, if the PIPKs indeed modulate actin assembly by synthesis of PIP 2 this would presumably require syntheses of PI-4,5-P 2 to be spatially and temporally regulated. If PIPKI activity is required for regulation of actin assembly, then a role for PLD and PA is also conceivable. Activation of PLD is mediated by many stimuli that also induce actin reorganization; in particular PLD is stimulated by the small G-protein Rho (55,56). Rac and Rho modulate actin assembly (55) and have also been reported to activate or associate with PIPKI isoforms (57)(58)(59)(60). These results again illustrate the potential for a set of ordered interactions between the PIPKs and other signaling molecules that in turn modulate PIP 2 production.
The MSS4-encoded PIPK is important for actin assembly in yeast (36,37). Yeast lacking Mss4p do not have the ability to form normal actin filaments and to properly localize their actin cytoskeleton during polarized cell growth. Interestingly, overexpression of Rho2p, a Rho GTPase, restored growth and polarized actin assembly (36). Yeast lacking Mss4p are rescued by a PIPKI isoform, and the enzymatic properties of Mss4p, including PA stimulation, are most similar to PIPKI isoforms (37).

Nuclear Phosphoinositide Signaling
Classical PI signaling pathways are based at the plasma membrane and linked to receptor activation by agonists. However, a distinct PI cycle exists in the nucleus and is regulated independently from the plasma membrane PI cycle (see for review see Ref. 61). Within nuclei the spatial organization of these signaling pathways has been vague. Biochemical evidence suggests that a fraction of the enzymes and phosphoinositides are retained in highly purified nuclei that have been stripped of their nuclear envelope with detergent (62). The explanation is that the enzymes and phosphoinositides are not associated with membrane structures within the nucleus.
Some PIPKI and PIPKII isoforms are concentrated in nuclei, spatially organized to "nuclear speckles" that are separated from known membrane structures (63). These speckle structures also contain enzymes required for mRNA processing and transcription (Fig. 3). These same speckles are reported to contain polyphosphoinositide(s), suggesting that the PIPKs generate PIP 2 in speckles (63). Nuclear speckles are highly dynamic, and their morphology is tightly linked to the state of mRNA transcription. Inhibition of mRNA transcription induces these structures to become larger and fewer in number; the PIPKs and PIP 2 reorganize identically. Although the function of the PIPKs at speckles is not known, this is additional evidence for spatial organization of the PIPKs that presumably are linked functionally to processes modulated by the FIG. 3. The immunofluorescent localization of the PIPKI␣ in nuclei (green) in A, compared with that of small nuclear ribonucleoprotein components detected by an Sm autoantibody that detects nuclear speckles in B (red), and the overlay in C (yellow) showing that the speckles colocalize. The PIPKI␣, PIPKII␣, and PIPKII␤ show an identical speckle-patterned nuclear localization, and monoclonal antibodies to PI-4,5-P 2 also stain these same nuclear structures (see Ref. 63 for details).

Minireview:
The PIP Kinase Family 9909 polyphosphoinositide messengers that they generate. There is also evidence for the regulated localization of diacylglycerol kinase to nuclei (64). Consequently, PA generated by diacylglycerol kinase could activate nuclear localized PIPKI isoforms.

Prospectus and Future Directions
These combined data suggest a model of how the PIPKs may assemble in cells at the membrane, in vesicles, on receptors, or within nuclei. The in vitro data demonstrate that PIPK substrate specificities are broad, suggesting that the PIPK family may generate all possible combinations of polyphosphoinositide signaling molecules. In vivo, the specificity for polyphosphoinositide substrates and the generation of different products by these kinases must be tightly regulated. One mechanism by which to regulate PIPK specificity is to channel phosphoinositide substrates to these kinases. This could occur by assembling both PI and PIPKs together into a macromolecular complex or within a cellular compartment. In this model, PI kinases would synthesize a specific PIP isomer, which would then be a substrate for the contiguous PIPKs (Fig. 4). PI-4,5-P 2 concentrations within cells are reported to not fluctuate greatly. Thus, it is difficult to understand how effectors of PI-4,5-P 2 can be modulated in vivo. The model suggests that PIPKs spatially organize within cells and that the spatial and temporal generation of PI-4,5-P 2 at these cellular sites or compartments is regulated and coupled to effectors also at these sites. As such, PIP 2 concentrations would vary at specific cellular compartments but not globally within cells. Clearly such a model needs refinement, but the concept illustrates the paths that can be taken.