Structural features of human inositol phosphate multikinase rationalize its inositol phosphate kinase and phosphoinositide 3-kinase activities

Human inositol phosphate multikinase (HsIPMK) critically contributes to intracellular signaling through its inositol-1,4,5-trisphosphate (Ins(1,4,5)P3) 3-kinase and phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P2) 3-kinase activities. This catalytic profile is not conserved; orthologs from Arabidopsis thaliana and Saccharomyces cerevisiae are predominantly Ins(1,4,5)P3 6-kinases, and the plant enzyme cannot phosphorylate PtdIns(4,5)P2. Therefore, crystallographic analysis of the yeast and plant enzymes, without bound inositol phosphates, do not structurally rationalize HsIPMK activities. Here, we present 1.6-Å resolution crystal structures of HsIPMK in complex with either Ins(1,4,5)P3 or PtdIns(4,5)P2. The Ins(1,4,5)P3 headgroup of PtdIns(4,5)P2 binds in precisely the same orientation as free Ins(1,4,5)P3 itself, indicative of evolutionary optimization of 3-kinase activities against both substrates. We report on nucleotide binding between the separate N- and C-lobes of HsIPMK. The N-lobe exhibits a remarkable degree of conservation with protein kinase A (root mean square deviation = 1.8 Å), indicating common ancestry. We also describe structural features unique to HsIPMK. First, we observed a constrained, horseshoe-shaped substrate pocket, formed from an α-helix, a 310 helix, and a recently evolved tri-proline loop. We further found HsIPMK activities rely on a preponderance of Gln residues, in contrast to the larger Lys and Arg residues in yeast and plant orthologs. These conclusions are supported by analyzing 14 single-site HsIPMK mutants, some of which differentially affect 3-kinase and 6-kinase activities. Overall, we structurally rationalize phosphorylation of Ins(1,4,5)P3 and PtdIns(4,5)P2 by HsIPMK.

Considerable attention is focused on the enzymes that regulate the metabolism and hence the myriad cell signaling activities of the inositol phosphates and the inositol lipids. These are two physicochemically and functionally distinct groups of intracellular signals (1), which typically each rely on separate families of kinases for their synthesis. The sole exception is the inositol phosphate multikinase, initially named for its ability to phosphorylate inositol phosphates (2,3), but later found to also phosphorylate PtdIns(4,5)P 2 (4). This "dual-specificity" has endowed the inositol phosphate multikinase (IPMK) 2 family with multiple biological activities. For example, IPMK is indispensable for connecting PLC-mediated Ins(1,4,5)P 3 release to the generation of InsP 5 (5)(6)(7); the latter is a precursor for InsP 6 and the inositol pyrophosphates, which each have many cellular functions (1,8). Activation of the inositol phosphate kinase activities of IMPK appears to be a key response in the Wnt/␤catenin signaling pathway (9). IPMK is mainly localized in the nucleus (4,10,11), where its kinase activities have been shown to mediate cellular differentiation programs (12), and transcript-selective mRNA export from the nucleus (13). Also in the nucleus, the PtdIns(4,5)P 2 3-kinase activity of IPMK stimulates the transcriptional activity of the nuclear receptor steroidogenic factor 1 (14). In addition, mammalian IPMK has moonlighting functions, unrelated to its catalytic activities, which are mediated through interactions with a number of protein-binding partners, such as mTOR (mechanistic target of rapamycin) (15), p53 (16), and AMP-activated protein kinase (17).
The wide-ranging importance of the IPMKs is underscored by the observation that knock-out of the IPMK gene in mice is embryonic lethal (5). There are also some pathological consequences for genetic defects in human IPMK that have been associated with a reduction in its kinase activities. For example, a heterozygous, frameshift mutation in the human IPMK gene has been identified in six members of the same family who all developed small intestinal neuroendocrine tumors; these individuals also exhibited a reduction in InsP 5 synthesis (18). Additionally, impaired IPMK transcription and a decrease in IPMK stability has been linked to the pathology of Huntington's disease, by virtue of an attenuation of the PtdIns(3,4,5)P 3 /AKT This work was supported by the Intramural Research Program of the National Institutes of Health, NIEHS. The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The atomic coordinates and structure factors ( cro ARTICLE signaling cascade (19). Genome-wide association studies have described that decreased IPMK expression in brain tissue is associated with the pathogenesis of inflammation-associated neurodegeneration (20).
IPMK is a member of the so-called IP-kinase family that includes IP3Ks and IP6Ks (21). Characterization of the structures of each of these enzymes can rationalize their alternate catalytic specificities, assist in deciphering evolutionary relationships, and permits rational design of enzyme-specific inhibitors. Previous work has described crystal structures of an IP3K (21,22) and an IP6K (23), each with substrates captured in the active site. Two previous studies have described the crystal structures of IPMK orthologues from Saccaromyces cerevisiae (24) and Arabidopsis thaliana (25), at resolutions of 2 and 2.9 Å, respectively. However, neither study captured inositol phosphate within the active site.
Molecular modeling has generated a consensus view that IPMKs host a relatively spacious and conformationally flexible substrate-binding pocket, in which mobile side chains of Lys and Arg residues play major roles (24,25). Here, we demonstrate that this model does not apply to HsIPMK; we describe a catalytic pocket that is more constrained than those of the plant and yeast orthologs. Also unique to mammalian IPMK is a catalytically important proline-loop, and a preponderance of Gln residues in the active site. These conclusions are drawn from our description, for the first time, of the crystal structure of HsIPMK. Moreover, we present the first structures of any IPMK in complex with inositol phosphate substrate: we describe two versions of Ins(1,4,5)P 3 within the active site, first as a free inositol phosphate, and second as the headgroup of a soluble analogue of PtdIns(4,5)P 2 . This allows a structural rationalization of 3-kinase activity toward Ins(1,4,5)P 3 and PtdIns(4,5)P 2 .

Structure of human IPMK
We produced crystals of the core catalytic domain of the enzyme that contains residues 50 to 416, from which we deleted an internal domain comprising residues 263 to 377 ( Fig. 2A); that deletion, which was necessary to obtain crystals, was replaced with a simple Gly-Gly-Ser-Gly-Gly linker ( Fig. 2A). Previous work has shown that this deletion does not compromise catalytic activity (26); this is a non-catalytic region of the protein. It contains a nuclear localization sequence, flanked by residues that host protein kinase phosphorylation sites that regulate nuclear localization sequence functionality (26).
The structure of the IPMK apoenzyme was determined by a molecular replacement approach using a model constructed from the template of ScIPMK (PDB accession code 2IF8). That information was then used for further elucidation of the structures of crystal complexes with ADP plus either Ins(1,4,5)P 3 (Fig. 2, A-D) or diC 4 -PtdIns(4,5)P 2 (see below).
For each asymmetric unit, there is one molecule of IPMK in space group P4 2 2 1 2 ( Table 1). Analysis of the overall fold of HsIPMK (Fig. 2B) reveals domains that are similar to the so-  (21). Note that amino acid residues in a domain responsible for nuclear localization (263 to 377) were replaced with a Gly-Gly-Ser-Gly-Gly linker. B, ribbon plot of the HsIPMK structure. NLS ϭ nuclear localization sequence. ADP and InsP 3 are shown as stick models within an 2F o Ϫ F c electron density, which is contoured at 1.5 . Two magnesium atoms are depicted as magenta balls. C, electrostatic surface plot with blue and red coloration to denote positive and negative electrostatic potentials, respectively, at physiological pH. D, a manual alignment of amino acid sequences (HsIPMK, NP_689416.1; HsIP6K2, NP_057375.2; and HsIP3KA, NP_002211.1), guided by the structural elements that have been observed in crystal structures, and in the case of HsIP6K2, secondary structural predictions. The secondary structural elements from HsIPMK are depicted above its sequence and are color-coded orange for the N-lobe, yellow for the C-lobe, and blue for the IP-helices. Residues are involved in ATP binding are highlighted as magenta for polar contacts and green for Van der Waals interactions. Residues that are involved in Ins(1,4,5)P 3 binding are highlighted in red. The loop that contains three prolines ("3-P"), and the hinge between the N-and C-domains, are also highlighted. PDB codes for HsIPMK are 5W2G, 5W2H, and 5W2I.

Description of the nucleotide-binding region of HsIPMK
We soaked Ins(1,4,5)P 3 , ATP, and magnesium into the apoenzyme crystals. Nucleotide binding did not alter protein conformation. The ADP moiety was observed (Fig. 3A), but not the ␥-phosphate of ATP. A similar result was obtained in an earlier study with ScIPMK (24). Perhaps in our study, the ␥-phosphate was hydrolyzed, or alternately, disordered in the crystal. The latter explanation is feasible, because the terminal phosphate of adenylyl-imidodiphosphate (AMP-PNP) was also the only portion of that non-hydrolyzable ATP analogue that we could not visualize, after it was soaked into the crystal structure with Ins(1,4,5)P 3 . 3 HsIPMK clasps the nucleotide between the N-and C-lobes, which are linked by a hinge that comprises residues Asp 132 to Pro 140 (Figs. 2, A and C, and 3A). The N 1 and N 6 atoms of adenine both make hydrogen bonds with the polypeptide backbone: N 1 contacts the amide nitrogen of Val 133 from the hinge, and N 6 interacts with the carbonyl oxygen of Glu 131 from the N-lobe (Fig. 3A). The ATP-ribose group is loosely confined by several van der Waals contacts with Leu 254 and Ile 384 , plus one hydrogen bond with Asp 144 . The ␣-phosphate of the nucleotide forms a salt bridge with Lys 75 (Fig. 3A). Asp 385 interacts with two magnesium ions to make contact with the ␣and ␤-phosphates of ADP. The particular importance of Asp 385 is reflected in it being part of an Ile-Asp-Phe tripeptide that is conserved throughout the IP-kinase family (Figs. 2D and 3A, and see Ref. 21).
Key residues in the nucleotide-binding pocket of HsIPMK were superimposed upon those of ScIPMK, revealing a high degree of conservation ( Fig. 3B; the same comparison could not be made with AtIPMK, because no nucleotide-bound crystal structures are available). Moreover, data in Fig. 3C show that five of the residues in HsIPMK that make contacts with ADP, namely, Lys 75 , Glu 131 , Val 133 , Asp 144 , and Asp 385 , are also represented in the nucleotide-binding domain of protein kinase A (PKA) (27). Furthermore, a high degree of conservation of the entire N-lobes was revealed by superimposition of the secondary structure elements of the HsIPMK structure upon those in PKA (Fig. 3D): a core r.m.s. deviation of 1.80 Å (51/69 comparable residues). The C-lobes (Fig. 3E) are less conserved: a core r.m.s. deviation of 3.35 Å (78/168 comparable residues). The major structural differences between the C-lobes of the two proteins reflects specialization of the alternative substratebinding pockets. The C-lobe of PKA, and indeed protein kinases in general, contains a greater degree of helical structure, and a wider binding site to accommodate a polypeptide (27, 28) (Fig. 3F). These data confirm and extend the idea (21, 29) that protein kinases and the so-called IP-kinase family share an evolutionary ancestry.

Description of inositol phosphate binding
The successful soaking of Ins(1,4,5)P 3 into HsIPMK crystals (Fig. 2, B and C) has yielded the first description of any inositol phosphate substrate captured in the active site of an IPMK. Simulated annealing omit maps and 2F o Ϫ F c maps clearly identify the inositol ring and individual phosphate groups for Ins(1,4,5)P 3 (Fig. 2B). The 2-5 axis of the Ins(1,4,5)P 3 substrate inserts vertically into a positively charged (at physiological pH) horseshoe-shaped pocket (Fig. 4A) constructed from (in anticlockwise rotation), a short 3 10 helix, the ␣3 helix, and a unique loop that is fabricated from three proline residues (Fig. 4B). Rigidity in the ␣3 helix is enhanced by virtue of a 2.8-Å hydrogen bond between the carboxyl and amine groups in the side chains of Gln 163 and Gln 164 , respectively (Fig. 5A). This constrains the conformations of the two side chains and introduces 3 H. Wang, unpublished data.

Structure of human IPMK
planarity, resulting in a stacking effect between the inositol ring and the ␣3 helix.
The proline loop of HsIPMK (Fig. 4B) is a structural element that is absent from HsIP3K (Figs. 2D and 4C) and both the plant and yeast IPMKs (Fig. 4, D and E, and 5B). In fact, in our IPMK sequence alignment (Fig. 5B), there are gaps in the yeast and plant sequences in the region corresponding to the HsIPMK proline loop. Interestingly, the equivalent Drosophila IPMK sequence is Lys-Pro-Glu, suggesting that a nascent version of this loop is present in this, and perhaps, other invertebrates (Fig. 5B). Arg 82 at the N terminus of the proline-loop makes 3 polar contacts with the 4-and 5-phosphates of Ins(1,4,5)P 3 , indicating its particular importance in substrate binding. Gln 78 , at the N terminus of the proline loop, interacts with two water molecules that coordinate with the ␤-phosphate of ADP and a magnesium atom (Fig. 4B).
We were unable to visualize the ␥-phosphate of ATP within the catalytic center, apparently because it is disordered in the crystal (see above), but the ADP moiety is only 6.3 Å from the 3-OH of Ins(1,4,5)P 3 that is phosphorylated (Fig. 5, A and C). Thus, ATP may phosphorylate the Ins(1,4,5)P 3 substrate by direct, in-line transfer (Fig. 5C). There are also two magnesium ions that are in a position to stabilize the negative charge that would develop on the leaving ␥-phosphoryl group. Additionally, His 388 is also only 7.7 Å from the ␤-phosphate of the ADP moiety, and so it is possible that His 388 may contribute to charge balance during catalysis. His 388 can also hydrogen bond with the 4-phosphate of Ins(1,4,5)P 3 (Fig. 5A).
Other residues that form polar contacts with Ins(1,4,5)P 3 include Lys 160 , Gln 163 , Gln 164 , and Lys 167 from the ␣3 helix, and Gln 196 from the 3 10 helix (Fig. 5A). A contact between Gln 164 and the axial 2-hydroxyl group appears to help locate the inositol ring near-parallel to the ␣3 helix (Figs. 4A and 5A). The interactions of Gln 164 , Lys 167 , and Gln 196 with Ins(1,4,5)P 3 may be particularly important for enforcing its phosphorylation at the 3-position. Thus, we have described a relatively constrained The backbones of the hinge residues are also shown as stick and ball models. B, superimposition of key residues for nucleotide binding between HsIPMK (orange or yellow stick, residues are numbered as in panel A) and ScIPMK (light green stick, with green-colored residues numbers). C, shows conservation of both the nature and the relative positions of the nucleotide-binding residues in HsIPMK (orange, blue, and yellow sticks) and PKA from the rat (white sticks; PDB code 1L3R). PKA-bound nucleotide is depicted in white. The magnesium atoms in HsIPMK are shown in magenta; those from PKA are colored pink. D-F, superimpositions of HsIPMK and PKA. D and E are ribbon plots of the N-and C-lobe, respectively, of HsIPMK (orange, blue, and yellow) and PKA (cyan). F, a surface representation of PKA (white) in complex with the peptide substrate; Ins(1,4,5)P 3 (green carbon stick and ball), from the HsIPMK structure (yellow and blue schematics), is located in the PKA catalytic center.

Structure of human IPMK
catalytic pocket in which a preponderance of Gln residues make contact with substrate. This contrasts with the description of the active site that emerged from the modeling of substrates into the active sites of AtIPMK (25) and ScIPMK (24). The latter two studies described a less-enclosed and conformationallyflexible substrate pocket in which mobile side chains of Lys and Arg residues play major roles in ligand binding (Fig. 5, D and E). Two such residues in particular, Lys 153 and Arg 156 in AtIPMK, were proposed to form key contacts with substrate (25). In a structural alignment (Fig. 5E), the latter two residues correspond to His 197 and Arg 200 in HsIPMK (Fig. 5E), but these do not play any role in substrate-binding in our crystal structures (Fig. 5, A and E).

Structure of human IPMK
protein (Fig. 6C). These comparisons suggest that, compared with Ins(1,4,5)P 3 , IMPK may have a lower binding affinity for Ins(1,3,4,5)P 4 ; this may be the reason that we were unable to soak Ins(1,3,4,5)P 4 into the active site (see above). Differences in binding affinity may also explain why the rate of Ins(1,3,4,5)P 4 phosphorylation is 90-fold slower than that for Ins(1,4,5)P 3 ( Table 2). We do not exclude the possibility that the orientations of the amino acid side chains might be affected by the nature of the bound substrate. Nevertheless, no such movements were necessary for us to model Ins(1,3,4,5)P 4 into the active site, compared with their positions in the Ins(1,4,5)P 3 -bound crystal complex. This situation contrasts with the conclusion that emerged after substrates were modeled into the plant and yeast IPMKs (24,25). In the latter studies, it was proposed that con-formational flexibility was likely an important aspect to accommodating the different substrates within a relatively spacious binding pocket. In those particular IPMK orthologs, such flexibility could be provided by the relatively long and mobile side chains of Lys and Arg (24,25). In contrast, in the case of HsIPMK, the smaller side chains of Gln have a larger role in substrate-binding.

Mutagenesis of HsIPMK
Elements of the nucleotide-binding domain of HsIMPK are well-conserved within the IP-kinase family (Fig. 2D) (21). In contrast, our structural analysis has revealed unique features of the inositol phosphate-binding site, which presumably enforce its own particular set of catalytic activities. We interrogated these new findings using site-directed mutagenesis. We mu-  (1,4,5)P 3 within the catalytic site of HsIMPK (stick and ball model; green stick for carbon, red for oxygen, and orange for phosphorus atoms. The phosphate groups are numbered.). B, multiple sequence alignment of inositol phosphate-binding regions of IPMKs from the indicated organisms. The tri-Pro sequence is given in bold. Residues highlighted in red are those involved in binding Ins(1,4,5)P 3 by HsIMPK. C, significance of His 388 in the catalytic center of HsIMPK (stick and ball model; green stick for carbon, red for oxygen, and orange for phosphorus atoms. D, superimposition of Ins(1,4,5)P 3 -binding residues in HsIPMK (blue stick) upon the aligned residues (see B) in ScIPMK (green stick); E, superimposition of Ins(1,4,5)P 3 -binding residues in HsIPMK (blue stick) upon the aligned residues (see B) in AtIPMK (pink stick), plus two additional AtIPMK residues (Lys 153 and Arg 156 , also pink stick), which have been implicated in Ins(1,4,5)P 3 binding. The latter two residues align with His 197 and Arg 200 in HsIPMK, which do not participate in Ins(1,4,5)P 3 binding; those two HsIPMK residues are shown as a transparent blue stick. Note that Arg 82 in HsIPMK does not have a corresponding residue in either ScIPMK or AtIPMK.

Structure of human IPMK
tated to Ala seven residues that are revealed to interact with Ins(1,4,5)P 3 substrate, together with His 388 at the catalytic center (Table 2). To varying degrees, each of these mutants exhibited reduced activity compared with wild-type enzyme, for both Ins(1,4,5)P 3 phosphorylation, as well as InsP 5 accumulation (Table 2). These mutagenic data validate the conclusions based on structural data (see above) that these particular residues are catalytically important. The H388A mutation had the largest effect, reflecting the critical nature of its role in the catalytic center (Fig. 5C).
We also performed a more subtle mutagenic approach to pursue further the particular significance of the three catalytically-important Gln residues at positions 163, 164, and 196: we mutated each to Arg and Lys, both of which have side chains that are larger and also positively charged at physiological pH. The results are quite dramatic (Table 3): in each case, the rate of Ins(1,4,5)P 3 3-kinase activity declined, but in contrast, the rate of Ins(1,3,4,5)P 4 6-kinase activity was not impaired; in fact, three of these mutants showed increased 6-kinase activity ( Table 3). From our data on Ins(1,4,5)P 3 binding within the crystal complex (Fig. 5A), and our model of Ins(1,3,4,5)P 4 binding (Fig. 6C), we can propose 3 possible explanations for why these mutants exhibit a switch in 3-kinase/6-kinase preferences. First, the 3-kinase activity could be impeded by the mutation of Gln 164 to the larger Lys or Arg residue, because that would provoke a steric conflict with the 2-OH of Ins(1,4,5)P 3 , with which Gln 164 has a favorable interaction in the wild-type enzyme (Fig. 5A). For Ins(1,3,4,5)P 4 binding, the 2-OH is rotated out-of-reach from any residue at position 164, because of the ring flip (Fig. 6C). Moreover, a gain of function of Ins(1,3,4,5)P 4 6-kinase activity could result from this substrate's 6-OH and 3-phosphate groups making contact with the Lys or Arg replacement (Fig. 6C). Second, the relative spatial position of Gln 164 would be perturbed by mutation of Gln 163 , because of elimination of the stabilizing electrostatic connection between the two Gln residues (Fig. 5A). Third, the substitution of Arg or Lys for Gln 196 would sterically disturb the latter's favorable interaction with the proximal 1-phosphate of Ins(1,4,5)P 3 (Fig.  5A); in the Ins(1,3,4,5)P 4 model, the 1-phosphate is rotated further away from Gln 196 , but this gap could be bridged by multiple contacts with the larger Lys or Arg, thereby potentially contributing to a 6-kinase gain of function (Fig. 6C). These data provide a foundation for the generation and utilization of substrate-selective HsIPMK mutations for a synthetic biology approach to understanding each of the individual biological activities of this multifunctional enzyme.

Structural rationalization of PtdIns(4,5)P 2 3-kinase activity
The determination of the position of the 1-phosphate of Ins(1,4,5)P 3 in the substrate pocket is of particular interest for understanding why PtdIns(4,5)P 2 is also a substrate for HsIMPK. This 1-phosphate, which is doubly ionized (21), makes contacts with both Lys 167 and Gln 196 (Fig. 5A). In this configuration, a single uncharged oxygen is exposed to the bulk phase; the esterification of this particular oxygen to a diacyclyglycerol backbone would not be expected to impose any steric hindrance to substrate binding. To pursue that idea, we next soaked a soluble diC 4 -analogue of PtdIns(4,5)P 2 (along with nucleotide and magnesium) into the HsIMPK apoenzyme crystal; the structure of the enzyme co-complex revealed that the Ins(1,4,5)P 3 headgroup of the inositol lipid was oriented in a near-identical configuration to that of free Ins(1,4,5)P 3 (Fig. 7). Furthermore, three mutations that compromised Ins(1,4,5)P 3 3-kinase activity, Q164A, K167A, and Q196A, had quantitatively similar effects upon PtdIns(4,5)P 2 3-kinase activity ( Table  2). These data indicate that there has been co-evolution of Ins(1,4,5)P 3 and PtdIns(4,5)P 2 3-kinase activities.
No electron density was observed for the C 4 -diacylglycerol moiety of the PtdIns(4,5)P 2 analogue, indicating that its mobility is not constrained upon binding to HsIPMK. This leaves the natural diacyglycerol backbone free to embed itself either into membranes, or the hydrophobic pockets of certain proteins (14). The position in the crystal complex of the Ins(1,4,5)P 3 headgroup of diC 4 -PtdIns(4,5)P 2 clarifies that it can make the same contacts with the protein as Ins(1,4,5)P 3 itself, with just the one exception that the diester phosphate of the lipid at position 1 is held less tightly, because it only carries one negative charge, in contrast to the two in Ins(1,4,5)P 3 (30). Indeed, GroPIns(4,5)P 2 is a 5-fold weaker substrate than Ins(1,4,5)P 3 ( Table 2).

Concluding comments
We have described several novel structural features of HsIMPK that clearly distinguish it from the orthologs in Arabidopsis and S. cerevisiae that predominantly phosphorylate the 6-hydroxyl of Ins(1,4,5)P 3 (24,25). First, the horseshoe-shaped catalytic site in the human enzyme is more physically constraining. Second, HsIMPK hosts a smaller substrate-binding

Structure of human IPMK
pocket in which Gln residues play major roles, unlike the plant and yeast orthologs that are more reliant on the longer and more flexible side chains of Arg and Lys. Third, the proline loop is a unique structural feature of the human enzyme that orients Arg 82 and Gln 78 into functionally-important positions within the catalytic pocket. Our results indicate that these are all adaptations that optimize Ins(1,4,5)P 3 phosphorylation predominantly at the 3-position. Our crystal complex data also demonstrate that the Ins(1,4,5)P 3 headgroup of PtdIns(3,4,5)P 3 is near-perfectly superimposed upon free Ins(1,4,5)P 3 itself (Fig.  7). Thus, we propose that 3-kinase positional specificity toward Ins(1,4,5)P 3 has co-evolved along with the functional significance of PtdIns(4,5)P 2 3-kinase activity of HsIPMK. Indeed, AtIPMK does not express PtdIns(4,5)P 2 3-kinase activity (14,31), whereas any PtdIns(3,4,5)P 3 product that might be formed in S. cerevisiae is not considered to be functional (32). In contrast, the biological importance of the PtdIns(4,5)P 2 3-kinase activity of IMPK in mammals has been well demonstrated (14,31,33).

Protein expression and purification
The cDNA of HsIPMK was purchased from Addgene (plasmid number 23666). The Gateway expression system (Invitrogen) was used to subclone HsIPMK into the pDest-566 vector. This vector encodes a His 6 tag, a maltose-binding protein tag, and a tobacco etch virus protease cleavage site at the N terminus. Mutants were prepared using a site-directed mutagenesis kit (Stratagene) or a Q5 site-directed mutagenesis kit (Biolabs); all mutants were verified by sequencing. Each pDest-566 vector was used to transform DE3 competent Escherichia coli cells (Stratagene) that were pre-transformed with chaperone plasmid pGro7 (Takara, Clontech). An overnight culture of the transformed E. coli cells was inoculated into nutrient-rich 2ϫ YT medium supplemented with 0.07% (w/v) L-arabionose at pH 7.5 and grown at 37°C to A595 ϭ 0.7. Isopropyl ␤-D-thiogalactopyranoside (0.1 mM) was then added and cultures were continued at 15°C for 20 h. The cells were disrupted using a con-stant cell disruption system (Constant Systems LTD) under 20 KPsi. Recombinant wild-type and mutant proteins were purified by several chromatographic procedures performed at 4°C. First, the protein was applied to a nickel-nitrilotriacetic acidagarose column (Qiagen), washed with buffer containing 300 mM NaCl, 20 mM Tris-HCl, pH 7.5, 20 mM imidazole, then eluted by increasing the imidazole concentration to 400 mM. Second, the eluate was applied to a HiTrap TM Heparin HP column (GE Healthcare) and eluted with 10 column volumes of a 50 -2000 mM NaCl gradient in 20 mM Tris-HCl (pH 7.5). Next, after tobacco etch virus protease cleavage, the protein was further purified using another HiTrap TM Heparin HP column, and finally, a Superdex TM 200 gel filtration column (GE Healthcare) that was eluted with 150 mM NaCl, 20 mM Tris-HCl, pH 7.5. Purified proteins were concentrated to either 0.5-2 mg/ml (for assaying catalytic activities) or 30 mg/ml (for crystallization); storage was at Ϫ80°C.

Crystallization
The core catalytic domain of HsIPMK (residues: 50 -262 Gly-Gly-Ser-Gly-Gly 378 -416 ; Fig. 2A) was initially screened for optimum crystallization conditions using the mosquito-LCP (TTP Labtech). Multiple conditions that each contained a high concentration of PEG 400 were identified. Final conditions were optimized by hanging drop vapor diffusion, against a well buffer of 35% (w/v) PEG 400, 0.1 M Li 2 SO 4 , 100 mM MES-imidazole buffer, pH 6.0, 50 mM ␤-mercaptoethanol at 25°C (2 l of 38 mg/ml of protein plus 2 l of well buffer in the crystallization drop). To obtain complex structures, apoenzyme crystals were further soaked for 1 day in 35% (w/v) PEG 400, 100 mM Li 2 SO 4 , 100 mM HEPES, pH 7.5, at 25°C, in the presence of 20 mM Ins(1,4,5)P 3 or a soluble diC 4 -analogue of PtdIns(4,5)P 2 , 10 mM MgCl 2 , and 5 mM of either Na 2 ATP or Li 2 AMP-PNP; mother liquid was used for cryoprotection. The same conditions were used in attempts to soak Ins(1,3,4,5)P 4 into the crystals, in the presence of either ATP or ADP, but we were unable to define any clear electron density for the inositol phosphate.

Data collection, structure determination, and refinement
Diffraction data were collected using APS beamlines 22-ID and 22-BM. All data were processed with the program HKL2000 (34). Initial phases for the structure were determined by molecular replacement with the autoMR program in the CCP4 package (35,36), using ScIPMK structure (PDB code 2IF8; sequence identity 31%) as a search model. This initial structure was manually rebuilt with COOT and refined with REFMAC from the CCP4 package. The other crystal structures were determined by using rigid body and direct Fourier synthesis, and refined with the equivalent and expanded test sets. The molecular graphics representations were prepared with the program PyMol (Schrödinger, LLC). Atomic coordinates and structure factors have been deposited with the Protein Data Bank with accession codes 5W2G, 5W2H, and 5W2I.