Integration of Inositol Phosphate Signaling Pathways via Human ITPK1*

Inositol 1,3,4-trisphosphate 5/6-kinase (ITPK1) is a reversible, poly-specific inositol phosphate kinase that has been implicated as a modifier gene in cystic fibrosis. Upon activation of phospholipase C at the plasma membrane, inositol 1,4,5-trisphosphate enters the cytosol and is inter-converted by an array of kinases and phosphatases into other inositol phosphates with diverse and critical cellular activities. In mammals it has been established that inositol 1,3,4-trisphosphate, produced from inositol 1,4,5-trisphosphate, lies in a branch of the metabolic pathway that is separate from inositol 3,4,5,6-tetrakisphosphate, which inhibits plasma membrane chloride channels. We have determined the molecular mechanism for communication between these two pathways, showing that phosphate is transferred between inositol phosphates via ITPK1-bound nucleotide. Intersubstrate phosphate transfer explains how competing substrates are able to stimulate each others' catalysis by ITPK1. We further show that these features occur in the human protein, but not in plant or protozoan homologues. The high resolution structure of human ITPK1 identifies novel secondary structural features able to impart substrate selectivity and enhance nucleotide binding, thereby promoting intersubstrate phosphate transfer. Our work describes a novel mode of substrate regulation and provides insight into the enzyme evolution of a signaling mechanism from a metabolic role.

classic example of these signaling activities is the release of Ca 2ϩ into the cytoplasm through an intracellular channel that is gated by inositol 1,4,5-trisphosphate (Ins(1,4,5)P 3 ) 4 (2). Additional roles are continually being discovered: inositol phosphates have recently been shown to be critical to the activity of RNA-editing enzymes (3), to participate in telomere maintenance (4), and to be the phosphate donors in certain protein phosphorylation events (5). It is of critical interest, therefore, to establish the regulatory mechanisms that govern the metabolism of inositol phosphates.
Recently, Miller et al. (15) described the crystal structure of an ITPK1 homologue in Entamoeba histolytica (eITPK1). This structure revealed an unusually versatile catalytic cleft that possesses little or no stereospecific constraints. Ligand binding was further proposed to be unaffected by either the presence of hydroxyl groups or by their orientation (i.e. equatorial versus axial); up to 18 inositol phosphate analogues could be modeled into the active site (15). Most of these characteristics of eITPK1 were proposed to apply to Homo sapiens ITPK1 (hITPK1) (15).
However, other data indicated that there are limitations in the extent to which this information can be exploited to improve our molecular understanding of the human enzyme. For example, the ligand specificity of hITPK1 depends on the stereochemistry of the inositol ring and the orientation of its hydroxyl groups (16). Most important of all, the eITPK1 structure did not offer any rationalization for how Ins(1,3,4)P 3 can stimulate Ins(1,3,4,5,6)P 5 dephosphorylation by mammalian ITPK1. In fact, in the current study we demonstrate that this characteristic has evolved specifically in higher organisms; we show that Ins(1,3,4)P 3 does not activate Ins(1,3,4,5,6)P 5 dephosphorylation in ITPK1 homologues from soybean and in E. histolytica. Moreover, we describe the crystal structure of the human enzyme and identify structural features likely to contribute to this and other important differences in the catalytic activities of different homologues of ITPK1.

EXPERIMENTAL PROCEDURES
A rapid Escherichia coli expression, purification, and crystallization procedure for hITPK1 has been developed. Human ITPK1 was cloned into an MH4 vector (17) by PCR and expressed in the HK100 strain of E. coli. Briefly, 12 ϫ 65-ml cultures were grown at 37°C in a Genomics Institute of the Novartis Research Foundation fermenter (18), and hITPK1 expression was induced by the addition of 0.02% L-arabinose for 3 h. Proteins were extracted by sonication into 20 ml of 50 mM Tris, pH 8, 100 mM NaCl, 1 mM tri(2-carboxyethyl)phosphine hydrochloride (TCEP), 10 mM imidazole, with 1 Complete EDTA-free protease inhibitor tablet (Roche Applied Science). Cell debris was removed by centrifugation at 30,000 ϫ g for 45 min. The recombinant enzyme was purified by nickel-nitrilotriacetic acid-agarose chromatography and eluted into a buffer composed of 20 mM Tris, pH 8, 100 mM NaCl, 10% glycerol, 0.5 mM TCEP, 100 mM imidazole. S-200 size exclusion chromatography followed. Selenomethionine-substituted protein was produced by inhibition of methionine biosynthesis (19) and subsequently purified in the same manner as the native protein.
Protein was concentrated in a Centriprep 15 10-kDa concentrator (Amicon) prior to crystal screening. Native protein was concentrated to 15-22 mg/ml for crystallization trials, whereas selenomethionine protein appeared insoluble beyond 12 mg/ml and was used at this concentration. 480 crystallization conditions were set up in Greiner low profile 96-well sitting drop vapor diffusion plates at two temperatures (4 and 20°C) (conditions described in Ref. 17). Full-length ITPK1 failed to crystallize despite extensive coarse crystallization screening and showed some proteolytic degradation during purification and storage. Proteolytic products, presumably a consequence of the action of endogenous E. coli proteases, were analyzed by mass spectroscopy, and sequences corresponding to the truncated products were cloned back into the MH4 vector. The construct composed of residues 1 to 335 proved amenable to crystallization and was used in subsequent structural studies. Crystals grew in 2 weeks at 4°C after mixing the protein solution 1:1 with, and subsequently equilibrating against, a reservoir solution of 0.2 M citrate, pH 5.6, 2.0 M (NH 4 ) 2 SO 4 , 0.2 M potassium/sodium tartrate. Further crystals were obtained by screening around this condition. Nucleotide complexes were obtained by co-crystallization with either 1 mM ATP or AMPPNP (Sigma), along with 5 mM MnCl 2 . Crystals were cryo-protected in reservoir solution containing 20% glycerol. Data were collected at the Advanced Light Source (ALS, Berkeley, CA) beamlines 5.0.2 and 5.0.3. Data were integrated, reduced, and scaled using HKL2000 (20) and the CCP4i suite (21). Crystallographic statistics are summarized in Table 1. ITPK1 was initially phased using a three-wavelength multiwavelength anomalous dispersion experiment on selenomethionine-derivative protein, and the structure was subsequently solved using Solve/Resolve (22). Subsequently, non-isomorphous structures were solved by molecular replacement using Phaser (23). Iterative rounds of manual model building in Coot (24) were followed by restrained refinement using Refmac5 (25). Ligand solvent accessibility was calculated using AREAIMOL (26) with a probe radius of 1.4 Å. The protein-protein interaction server was used to evaluate the hITPK1 crystal interactions (27). Figures were prepared using PyMOL.
All in vitro assays of inositol phosphate metabolism were performed as previously described (28). For these experiments, recombinant hITPK1 was prepared after expression in Sf9 cells as previously described (28) except that the gene was first shuttled into the pDEST605 kinase vector, which also contains a green fluorescent protein reporter gene that was then used to generate bacmid insect viral DNA (Invitrogen). Site-directed mutagenesis was performed as previously described (28).
An expressed sequence tag clone encoding gmITPK4 was obtained from the Public Soybean Data Base (Biogenetic Services, Brookings, SD). PCR amplifications were performed with a 5Ј-primer containing a SmaI restriction site and the 3Ј-primer containing XhoI. PCR products were digested and ligated into pGEX4T-1 (encoding an N-terminal glutathione S-transferase tag), which was used to transform competent TOP10 bacterial cells. Protein expression was induced with isopropyl-1-thio-␤-D-galactopyranoside, cells were lysed, and protein was purified using glutathione S-transferase-Sepharose
The assays described above contained a physiologically relevant concentration of ADP (1 mM) (30) but no ATP. In further experiments, non-radioactive ATP was added to 5 mM and it is striking that there was only a 30% decrease in both the extent of Ins(1,3,4,5,6)P 5 dephosphorylation and the rate of [ 32 P] transfer from Ins(1,3,4,5,6)P 5 to Ins(1,3,4,6)P 4 (Fig. 1E) compared with incubations containing 1 mM ADP and no ATP. These results support our hypothesis that there is a relatively slow rate of exchange of bulk phase ATP with either enzyme-bound ADP or enzyme-bound [ 32 P]ATP. Clearly, the intersubstrate phosphate transfer activity of ITPK1 occurs even in the presence of physiologically relevant levels of adenine nucleotides (i.e. [ATP] Ͼ [ADP]) (30).
Simultaneous binding of both Ins(1,3,4)P 3 and Ins(1,3,4,5,6)P 5 to the same active site seems implausible because of the topological and electrostatic constraints imposed by the binding pocket (described below). In order for the same phosphate to be transferred between different substrates via the same binding site, a sequential reaction model has been proposed ( Fig. 2A). The efficient operation of this proposed nucleotide-mediated FIGURE 2. Sequential intersubstrate phosphate transfer hypotheses for hITPK1. This graphic shows hypothetical intersubstrate phosphate transfer mechanisms by which levels of Ins(1,3,4)P 3 could regulate the synthesis of Ins(3,4,5,6)P 4 . Panel A shows the two candidate phosphate carriers with enzyme represented as "E" with the transferred phosphate highlighted in red. Panel B expands the nucleotide-mediated phosphate transfer hypothesis in which enzyme-bound nucleotide acts as the phosphate carrier. Unliganded ITPK1 is represented by E; phosphotransferase reactions are indicated by red graphics. The position of the [ 32 P] group that is transferred between inositol phosphates is shown in red. An asterisk indicates a reaction for which Ins(1,3,4,5)P 4 is also produced at a reduced rate. We do not rule out the possibility that a phosphoenzyme intermediate might also participate in these reactions.
intersubstrate phosphate transfer may be facilitated by the enzyme withholding the nucleotide from the bulk phase during this series of reactions (Fig. 2B).

)P 3 Does Not Stimulate This
Reaction-Although the inositol phosphate kinase activity of eITPK1 has been well documented in previous reports (15,31), the data shown in Fig. 3A represent the first demonstration that eITPK1 can dephosphorylate Ins(1,3,4,5,6)P 5 to Ins(3,4,5,6)P 4 . It is particularly notable that the rate of Ins(1,3,4,5,6)P 5 dephosphorylation was not stimulated by Ins(1,3,4)P 3 (Fig. 3A). This is an important functional difference between eITPK1 and hITPK1 and may represent a divergence in cellular function. Homologues of hITPK1 are also present in plants (32). Plant enzymes have higher homology with hITPK1 than does eITPK1 (Fig. 4C). The plant enzymes are also functionally more similar to hITPK1 in that neither show physiologically significant kinase activity toward Ins(1,4,5)P 3 ( Table 2 and Ref. 32), unlike eITPK1 which shows robust Ins(1,4,5)P 3 3-kinase activity (31). We therefore felt useful information could be obtained by comparing some of the characteristics of hITPK1 with a plant homologue.
Structural Overview of hITPK1 -To rationalize the species differences in substrate regulation, the crystal structure of hITPK1 was determined. The structure of hITPK1 was initially solved by selenomethionine multiwavelength anomalous dispersion and has been subsequently refined in a variety of ligand-bound forms and crystal forms to a maximum resolution of 1.6 Å ( Table 1). The entire catalytic domain is defined as far as residue 335. In both crystal forms the C-terminal residues form a small dimer interface. Although this interface corresponds to ϳ10% of the total surface area of the protein, other biophysical methods such as gel filtration and static light scattering indicate a monomeric state for ITPK1 (data not shown). hITPK1 exhibits the same overall topology as the eITPK1 structure, which is related to the former by 24% amino acid sequence identity between the catalytic domains (Fig. 4C). Superposition of the hITPK1 structure with eITPK1 results in a root mean square deviation of 2.2 Å on 242 topologically equivalent C␣ atoms (Fig. 4A). As would be expected for enzymes with variations in enzymatic function, structural diversity is exhibited in regions proximal to the active site.
ATP Binding Site-The ATP binding sites of eITPK1 and hITPK1 are each sandwiched between two sets of fourstranded anti-parallel ␤-sheets (Fig. 4A), a structure commonly described as an ATP-grasp fold (34,35). In hITPK1, one ␤-sheet is composed of ␤11, ␤10, ␤13, and ␤14; the opposed sheet is composed of ␤12, ␤8, ␤7, and ␤9 (Fig. 4, A and C). A further two extended strands provide contacts to the purine ring of the bound nucleotide. Two well defined Mn 2ϩ ions are present in the active site of the hITPK1 structure. One Mn 2ϩ ion is coordinated by Asp-281 and Asp-295 and interacts with oxygens from both the ␤and ␥-phosphates of the bound ATP, adjacent to the scissile bond. The second Mn 2ϩ ion is coordinated by the side chains of Asp-295 and Asn-297. These Mn 2ϩ ions are structurally equivalent to the Mg 2ϩ ions in the eITPK1 structure (15).
It was previously noted that the nucleotide binding site of eITPK1 encloses much of the ligand; only 22% was accessible to solvent (15). There is only one entry and exit site for nucleotides in both eITPK1 and hITPK1 structures that would not require considerable rearrangement of the protein backbone, and this opening occurs next to active site residues Asp-295 and Asn-297. In eITPK1, one face of the entry/exit site is formed by a turn in the polypeptide chain just after the conserved structural element ␤12 and is followed by two helices (eITPK1 residues 212 to 236) arranged anti-parallel to each other and oriented roughly parallel to ␤12 (Fig. 4B). However, the structure of hITPK1 diverges after ␤12, with a 14-residue deletion relative to the eITPK1 sequence. Instead, hITPK1 substitutes a 3 10 helix (3 10 4), composed of residues 232 to 236, in a lateral orientation to the ATP phosphates (Fig. 4B). The two crystal forms (Table  1) each contain two unique hITPK1 molecules in the asymmetric unit, and in each crystal form these molecules vary in the degree to which helix 3 10 4 encloses the nucleotide, indicating flexibility within this motif. In the conformation in which the nucleotide is most enveloped, the side chain of His-233 is able to reach hydrogen bonding distance to the AMPPNP ␥-phosphate, although the equivalent distance is 6 Å in the second molecule, beyond interacting range. Nevertheless, in both cases the entry/exit site for nucleotide is more restricted for hITPK1 than eITPK1, with solvent accessibilities of only ϳ5% for AMPPNP (Fig. 5, A and B).
The 3 10 4 helix also offers several extra interactions with the bound nucleotide that are not conserved in eITPK1. These include a bond from Ser-236 to the ADP-ribose o3 hydroxyl (2.7 Å, Fig. 4B), and extra Van der Waals interactions from Val-235 to the ribose ring (4.2 Å, Fig. 4B). hITPK1 Ser-232 makes a tight polar contact with the ␤-phosphate (2.5 Å) and also approaches the side chain of His-167 (3.4 Å). Note also that the helix 3 10 4 in hITPK1 aligns with a sequence in gmITPK4 that contains a proline residue (Fig. 4C), which would be expected to disrupt the intra-helix hydrogen bonding and geometry.
Inositol Phosphate Binding Site of hITPK1-A particularly interesting feature of the eITPK1 structure is that there are no direct interactions between inositol hydroxyl groups and the protein, reflecting a limited capability for stereochemical substrate discrimination (15). In contrast, experiments with hITPK1 indicate a much greater ability to discriminate between substrates, including those that are stereoisomers (16). The inositol phosphate binding site of hITPK1 can be visualized as a strongly electropositive cleft that is stationed at the entrance to the ATP binding pocket (Fig. 5). The hITPK1 inositol phosphate binding site is composed of residues from helices ␣1, ␣3, and 3 10 4, Arg-212 from sheet ␤11, and loop regions between ␤14-␣9 and ␤7-␤8 (Fig. 4, A and C). Despite repeated attempts, we have been unable to co-crystallize or soak inositol phosphates into the hITPK1 crystals, possibly due to the highly polar nature of the binding site in combination with the presence of ammonium sulfate in the crystallization buffer. Sulfate molecules are, in fact, bound in the inositol phosphate binding site (Fig. 5C).
In comparing the inositol phosphate binding sites of hITPK1 and eITPK1, structural rearrangements and amino acid substitutions contribute to a considerable reduction in binding pocket volume for hITPK1 (Fig. 5, A and B). Helix 3 10 4 in hITPK1 is followed by a loop composed of residues Ser-236 to Ser-258 where the eITPK1 binding site is open to solvent. The consequences of this loop are the further encapsulation of the bound nucleotide, the constriction of the inositol phosphate binding site, and the introduction of residues capable of mediating novel interactions with the bound inositol phosphate. For example, Lys-237 reaches from this loop into the inositol binding pocket where it would be able form a bond with either a phosphate group or with a hydroxyl group (Fig. 5, B and C). From the opposite face of the inositol phosphate binding cleft, two additional lysine residues, Lys-17 and Lys-21, both project from the ␣1 helix into the inositol phosphate binding pocket (Fig. 5, B and C). Lys-21 is a substitution for Thr-20 in eITPK1 and in hITPK1 could form a bond with a phosphate or hydroxyl group on an inositol ring modeled into the binding site. Lys-17 is also capable of interacting with inositol phosphate groups (Fig. 5C). A kink in the loop formed from residues Ile-296 to FIGURE 4. Sequence and structural comparisions between hITPK1 and eITPK1. Panels A and B show a structural alignment of hITPK1 in complex with AMPPNP (blue) with eITPK1 (orange). Panel B highlights topological differences proximal to the ATP binding site. Selected hITPK1 residues are shown in stick representation and selected hydrogen bonds as broken lines. AMPPNP is from the hITPK1 structure and is shown in stick form, with carbons shown in yellow. Panel C shows a sequence alignment of hITPK1, eITPK1, and gmITPK4. Secondary structural elements from hITPK1 are shown in blue above the alignment, and eITPK1 secondary structure is shown in orange below the alignment. aX, ␤X, and 3 10 X refer to ␣ helices, ␤-sheets, and 3 10 helices, respectively, numbered in order from the N terminus. The position of His-162 is indicated by an asterisk.

TABLE 2 Catalytic comparison between WT and H162D mutant of hITPK1
Assays were performed as described under "Experimental Procedures." The units for the kinase reactions are 10 3 ϫ k (the first-order rate constant)/ng protein.

Reaction WT H162D
Ins ( Glu-307, and a substitution from a glycine in eITPK1 to an glutamate at position 303 in hITPK1, closes one further face of the inositol binding site (Fig. 5, B and C).
A further reduction of the inositol phosphate binding pocket occurs as a consequence of structural changes at hITPK1 ␣3 and the loop between ␤7 and ␤8 (Fig. 4, A and C) that contains His-162 (Fig. 5, B and C). Helix ␣3 is composed of residues 60 -69 and corresponds to a loop region in eITPK1. In hITPK1, Asp-62 protrudes from ␣3 and forms a tight polar interaction with His-162 from the loop between ␤6 and ␤7 (2.6 Å, Fig. 5C). The side chain of His-162 is positioned over the side chain amine of residue Lys-59 at a distance of 3.4 Å, which lines the inositol phosphate binding site below the imidazole ring. Interestingly, when Ins(1,3,4)P 3 is modeled into hITPK1 in the same binding mode as eITPK1, the imidazole ring of His-162 clashes with the 4-phosphate group of the ligand (Fig. 5, B and C). His-162 is equivalent to Gln-141 in eITPK1, which also hydrogen bonds with the 4-phosphate of Ins(1,3,4)P 3 (15). The conformation of ␣3 in hITPK1 leads to a shift of 5.0 Å for Thr-61 (equivalent to Thr-59 in eITPK1) toward the inositol phosphate binding site and within 3.6 Å of the imidazole ring of His-162. Thus, a steric barrier exists preventing His-162 from accommodating Ins(1,3,4)P 3 in the manner of eITPK1 and indicating that rearrangements of protein and/or substrate are necessary to facilitate binding.
His-162 in hITPK1 Contributes to Intersubstrate Phosphate Transfer-Ins(1,3,4)P 3 stimulated Ins(1,3,4,5,6)P 5 dephosphorylation by hITPK1, but this was not observed for either eITPK1 or gmITPK4 (Fig. 3). Based on comparative structural and sequence analysis, residues were identified that may contribute to this phenomenon. His-162 and His-233 emerged as especially interesting candidates as in our comparative analysis both are unique to hITPK1 (Fig. 4C). Our structural studies show that His-162, and its polar interaction with Asp-62, makes a significant contribution to the geometry of the substrate binding cleft (Fig. 5, B and C). His-233 is present on helix 3 10 4 and protrudes into the substrate binding site (Fig. 5, B and C). However, the ability of Ins(1,3,4)P 3 to enhance Ins(1,3,4,5,6)P 5 dephosphorylation was preserved in a H233S mutant of hITPK1 (data not shown). In contrast, when His-162 in hITPK1 was mutated to its equivalent residue in the soybean enzyme, Asp, Ins(1,3,4)P 3 did not stimulate Ins(1,3,4,5,6)P 5 dephosphorylation (Table 2). This loss-of-function for Ins(1,3,4)P 3 in the H162D mutant was matched by a 200-fold reduction in its Ins(1,3,4)P 3 kinase activity (Table 2), yet the H162D mutant dephosphorylated Ins(1,3,4,5,6)P 5 almost twice as fast as the wild-type enzyme ( Table 2). In previous work we found that the ability to dephosphorylate Ins(1,3,4,5,6)P 5 was compromised whenever kinase activity was reduced (28), making H162D unique ( Table 2). The H162D mutant also displayed very limited ability to transfer phosphate from Ins(1,3,4,5,6)P 5 to Ins(1,3,4)P 3 in the presence of physiological concentrations of nucleotide (Table 2). Instead, most of the phosphate removed from Ins(1,3,4,5,6)P 5 was trapped in ATP and released into the bulk phase (Table 2; data not shown). The position of His-162 on a loop forming part of both the inositol phosphate binding site and the nucleotide binding site indicates that a role in substrate binding or catalysis is plausible. Our data indicate that the structural motif featuring His-162 plays an important role in the intersubstrate phos- phate transfer activity of hITPK1, in part by promoting the phosphorylation of Ins(1,3,4)P 3 and also regulating ATP exchange with the bulk phase.
Concluding Comments-This study provides a molecular basis for understanding how hITPK1 couples steady-state levels of Ins(3,4,5,6)P 4 to a metabolically separate pool of inositol phosphates, including Ins(1,4,5)P 3 and Ins(1,3,4)P 3 , that arise from receptor-regulated phospholipase C activity (9). During metabolic coupling, a phosphate is transferred between inositol phosphates by hITPK1, but not by eITPK1 or gmITPK4. Although our studies reveal that ADP does act as a phosphate carrier, we do not exclude the possibility that a phosphoenzyme intermediate might also play a role in the catalytic cycle. hITPK1 residues capable of carrying a phosphate and that are proximal to the ATP ␥-phosphate are shown in yellow in Fig. 5C.
We propose that these differences between ITPK1 homologues represent the evolution in mammals of a mechanism for regulating the signaling activities of Ins(3,4,5,6)P 4 . This function may be superimposed upon a more general role of ITPK1 homologues in eukaryotes: contributing to the synthesis of higher inositol polyphosphates such as InsP 6 (33,37). However, the extent to which ITPK1 contributes to higher order inositol phosphate synthesis in mammals remains to be demonstrated in a range of conditions and cell types. For example, it has been shown that two other inositol kinases, IPK1 and IPK2, are sufficient to recapitulate InsP 6 synthesis from Ins(1,4,5)P 3 (38,39). In contrast, the function of gmITPK4 in soybean is likely restricted to the synthesis of higher inositol polyphosphates. Furthermore, Ins(1,3,4)P 3 has not been identified in any plant, nor has an enzyme that can synthesize it.
It seems likely the capacity for inositol phosphate signal integration has co-evolved as Ins(3,4,5,6)P 4 developed into an intracellular signal that, through its control over Cl Ϫ channel conductance, regulates many biological processes, including epithelial salt and fluid secretion, insulin secretion, and likely smooth muscle contraction and neurotransmission. Together, the biochemical and structural studies on ITPK1 provide an example how evolution can transform a metabolic enzyme into one with a signaling function. The characterization of the structure of hITPK1 and its mechanism of substrate regulation together offer new avenues for intervention in the inositol phosphate metabolic pathway.