Crystal Structure of Inositol Phosphate Multikinase 2 and Implications for Substrate Specificity*

Inositol polyphosphates perform essential functions as second messengers in eukaryotic cells, and their cellular levels are regulated by inositol phosphate kinases. Most of these enzymes belong to the inositol phosphate kinase superfamily, which consists of three subgroups, inositol 3-kinases, inositol phosphate multikinases, and inositol hexakisphosphate kinases. Family members share several strictly conserved signature motifs and are expected to have the same backbone fold, despite very limited overall amino acid sequence identity. Sequence differences are expected to play important roles in defining the different substrate selectivity of these enzymes. To investigate the structural basis for substrate specificity, we have determined the crystal structure of the yeast inositol phosphate multikinase Ipk2 in the apoform and in a complex with ADP and Mn2+ at up to 2.0Å resolution. The overall structure of Ipk2 is related to inositol trisphosphate 3-kinase. The ATP binding site is similar in both enzymes; however, the inositol binding domain is significantly smaller in Ipk2. Replacement of critical side chains in the inositolbinding site suggests how modification of substrate recognition motifs determines enzymatic substrate preference and catalysis.

Inositol polyphosphates perform essential functions as second messengers in eukaryotic cells, and their cellular levels are regulated by inositol phosphate kinases. Most of these enzymes belong to the inositol phosphate kinase superfamily, which consists of three subgroups, inositol 3-kinases, inositol phosphate multikinases, and inositol hexakisphosphate kinases. Family members share several strictly conserved signature motifs and are expected to have the same backbone fold, despite very limited overall amino acid sequence identity. Sequence differences are expected to play important roles in defining the different substrate selectivity of these enzymes. To investigate the structural basis for substrate specificity, we have determined the crystal structure of the yeast inositol phosphate multikinase Ipk2 in the apoform and in a complex with ADP and Mn 2؉ at up to 2.0 Å resolution. The overall structure of Ipk2 is related to inositol trisphosphate 3-kinase. The ATP binding site is similar in both enzymes; however, the inositol binding domain is significantly smaller in Ipk2. Replacement of critical side chains in the inositolbinding site suggests how modification of substrate recognition motifs determines enzymatic substrate preference and catalysis.
Based on sequence homology, IP kinases are classified into three families, inositol 5/6-kinases, inositol 2-kinases, and the IPK superfamily, which includes three subgroups: inositol 3-kinases (IP3Ks), inositol multikinases (IPMKs), and inositol hexakisphosphate kinases. Members of this superfamily share several strictly conserved signature motifs with each other and are predicted to assume the same overall fold (16), although their sequence conservation is very low (Fig. 1).
Yeast inositol multikinase Ipk2 was cloned and characterized by two groups (16,23,26). Knock-out experiments showed that Ipk2⌬ cells are viable when grown at 30°C but exhibit a drastic reduction of cellular inositol 1,3,4,5-tetrakisphosphate, inositol 1,3,4,5,6-pentakisphosphate, and inositol hexakisphosphate levels, together with a 170-fold increase of InsP 3 concentration, demonstrating that Ipk2 function is essential for production of higher phosphorylated InsPs in yeast (26). Yeast Ipk2 seems to perform a unique second function among the IPMKs in that it had previously been characterized under the name ArgRIII (or Arg 82 ) as part of a transcriptional regulatory complex in arginine metabolism (27). The main function of Ipk2 in this transcriptional complex appears to be the stabilization of the transcription factors ArgRI and Mcm1 from degradation (28). Protein-protein interaction is mediated by a unique inserted region containing a stretch of 15 aspartates in 16 residues (poly-D loop) that is required for the interaction with ArgRI and Mcm1 (28). The poly-D loop insertion is not conserved in the mammalian IPMKs, and this second function might be restricted to the Saccharomyces cerevisiae enzyme. A, sequence diagram of IPK signature motifs and insert regions for the three IPK subgroups. The N-terminal domain, the IP-binding domain and the C-terminal domain are colored in pink, yellow, and cyan, respectively. Positions of insert regions are indicated in darker colors, and the length of the inserted region for each protein is indicated below the insert box. B, structure-based sequence alignment of Ipk2 with IP3K, human IPMK, and human inositol hexakisphosphate kinase 1 (IP6K1). Residues conserved in all four sequences are colored in dark blue, and residues conserved in two or three sequences are colored in cyan. Secondary structure elements of Ipk2 are indicated above the alignments. Strictly conserved signature motifs involved in cofactor binding are indicated with blue boxes. Initial alignments were performed with ClustalW, the alignment between Ipk2 and IP3K was manually adjusted based on structural analysis. Accession numbers were as follows: Ipk2: S. cerevisiae Ipk2, NP_010458; Ip3ka_hs: human IP3K-A, P23677; IPMK_hs: human IPMK, NP_689416; IP6K1_hs: human IP6K1, AAH12944. able for the IPMK subgroup. To further investigate the structural determinants for catalytic activity and substrate specificity of inositol phosphate multikinases, we have determined the crystal structure of Ipk2 in the native form and in a complex with ADP and Mn 2ϩ . This is the first structure of an IPMK family member. The overall structure is related to IP3K, despite the low sequence identity of 17.7% between both enzymes. Structural comparison with the highly specific inositol 3-kinase highlights critical binding interactions in the active site, which are consistent with substrate binding in several possible orientations as required for the Ipk2 multikinase catalytic activity.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, and Crystallization-Fulllength Ipk2 from S. cerevisiae was cloned from genomic DNA into the pET26b vector (Novagen). The expression construct contained an C-terminal hexahistidine tag. Protein was overexpressed in Escherichia coli strain BL21Star (Invitrogen) at 20°C. Bacterial cells were lysed by ultrasonification on ice. The soluble protein was bound to nickel-agarose affinity resin (Qiagen), washed with a buffer containing 20 mM Tris (pH 8.5), 250 mM NaCl, and 10 mM imidazole. His-tagged protein was then eluted with a buffer containing 20 mM Tris (pH 8.5), 250 mM NaCl, and 150 mM imidazole. The protein was further purified with anion exchange chromatography at pH 8.5, using a linear gradient of 10 mM to 1 M NaCl concentration, and size exclusion chromatography at pH 8.5 and 200 mM NaCl. The purified protein was concentrated to 25 mg/ml in a buffer containing 10 mM Tris (pH 8.5), 20 mM NaCl, and 7% glycerol. The sample was flashfrozen in liquid nitrogen and stored at Ϫ80°C. The C-terminal His tag was not removed for crystallization.
For the production of selenomethionyl proteins, the expression construct was transformed into B834(DE3) cells (Novagen). The bacterial growth was carried out in defined LeMaster medium (32), and the protein was purified using the same protocol as for the wild-type protein.
Crystals of Ipk2 were obtained at 20°C with the sitting drop vapor diffusion method. The reservoir solution contained 100 mM HEPES, pH 7.7, 200 mM CaCl 2 , and 28% polyethylene glycol 400 (w/v). The crystals were cryoprotected by rapid soaking in a solution containing mother liquor with the addition of 20% (v/v) ethylene glycol and flash-frozen in liquid nitrogen. Crystals grew to maximum dimensions of 0.45 ϫ 0.12 ϫ 0.12 mm.
Data Collection-X-ray diffraction data were collected on an ADSC CCD detector at the X4A beamline of the National Synchrotron Light Source in Brookhaven. For the initial structure determination, a selenomethionyl single wavelength anomalous diffraction data set to 3.1 Å resolution was collected at a wavelength of 0.9790 Å at 100 K. The diffraction images were processed and scaled with the HKL package (33). The crystals belong to the space group P6 3 , with cell dimensions of a ϭ b ϭ 185.6 Å and c ϭ 50.1 Å. There are two molecules in the asymmetric unit, giving a V m of 3.1 Å 3 /dalton. To obtain higher resolution data, a native data set was collected on a larger crystal to 2.0 Å resolution. Cell dimensions of the native crystal were a ϭ b ϭ 186.5 Å and c ϭ 50.1 Å. Diffraction data for the ATPbound form were collected to 2.4 Å resolution from a crystal that had been soaked in a solution of 40 mM ATP and 20 mM MnCl 2 in mother liquor for 15 min. The data processing statistics are summarized in Table 1.
Structure Determination and Refinement-The locations of selenium atoms were determined with the program Solve (34), based on the anomalous differences in the single wavelength anomalous diffraction data set. Reflection phases to 3.1 Å were calculated with Solve, and phases were extended to 2.0 Å resolution of the native data set with DM (35) from the CCP4 program suite (36). After phase extension and noncrystallographic symmetry averaging, the resulting experimental electron density map was of excellent quality. The initial atomic model was built with the program Arp/wArp (37). Six of the eight residues of the His tag are visible in molecule A, and two residues are visible in molecule B. The initial protein model was rebuilt manually with the program O (38). The initial structure refinement was carried out with the program CNS (39). After simulated annealing refinement, the last stages of crystallographic refinement were performed with the program Refmac without noncrystallographic symmetry restraints (40). The structure refinement statistics are summarized in Table 1. All molecular figures were produced with the program PyMOL (41). Molecular surfaces were generated using Gromacs (42) and Mead (43).

RESULTS
Overall Structure of Ipk2-We have determined the structure of the full-length inositol multikinase Ipk2 from yeast in space group P6 3 at 2.0 Å resolution by single anomalous dispersion methods of selenomethionyl-substituted protein ( Table 1). The crystallographic R/R free factors are 20.8/25.3 and 19.5/24.6% for the native and the ADP-complex structure, respectively. The majority of residues (93.4 and 91.9%) are in the most favored region of the Ramachandran plot. There are two molecules in the asymmetric unit. The contacts among the noncrystallographically related molecules are generally weak and hydrophilic in nature. The protein migrates as monomers on a gel filtration column (data not shown). Several loop regions in the Ipk2 structure are disordered, and we have not included the corresponding residues in our current model. These loop regions comprise residues 1-26, 46 -57 (between ␣1 and ␣2), and 76 -110 (between ␤2 and ␤3) in the N-terminal domain and 287-316 in the C-terminal domain. Not surprisingly, the unique aspartate-rich poly-D loop region between residues 287 and 316 is disordered in our structure and was omitted from the final model. This loop is located on the opposite side of the active side between helix ␣9 and ␤10. The C-terminal His tag (-LEHHHHHH) is partially ordered in our structure. Three histidine residues were included in molecule A, but only the Leu and Glu of the His tag were ordered in molecule B.
The C-domain is formed by residues 127-134 and 165-355. It is an ␣ ϩ ␤ fold with a central six-stranded ␤ sheet. The DECEMBER 8, 2006 • VOLUME 281 • NUMBER 49 strand order of this sheet is 6-5-1-2-4-3. One side of the ␤ sheet is stabilized by three helices. Helices ␣7 and ␣8, together with the additional small helix ␣6, connect strands ␤7 and ␤8. The third helix is ␣10 at the C-terminal end of the protein. An addi-tional small ␤-sheet is formed by strands ␤8 and ␤11, where ␤8 is inserted between the stabilizing helix ␣8 and strand ␤9 of the central ␤ sheet, and ␤11 connects the last strand ␤10 with the last helix ␣10. The N-and C-domains are connected by a loop from residues 119 to 126.

Structure of Yeast Inositol Multikinase Ipk2
The inositol-binding domain is inserted between ␤4 and ␤5 of the C-terminal domain. In the case of Ipk2, the inositol-binding domain consists of only two helices, and we have designated this part of the structure as a separate domain based on the homology with the larger inositol binding domain of IP3K.
A data base search with the program Dali (44) confirmed that the structure of IP3K (30,31) is the closest related protein structure. IP3K has been classified as a member of the Saicar-synthase like superfamily in the SCOP data base (45), which also contains the phosphatidylinositol kinase PIPIIK␤ (46) and shares structural and functional similarity with ATP-grasp fold proteins and protein kinases.
The ATP Binding Site-We have obtained the ADP-bound form of Ipk2 by soaking native crystals with 40 mM ATP and 20 mM MnCl 2 . In a 2.4 Å resolution data set, we observed electron density for a molecule of ADP and a Mn 2ϩ ion in the active site of one of the two Ipk2 molecules in the asymmetric unit (Fig. 3A). The active site region of the unoccupied molecule is partially blocked by crystal lattice contacts and it might therefore not be accessible to the cofactor. Despite the short soaking time, we observed only very weak density for the ␥-phosphate group, indicating that either the majority of bound ATP was hydrolyzed or the position of the ␥-phosphate is disordered in the absence of the substrate. Only minor structural differences are observed between the apo-form and the ADP-bound form. The overall root mean square deviation between both structures is 0.3 Å for 255 C␣ atoms.
The ATP binding site is located between the N-and C-domains of Ipk2 (Fig. 3B). The adenosine moiety of ADP is positioned in a hydrophobic pocket that is lined by residues Ile 29 and Leu 117 from the N-domain and residues Leu 121 , Leu 260 , and Ile 324 from the C-terminal domain. Leu 260 is part of the SLL signature motif, and Ile 324 is part of the IDF signature motif, both strictly conserved in the IPK family. The N1 and N6 atoms of adenine are recognized specifically by the kinase through hydrogen bonds to the amide nitrogen and carbonyl oxygen of Leu 120 and Glu 118 , respectively, which anchors the adenosine in the hydrophobic pocket. The ADP-ribose group is hydrogen-bonded to Asp 131 , which is part of the PXXXDXKXG signature motif. The ADP phosphate groups form hydrogen bonds with Lys 31 from the N-terminal domain and Asp 325 from the C-terminal domain. Asp 325 and Lys 131 coordinate the binding of Mn 2ϩ to the ADP phosphates. Residues in the ATP binding site, with the exception of Asp 325 , assume a very similar side chain conformation in the apoenzyme form and the ADPbound form.

DISCUSSION
Structural Homology with Inositol 1,4,5-Trisphosphate 3-Kinase-As predicted from previous sequence alignments, Ipk2 and IP3K (Protein Data Bank code 1W2C) show a high structural similarity despite their overall lower sequence identity of 17.7% (Fig. 4A). A schematic alignment of the domain organization and the location of signature motifs for the three IPK subgroups is shown in Fig. 1A, and a structurebased sequence alignment is given in Fig. 1B. Ipk2 is 355 residues long, whereas the 461-residue-long IP3K contains an additional N-terminal Ca 2ϩ /calmodulin activation domain that is   DECEMBER 8, 2006 • VOLUME 281 • NUMBER 49 not present in Ipk2. 191 C␣ atoms can be aligned between both structures to a root mean square deviation of 1.3 Å.

Structure of Yeast Inositol Multikinase Ipk2
Functionally important secondary structure elements are well conserved between both enzymes, but most of the loop regions assume different conformations, and there are two regions with significant structural differences. First, the central ␤-sheet of the Ipk2 C-terminal domain contains an additional short strand (␤6) and two helices, ␣5 and ␣9. These inserted structural elements are located remote from the active site and could have a functional role to mediate protein-protein interactions in the Ipk2 transcriptional complex.
Second, the inositol-binding domain of Ipk2 is significantly smaller than the IP-binding domain in IP3K. Only two of the five ␣-helices present in IP3K are conserved in Ipk2. However, these two helices form the core region of the inositol-binding site, and their orientation remains similar in both enzymes (Fig.  4A). Thus, the overall conformation of the inositol-binding site is comparable in both enzymes, but significant differences regarding nature and orientation of protein side chains are observed between both enzymes (discussed below). It should also be noted that the larger IP-binding domain in IP3K is not conserved in other IPK family members and could conceivably be required as an interdomain surface to coordinate the additional regulatory domain in the IP3Ks.
A comparison of the ATP binding site of both enzymes shows that the position of the cofactor is similar and that the functionally important residues are conserved (Fig. 4B). Surprisingly, Lys 171 in Ipk2 remains in a conformation similar to the apoenzyme and does not form hydrogen bonds with the ribose hydroxyl groups. Instead, a solvent water molecule (W255) is observed in a position equivalent to the amino group of the IP3K Lys 336 side chain. In Ipk2, Phe 197 (not conserved in IP3K) disrupts the formation of a solvent water network similar to IP3K, and this might contribute to the different conformation of Lys 171 . However, the equivalent Lys 336 in IP3K assumes a similar side chain conformation in the ADP-bound structure determined by Miller and Hurley (Protein Data Bank code 1TZD) (31), suggesting that this residue is not strictly required for coordinating ATP in Ipk2.
Although the coordination of the ADP ␣and ␤-phosphate groups is similar in both enzymes, the position of the metal ion and the Asp 325 side chain are slightly shifted by 1.5 Å. Interestingly, Lys 133 , which is crucial for positioning the substrate and the ATP ␥-phosphate during catalysis, is rotated toward the ADP molecule. It coordinates to Asp 325 and Ser 258 and is in close proximity to the Mn 2ϩ ion in the absence of inositol. A solvent water molecule is located in the position of the lysine amine in IP3K so that the overall geometry of cofactor-binding interactions remains similar in both enzymes.
In IP3K, an additional N-terminal ␤-strand and ␣-helix were observed that partially obstruct the ATP-binding site in the absence of the cofactor. This autoinhibitory interaction is probably part of the N-terminal Ca 2ϩ /calmodulin-regulated activation mechanism of this enzyme. Ipk2 is not regulated by Ca 2ϩ / calmodulin, but it contains an N-terminal segment of similar length. However, the first 26 residues are disordered in both Ipk2 structures, and we did not observe a similar order-disorder transition between the apoenzyme and the ADP-bound form.
Inositol Binding Site-Despite considerable efforts, we have not been able to produce inositol-bound crystals of Ipk2. The current crystallization conditions contain 200 mM CaCl 2 , which is required for crystal formation, since a Ca 2ϩ ion forms a bridging crystal contact between two crystallographically related molecules. The formation of a calcium complex with InsP 3 in the soaking solution might either limit substrate solubility or interfere with substrate binding. Given the close structural homology with IP3K, we have modeled the position of InsP 3 in Ipk2 from the least squares alignment between both enzymes to evaluate substrate binding.
The substrate-binding site of Ipk2 is located between the C-terminal domain and the inositol binding domain. The overall architecture is well conserved between both enzymes (Fig. 5A). In Ipk2, helix ␣3 of the inositol-binding domain, which delineates one side of the substrate binding site, is shifted toward the ATP position when compared with the substrate-bound structure of IP3K. However, a shift of the corresponding helix and an opening movement of the inositol-binding domain upon substrate binding was observed in IP3K, suggesting a similar conformational flexibility for Ipk2. The inositol-binding pocket is a highly positively charged shallow groove on the protein surface. At the bottom of this groove, Met 151 provides a hydrophobic patch that stacks against the inositol ring. The hydrophobic side chain of Met 151 will also interfere with the presence of a phosphate group in the 2-position and prevent binding of 2-phosphorylated inositol phosphates. The inositol 3-position in IP3K is hydrogen-bonded to Lys 133 in the PXXXDXKXG signature motif. The 4-phosphate interacts with Lys 419 and Lys 312 in IP3K. Lys 419 is replaced with His 328 in Ipk2, which is oriented in the same direction but will contribute less favorably to phosphate group binding. Considering the fact that IPMKs need to bind substrates in several orientations, a strong interaction in this position might interfere with enzymatic 6-kinase activity. Lys 147 , which is equivalent to Lys 312 in IP3K, is directed toward the inositol 5-position. A conformational change of helix ␣3 could position this side chain to interact with the 4-and 5-position of the inositol ring, similar to IP3K.
Consistent with the Ipk2 6/3-kinase specificity, the inositol 6-OH is recognized differently between both enzymes. One of the main determinants of substrate selectivity in IP3K is Met 288 , which interferes with an inositol 6-phosphate group and thus precludes binding of inositol 1,4,5,6-tetrakisphosphate. The helix with Met 288 is not conserved in Ipk2, but Arg 150 , which replaces Tyr 315 , occupies a similar position. Assuming a conformational shift of helix ␣3 in Ipk2, Arg 150 will create a favorable binding environment at the 6-position and should also be in a position to coordinate a phosphate group in the inositol 1-position. The inositol 1-phosphate group, finally, is recognized by Arg 319 in IP3K. Surprisingly, this important residue is replaced by Val 154 in Ipk2, but the equivalent hydrogen bonding position is occupied by Arg 204 from helix ␣6 of the C-terminal domain. Helix ␣6 is one of the three additional helices in the Ipk2 C-domain not present in IP3K, and the position of Arg 204 in the inositol-binding site underlines the importance of this helix for enzymatic function. The structural comparison with IP3K clearly shows that the Ipk2 active site provides favorable binding partners for inositol phosphate groups of inositol 1,4,5-trisphosphate that is bound in an orientation for either 6or 3-phosphorylation as well as for the two different inositol 1,3,4,5-tetrakisphosphates for the synthesis of inositol 1,3,4,5,6-pentakisphosphate. Although our structural model illustrates the general nature of substrate binding interactions in Ipk2, it should be noted that inositol binding is likely to induce a shift of helices ␣3 and ␣6 concomitant with slight reorientations of substrate binding side chains to accommodate the inositol phosphate groups.
Implications for Substrate Specificity-Phosphorylation of the inositol 6-hydroxyl requires substrate binding in one of two possible orientations to position the 6-hydroxyl group toward the ATP. In the first orientation, the inositol ring is rotated by 180°around a vertical axis, exchanging the 5-phosphate group with the 4-phosphate group and the 1-phosphate with the 2-phosphate (Fig. 5B). In the other orientation, the inositol ring is additionally rotated around a horizontal axis, bringing the 5-phosphate group onto the 2-position and the 4-phosphate onto the 1-position (Fig. 5C). In both orientations, there is a phosphate group present in the position previously occupied by the inositol 2-hydroxyl. In comparison with IP3K, Ipk2 contains a two-residue insertion at the beginning of helix ␣6 that projects Lys 200 toward a phosphate group in the place of the 2-position, creating a favorable binding environment. In a second replacement in this region, Glu 333 , which is oriented toward the potential phosphate group in IP3K, is replaced with the shorter and uncharged Cys 168 in Ipk2. Both changes create a favorable binding environment for phosphate groups in the 2-position, consistent with the requirement to bind inositol substrate in various positions.
In summary, our results provide a first view of the highly versatile active site of inositol phosphate multikinases. Our structural information has significantly advanced our understanding of the molecular details governing their multiple catalytic activities. The Ipk2 active site provides favorable binding interactions at all five available binding positions for substrate phosphate groups, enabling the enzyme to interact with differently phosphorylated inositol polyphosphates in different orientations as proposed by Shears (47). Modification of this binding pattern in the related IP3Ks rationalizes the increased substrate specificity of these enzymes.