Crystal Structure of the C2 Domain of Class II Phosphatidylinositide 3-Kinase C2α*

Phosphatidylinositide (PtdIns) 3-kinase catalyzes the addition of a phosphate group to the 3′-position of phosphatidyl inositol. Accumulated evidence shows that PtdIns 3-kinase can provide a critical signal for cell proliferation, cell survival, membrane trafficking, glucose transport, and membrane ruffling. Mammalian PtdIns 3-kinases are divided into three classes based on structure and substrate specificity. A unique characteristic of class II PtdIns 3-kinases is the presence of both a phox homolog domain and a C2 domain at the C terminus. The biological function of the C2 domain of the class II PtdIns 3-kinases remains to be determined. We have determined the crystal structure of the mCPK-C2 domain, which is the first three-dimensional structural model of a C2 domain of class II PtdIns 3-kinases. Structural studies reveal that the mCPK-C2 domain has a typical anti-parallel β-sandwich fold. Scrutiny of the surface of this C2 domain has identified three small, shallow sulfate-binding sites. On the basis of the structural features of these sulfate-binding sites, we have studied the lipid binding properties of the mCPK-C2 domain by site-directed mutagenesis. Our results show that this C2 domain binds specifically to PtdIns(3,4)P2 and PtdIns(4,5)P2 and that three lysine residues at SBS I site, Lys-1420, Lys-1432, and Lys-1434, are responsible for the phospholipid binding affinity.

of the C2 domain of PtdIns 3-kinases, which helps us to understand the regulatory mechanism of class II PtdIns 3-kinases.

MATERIALS AND METHODS
Construction of Plasmids and Mutagenesis-DNA fragments encoding the C2 domain of class II PtdIns 3-kinase, mCPK (residues 1384 -1509), was PCR cloned into a modified vector with a His 6 tag at its N terminus with the restriction enzymes EcoRI and HindIII as reported previously (15). The K1420A, K1432A/K1434A, and R1439A/K1440A mutants of the His-tagged mCPK-C2 domain (called C2-K-A, C2-KRK-ARA, and C2-RK-AA) were generated with the overlapping polymerase chain reaction. All cloned plasmid DNA was confirmed by DNA sequencing.
Expression and Purification-The plasmid DNA encoding the mCPK-C2 domain and its mutants were transformed into Escherichiacoli BL21(DE3) cells. The cells were grown at 37°C in 1 liter of Luria Bertani medium with 100 g/ml ampicillin. Protein expression was induced with 0.3 mM isopropyl-1-thio-␤-D-galactopyranoside when A 595 nm reached 0.5-0.6. After further growth at 22°C for 8 h, cells were pelleted by centrifugation and frozen in liquid nitrogen. For protein purification, the frozen cells were thawed on ice, resuspended in 40 ml of extraction buffer (25 mM Tris-HCl, pH 8.0, 5% glycerol, and 0.2% Triton X-100), and disrupted with sonication. After centrifugation, the supernatant was incubated with Ni 2ϩ -NTA agarose beads (Qiagen) at 4°C overnight with constant shaking. The beads were washed several times with 25 mM Tris-HCl, pH 8.0. The fusion protein was finally eluted from the beads with elution buffer (250 mM imidazole, 25 mM Tris-HCl, pH 8.0, and 5% glycerol). The pure protein was combined and dialyzed against and concentrated in the harvesting buffer (25 mM HEPES, pH 7.0 and 5% glycerol).
Crystallization and Data Collection-Crystal growth was performed by the vapor diffusion hanging drop method. Crystallization drops were obtained by mixing 2 l of mCPK-C2 domain (15 mg/ml) and 2 l of reservoir solution (23% polyethylene glycol 6000, 200 mM Li 2 SO 4 , and 100 mM Tris-HCl, pH 8.5) and were equilibrated against 0.5 ml of reservoir solution at 4°C. Crystals were obtained in 2 months. Diffraction data were collected at the BL41XU beam line in SPring-8 (Proposal 2003A0438-NL1-np; Harima, Japan) and at the Gulf Coast Consortium for Protein Crystallography beam line at the Center for Advanced Microstructures and Devices (Baton Rouge, LA). Data were processed using DENZO/SCALEPACK (31) and CCP4 programs (32). The crystals belong to the tetragonal space group P4 1 2 1 2 with cell constants a ϭ b ϭ 53.365 Å and c ϭ 200.110 Å. There are two molecules per asymmetric unit. The best data set was taken in the resolution range 50 -2.3 Å (2.38 -2.30 Å), with completeness, I/(I), and R merge of 99.7% (100%), 43.9 (10.4), and 5.6% (16.5%), respectively.
Structural Determination-The structure was determined by molecular replacement with the program AMORE (32). It was demonstrated that the polyalanine model of the Rattus norvegicus SytI C2 domain (20), adding identical side chains but omitting gap regions as well as five residues at both N-and C-terminal ends, serves as an effective search model. Cross-rotational and translational searches identified the first molecule at 3.0 Å. The second molecule was located by an additional round of translational searching holding the first molecule fixed. After an initial rigid body refinement, subsequent refinements were performed with the program CNS (33). 5% of randomly selected data were used to calculate the R free factor to monitor refinements. Non-crystallographic symmetry restraints were used in the refinements and were gradually released. Modeling was performed with the program O (34). Omitted residues and excluded side chains were gradually added based on 2 F o-Ϫ F c and omit maps. Positional refinements (simulated annealing and energy minimization) were performed first in the resolution range of 47-3.0 Å and finally to 2.3 Å. Water molecules and sulfate ions were then assigned and refined. Subsequently, individual B-factor refinement was performed. The final model was obtained with R and R free factors of 19.7 and 23.3%, respectively. The final model was checked with PROCHECK (32), and the root mean square deviations from ideal geometry are 0.009 Å for bond lengths, 1.6°for bond angles, and 26.6°for dihedral angles. The coordinates of the mCPK-C2 structure have been deposited in the Protein Data Bank (2B3R).
Liposome Binding Assays-Liposome binding assays were performed as described before (35). Phosphatidylinositol, phosphatidylcholine, phosphatidylserine, and phosphatidylethanolamine were purchased from Avanti Polar Lipids. PtdIns (4)P, PtdIns(5)P, and PtdIns(3,5)P 2 were purchased from Echelon, Inc. PtdIns (3)P, PtdIns(3,4)P 2 , PtdIns(4,5)P 2 , and PtdIns(3,4,5)P 3 were purchased from Matreya, Inc. The published liposome binding assay protocol was used for the binding experiments in this report (24) and was summarized as follows: The phospholipid mixture (100 g/reaction) was dried in a SpeedVac. The dried mixture was then resuspended in 100 l of liposome buffer (50 mM Hepes, 100 mM NaCl, pH 7.2), sonicated in a bath sonicator for 15 min, and spun for 10 min at 14,000 rpm, 4°C. The liposomes were resuspended in binding buffer (50 mM Hepes, 100 mM NaCl, 1 mM MgCl 2 , pH 7.2) at the concentration of 1 mg/ml. The liposomes were then incubated with 8 g of purified protein for 15 min at room temperature and centrifuged. The supernatant was saved, and the pellet was resuspended in 100 l of binding buffer. 25 l of both fractions was used for analysis by SDS-PAGE gel and Coomassie Blue staining. The intensity of the stained bands was quantitatively determined with an LKB Ultrascan laser densitometer.

RESULTS AND DISCUSSION
Overall Structure of the mCPK-C2 Domain-We cloned, expressed, and purified the C2 domain (residues 1384 -1509) of the class II PtdIns 3-kinase from M. musculus, mCPK, with a His 6 tag linked by a short peptide (-Gly-Ser-) at the N terminus by the reported method (15). The crystal structure of the mCPK-C2 domain was determined at 2.3 Å resolution with an R-factor of 19.7% (Fig. 1A). Residues 1384 -1505 were located with the electron density map, and other residues at both termini, including the His 6 tag, were disordered. The mCPK-C2 domain folds as a typical type-I anti-parallel eight-stranded ␤-sandwich, with an extra short ␤-strand (␤4Ј) located at the C terminus of strand ␤4 and an extra ␣ helix (␣1) connected to strands ␤5 and ␤6 (Fig. 1B). Structurebased sequence alignments of the mCPK-C2 domain with other four Ca 2ϩ -binding C2 domains containing type-I ␤-sandwich cores (24, 28 -30) and with the C2 domains of the three types of human class II PtdIns 3-kinases (C2␣, C2␤, and C2␥) are shown in Fig. 1C (10,27,28). Almost none of the five aspartate residues involved in Ca 2ϩ binding in the C2 domains of SytI and synatotagmin III (SytIII) and of PKC␣ and PKC␤ is conserved in the C2 domains of class II PtdIns 3-kinases (Fig. 1C).
Although the ␤-sandwich cores of various C2 domains are structurally well conserved, their loop regions at both ends of the cores are highly flexible, especially among the C2 domains from different protein families. The mCPK-C2 domain is folded mostly similarly to the SytI-C2 domain (20), with two ␤-sheets, ␤8-␤1-␤2-␤5 and ␤7-␤6-␤3-␤4, forming the convex and concave sides of the sandwich, respectively. The root mean square deviation between the ␣-carbons of the mCPK-C2 domain and the SytI-C2 domain is 0.87 and 1.05 Å for molecules A and B, respectively. In addition, a five-residue ␣-helix is inserted between strands ␤5 and ␤6, and an additional ␤4Ј-strand is located at the C-terminal side of ␤4 in the mCPK-C2 domain. Sequence alignments show high homology among the C2 domains of class II PtdIns 3-kinases, as compared with those of PKCs and synaptotagmins (ϳ30%). For example, the C2 domains of PtdIns 3-kinase C2␣ share 58 and 40% sequence identity with those of PtdIns 3-kinase C2␤ and PtdIns 3-kinase C2␥, respectively, suggesting a similar structure and function of these C2 domains (Fig. 1C). In addition, the mCPK-C2 domain is almost identical to that of the human PtdIns 3-kinase C2␣, with only 3 of 136 pairs of residues being non-identical.
The Sulfate-binding Sites of the mCPK-C2 Domain-Three sulfatebinding sites are located at the two ends and at the concave region of the ␤-sandwich of the mCPK-C2 domain (Fig. 2). Each sulfate-binding site is associated with two or three lysine/arginine residues, with a positive electrostatic potential surface on the binding sites (Figs. 1C and 2). Among them, either one or both of the two binding sites on the ␤3/␤4strands (SBS I) and on the ␤4Ј-strand (SBS II) were observed in the reported structures of other C2 domains (20 -21, 29 -30).
The first binding site, SBS I, is located at the concave side of the ␤-sandwich and is composed of the side chains of Tyr-1418 on ␤3 and three lysine residues (Lys-1420 on ␤3 and Lys-1432/Lys-1434 on ␤4), which are conserved among the C2 domains of class II PtdIns 3-kinases (Figs. 1C and 2A). The three oxygen atoms in the sulfate ion interact with the side chains of these lysine residues. In addition, one of the oxygen atoms also forms a hydrogen bond with the OH group of Tyr-1418. Similar tripod cluster structures formed by three-lysine residues, interacting with a PO 4 3Ϫ ion and a PS molecule, were also observed in the structures of the C2 domains of PKC␣ and PKC␤, respectively (21,30).
The second sulfate-binding site, SBS II, is located at the CBR end of the ␤-sandwich (Fig. 2B). This site is located at the dimer interface and is determined by a two-residue cluster (Arg-1439/Lys-1440) on the ␤4Јstrand of both the mCPK-C2 domain and its symmetry-related dimer molecule. Lys-1440 is conserved among the C2 domains of class II PtdIns 3-kinases. A phosphatidylserine molecule was bound to a similar region in the PKC␣-C2 domain structure (31). The third sulfate-binding site, SBS III, is located at the end of the molecule opposite the CBR (Figs. 2C and 4A) and is formed with a lysine (Lys-1457) and an arginine (Arg-1461) at the two ends of an ␣1 helix. However, both the ␣1 helix and the two residues defining SBS III are not conserved among other C2 domains of class II PtdIns 3-kinases. Although the three SBSs have positive electrostatic potential surfaces (Fig. 2, D and E), no evidence supports the existence of an extensive and deep binding pocket similar to that observed in the p47phox PX domains (40).
Dimer Interface of the mCPK-C2 Domain-The mCPK-C2 domain exists as a dimer in its crystal structure, as there are two molecules/ asymmetric unit. Examination of contacts between two independent molecules reveals two possible dimer interfaces of the mCPK-C2 domains (Fig. 3). One dimerization interface is located at the convex side of the ␤-sandwich (Fig. 3A) and is stabilized by ionic interactions (Arg-1393/Glu-1496Ј and Glu-1496/Arg-1393Ј), van der Waals interactions (Met-1449/Met-1449Ј), and hydrogen bonding (Gln-1447/Ser-1453Ј and Ser-1453/Gln-1447Ј). The other dimerization interface is located at the concave side of the ␤-sandwich (Fig. 3B). Here, the ␤6-␤7 loop of each monomer contacts the concave side of the other molecule. Two residues (Glu-1477 and Tyr-1432) on the ␤6-␤7 loop interact with a lysine residue (Lys-1420Ј) on the ␤7Ј-strand of the other monomer to stabilize the dimer interface. Each dimerization decreases the accessible FIGURE 1. Structure of the mCPK-C2 domain. A, a ribbon representation of the overall structure of the mCPK-C2 domain. The structure is colored from blue at the N terminus to green at the C terminus, with secondary structure elements labeled in order. Three sulfate ions bound to mCPK-C2 are labeled and rendered as bond models with oxygen atoms colored in red and sulfur atoms in yellow. B, schematic drawing of the topological structure (type I) of the mCPK-C2 domain. ␤-strands are shown as arrows and the ␣-helix as a rectangle. C, secondary structure-based sequence alignment of eight C2 domains. The top four C2 domains belong to different subclasses (␣, ␤, and ␥) of class II PtdIns 3-kinase C2; the bottom four C2 domains (Synaptotagmins I and III, PKC␣, and PKC␤) all have type I folds as seen in their crystal structures (20 -21, 29 -30). Secondary structure elements are assigned and labeled in order. Identical residues are highlighted in cyan, whereas those residues with high homology to non-mCPK-C2 domains are highlighted in yellow. Key residues in the sulfatebinding sites are colored in red and labeled S1, S2, and S3 for SBS I, SBS II, and SBS III, respectively. The aspartate residues associated with the Ca 2ϩ binding regions are indicated with red triangles at the bottom of the sequences. MOLSCRIPT (36) and RASTER3D (37) were used to generate Fig. 1A and Figs. 2-4. surface by 776 and 943 Å 2 /molecule, corresponding to 11.3 and 13.9% of the monomer surface, respectively. Each molecule of the dimer is related to the other one via a 2-fold non-crystallographic symmetry axis, with a root mean square deviation of 0.32 Å between their ␣-carbon atoms. Further examination reveals no other extensive contacts between the neighboring molecules. The observed mCPK-C2 dimer is different from the dimer of the PKC␤-C2 domain, which has the dimer interface between the flank of the ␤-sandwich core and a C-terminal ␣-helix (21).
Dimerization plays an important role in the regulation of enzymatic activity, which often serves as an early step in signal transduction for many kinases/phosphatases (41). For example, class I PtdIns 3-kinases appear to form a heterodimeric complex of p110 and p85 (5). The het-erodimer can further dimerize in response to stimulation with diphosphotyrosine-containing peptides, and such dimerization is dependent on PtdIns 3-kinase concentration, but not on diphosphopeptide concentration (42). In addition, dimerization of the C2 domain of synaptotagmin, although Ca 2ϩ dependent, was shown to play an important role in the efficient regulation of exocytosis (43). However, no evidence has shown that dimerization of class II PtdIns 3-kinases is involved in the regulation of this class of kinases. Nevertheless, the crystal structure of the mCPK-C2 domain suggests that the C2 domain of class II PtdIns 3-kinases might play a role in the regulation of the kinase through dimerization at higher concentrations, similar to those reported for class I PtdIns 3-kinases (42), as the C2 domain-containing class II PtdIns 3-kinases were suggested to be linked to diverse receptor-mediated sig-  naling processes such as those involving insulin, epidermal growth factor, and platelet-derived growth factor (26,44). In these processes, dimerization serves as a basic step in enzyme activation. More experiments will be carried out to explore this hypothesis.
Comparison with Other C2 Domains-Although C2 domains in different proteins share relatively low sequence homology (ϳ35%), structural analyses reveal a conserved eight-stranded anti-parallel ␤-sandwich core (45), which can be classified into two main types of folds. The first ␤-strand in the type I fold has a similar location as the last one in the type-II fold (45). The mCPK-C2 domain has a type-I fold. Therefore, its structure was compared with the structures of four other C2 domains with type-I ␤-sandwich cores (Figs. 1C and 4). Among them, the mCPK-C2 and SytIII-C2 domains do not bind Ca 2ϩ ions, whereas the SytI-C2, PKC␣-C2, and PKC␤-C2 domains do bind Ca 2ϩ ions.
A superposition of these C2 domains with respect to the ␤-sandwich core reveals the stability of the core and flexibility of the loops at both ends of the sandwich (Fig. 4). Although these C2 domains differ in their Ca 2ϩ dependence, the loops at the CBR end are relatively stable in comparison with those on the other end (Fig. 4A), which show large structural variation among different C2 domains. Ca 2ϩ -binding sites in the PKC␣ structure were formed with five aspartate residues. Sequence alignments show that the mCPK-C2 domain has two glutamate residues (Glu-1410 and Glu-1473) instead of aspartate residues at the corre-sponding sites of the CBR end (Fig. 1C). In contrast, the corresponding sites in the C-terminal C2 domains of human PtdIns 3-kinase C2␤ and PtdIns 3-kinase C2␥ contain two glutamine and proline/isoleucine residues, respectively. Among the loops containing these potential CBRs, the mCPK-C2 domain did show a large structural change in the ␤2-␤3 loop but not in the ␤4Ј-␤5 and ␤6-␤7 loops, in comparison with four other C2 domains with a type I ␤-sandwich (Fig. 4A).
Structural imposition of these type-I C2 domains also reveals relatively conserved anion binding pockets. As described above, such pockets could be classified into two groups, one located at the ␤-strands in the concave side of the ␤-sandwich core (SBS I) and the other one formed by loops in the CBR end (SBS II). In terms of structural stability observed within the ␤-sandwich cores, the SO 4 2Ϫ /PO 4 3Ϫ /PS binding pockets composed of residues on the concave side of the ␤-sandwich core are similar (Fig. 4A). Detailed comparisons of the SBS I sites among the three complex structures are shown in Fig. 4, B-D. The tripod lysine clusters and the tyrosine residue are structurally conserved. In contrast, the second binding site shows significant positional movement, and/or some sites were even associated with different loops in different C2 domains. The third binding site in the mCPK-C2 domain structure was not observed in the structures of other C2 domains, and the lysine/ arginine residues that interact with sulfate ions were not conserved. Lipid Binding Properties of the Wild-type and Mutated mCPK-C2 Domains-As the PX domain of the class II PtdIns 3-kinase C2␣ binds to PtdIns(4,5)P 2 (15), and some C2 domains in SytI and PKC␤ bind PA and PS (10, 23-26), we hypothesized that the mCPK-C2 domain would also have affinity for specific phospholipids. The three sulfate-binding sites identified from the crystal structure of mCPK-C2 domain are potential binding sites for phosphate(s) in the head groups of different phospholipids. To address this question, the lipid binding profile of the mCPK-C2 domain was studied with liposome binding assays. The Histagged mCPK-C2 domain was purified and mixed with liposomes containing 10% of each tested phospholipid. The amount of protein bound to specific phospholipids was quantified by densitometry and summarized in Fig. 5.
Our results show that the mCPK-C2 domain indeed bound specifically to PtdIns(3,4)P 2 and PtdIns(4,5)P 2 . The binding site(s) for phospholipids is predicted to be one or more of the three sulfate-binding sites identified in the mCKP-C2 structure. However, the SBS sites in the mCPK-C2 domain are shallow, compact, and exposed to the surface, as compared with the deep, spacious binding pocket of the PX domains for the large head groups of other phosphatidylinositol phosphates (40). In the mCPK-C2 domain, all three SBSs are Ͻ5 ϫ 5 Å in size and Ͻ4 Å in depth. In contrast, a PtdIns (3)P binding model of the p47phox PX domain shows that the head group of PtdIns (3)P occupies ϳ2/3 of a pocket that is ϳ7 Å in depth with an external opening of 10 ϫ 8 Å (40). These results suggest that the inositol ring is not buried in the cavity of the mCPK-C2 domain, and therefore the C2 domain binds lipids in a different fashion than the PX domain. This kind of binding mode is primarily defined by the ionic interaction between lysine side chains and phosphates (as sulfate observed in structure), which is consistent with the observation for other C2 domains that bind to PA and/or PS within SytI and PKC␤ (10,(23)(24)(25)(26).
To further identify the binding site for PtdIns(3,4)P 2 and PtdIns(4,5)P 2 , the lysine/arginine residues in SBS I (Lys-1420, Lys-1432/   Lys-1434) and SBS II (Arg-1439/ Lys-1440) were mutated to alanines. The residues in SBS III were not mutated because they were not conserved within the class II PtdIns 3-kinases (Figs. 1C and 4A). The mutated mCPK-C2 proteins were purified, and their lipid binding properties were studied using liposome binding assays. As shown in Fig. 5, the R1439A/K1440A double mutant (C2-RK-AA) in SBS II of the mCPK-C2 domain still binds specifically to both PtdIns(3,4)P 2 and PtdIns(4,5)P 2 . The K1421A mutant (C2-K-A) and the K1432A/K1433A double mutant of the C2 domain (C2-KRK-ARA) in SBS I have, however, decreased their binding affinity for these phospholipids. This suggests that SBS I may be primarily responsible for the phospholipid binding.
To further characterize this lipid binding behavior, the binding affinity was studied with liposomes containing different concentrations of PtdIns(3,4)P 2 and PtdIns(4,5)P 2 for mCPK-C2 domain and for the three mutants as well (Figs. 6 and 7). For both PtdIns(3,4)P 2 and PtdIns(4,5)P 2 , the C2-RK-AA mutant showed a similar binding affinity to the wildtype mCPK-C2 domain at various tested concentrations, also suggesting that the SBS II does not contribute to the phospholipid binding. In contrast, both the C2-K-A and the C2-KRK-ARA mutants in SBS I have shown decreased binding affinity for PtdIns(3,4)P 2 and PtdIns(4,5)P 2 from the wild-type mCPK-C2 domain, clearly indicating that the SBS I plays an important role in the PtdIns(3,4)P 2 and PtdIns(4,5)P 2 binding.
As shown in Figs. 6B and 7B, following the increase in concentration of the tested phospholipids, the mCPK-C2 domain and C2-RK-AA mutant tend to bind to liposome in a saturation mode, with the apparent half-saturation concentration as ϳ7.5% for both proteins. In contrast, at low phospholipid concentration, the C2-K-A and C2-KRK-ARA mutants showed low binding affinity. With increasing phospholipid concentrations, they did not bind liposome in a saturation mode. In addition, beyond 20% lipid concentration, their binding affinity was significantly increased. Meanwhile, the similarity between the binding property of these two mutants in SBS I suggests that a loss of the binding residues causes a significant reduction in binding affinity, which in turn suggests a collaboration among three lysine residues involved in SBS I, as observed in the crystal structure. Both crystallographic and biochemical data suggest that the 4Ј-phosphate in PtdIns(3,4)P 2 and PtdIns(4,5)P 2 most likely dominates the binding to SBS I. In addition, the existence of another phosphate group, the 3Ј-or 5Ј-group, on the inositol ring dramatically increases the binding affinity to the mCPK-C2 domain. However, it could not be both because PtdIns(3,4,5)P 3 only showed a medium binding affinity for mCPK-C2 domain (Fig. 5B), probably due to a potential space resistance effect.