Structural Insights into Activation of Phosphatidylinositol 4-Kinase (Pik1) by Yeast Frequenin (Frq1)*

Yeast frequenin (Frq1), a small N-myristoylated EF-hand protein, activates phosphatidylinositol 4-kinase Pik1. The NMR structure of Ca2+-bound Frq1 complexed to an N-terminal Pik1 fragment (residues 121-174) was determined. The Frq1 main chain is similar to that in free Frq1 and related proteins in the same branch of the calmodulin superfamily. The myristoyl group and first eight residues of Frq1 are solvent-exposed, and Ca2+ binds the second, third, and fourth EF-hands, which associate to create a groove with two pockets. The Pik1 peptide forms two helices (125-135 and 156-169) connected by a 20-residue loop. Side chains in the Pik1 N-terminal helix (Val-127, Ala-128, Val-131, Leu-132, and Leu-135) interact with solvent-exposed residues in the Frq1 C-terminal pocket (Leu-101, Trp-103, Val-125, Leu-138, Ile-152, and Leu-155); side chains in the Pik1 C-terminal helix (Ala-157, Ala-159, Leu-160, Val-161, Met-165, and Met-167) contact solvent-exposed residues in the Frq1 N-terminal pocket (Trp-30, Phe-34, Phe-48, Ile-51, Tyr-52, Phe-55, Phe-85, and Leu-89). This defined complex confirms that residues in Pik1 pinpointed as necessary for Frq1 binding by site-directed mutagenesis are indeed sufficient for binding. Removal of the Pik1 N-terminal region (residues 8-760) from its catalytic domain (residues 792-1066) abolishes lipid kinase activity, inconsistent with Frq1 binding simply relieving an autoinhibitory constraint. Deletion of the lipid kinase unique motif (residues 35-110) also eliminates Pik1 activity. In the complex, binding of Ca2+-bound Frq1 forces the Pik1 chain into a U-turn. Frq1 may activate Pik1 by facilitating membrane targeting via the exposed N-myristoyl group and by imposing a structural transition that promotes association of the lipid kinase unique motif with the kinase domain.

Frq1 and other frequenins belong to the neuronal calcium sensor (NCS) branch of the calmodulin superfamily, which includes recoverin and neurocalcin (28 -31). These proteins are small (Յ25 kDa) and characterized by a consensus signal for N-terminal myristoylation and four EF-hand Ca 2ϩ -binding sites (Fig. 1). We have shown previously that, at saturation, Frq1 binds only three Ca 2ϩ (32). Frq1, which is itself essential for the viability of yeast cells (20), associates with membranes in a manner that depends on both the N-myristoyl group and conformational changes induced upon Ca 2ϩ binding, suggesting that Frq1, like other NCS proteins, may possess a Ca 2ϩ -myristoyl switch (32). Indeed, prior work indicated that N-myristoylation of Frq1 is important (but not essential) for stimulating both the catalytic activity (20) and the membrane recruitment of Pik1 (17).
Three-dimensional structures for Frq1 and other NCS proteins have been determined by x-ray crystallography (23,(33)(34)(35)(36)(37) and NMR spectroscopy (32, 38 -40). The structure of Frq1 in solution revealed that calcium is bound at EF-2, EF-3, and EF-4, and the overall main chain fold is similar to that seen previously for Ca 2ϩ -bound forms of recoverin, neurocalcin, GCAP2, and KChIP1. The four EF-hands form two domains packed in a globular arrangement that contrasts with the dumbbell-shaped arrangement of EF-hand domains seen in calmodulin and troponin C (41,42). A striking feature of the NCS structures is a solvent-exposed hydrophobic groove formed by residues Phe-22, Trp-30, Phe-34, Phe-48, Ile-51, Tyr-52, Phe-55, Phe-85, and Leu-89 in the N-terminal domain of Frq1 that are invariant in all other NCS proteins (Fig. 1). The corresponding hydrophobic residues in GCAP2, recoverin, and KChIP1 have been implicated previously in target recognition (35,40,43,44).
We report here the NMR-derived structure of Frq1 in solution bound to a fragment (residues 121-174) corresponding to the Frq1-docking site in Pik1 (21), hereafter referred to as Pik1-(121-174). This is the first atomic resolution structure of a Ca 2ϩ -myristoyl switch protein bound to a lipid kinase target protein. The structure reveals that Pik1-(121-174) forms two antiparallel ␣-helical segments that interact with bipartite binding sites on the surface of Frq1. In essence, binding of Frq1 generates a U-turn in the Pik1 polypeptide, shedding considerable light on how Frq1 may potentiate the activity of this enzyme. Moreover, the structure of Frq1 bound to Pik1-(121-174) is somewhat different from the target complexes of other EF-hand proteins, like recoverin (40), calmodulin (45), KChIP1 (35,36), calcineurin B (46), and troponin C (47), and thus provides new insights about the molecular basis of target recognition specificity in this class of regulatory proteins.

EXPERIMENTAL PROCEDURES
Protein Preparation-To prepare recombinant Frq1 uniformly labeled with nitrogen-15 and/or carbon-13, Frq1 tagged with a C-terminal His 6 tract was expressed in Escherichia coli strain BL21(DE3) carrying a derivative of the pET23d vector (Novagen) harboring the FRQ1 coding sequence, constructed as described previously (20) (48 -50). Labeled Frq1 was purified from the soluble fraction of bacterial cell lysates using Ni 2ϩ -chelate affinity chromatography on a nitrilotriacetate resin (Qiagen), according to the manufacturer's instructions. Peak fractions were then applied to an anion-exchange column (50-ml bed, DEAE-Sepharose, GE Healthcare) equilibrated in buffer A (2 mM CaCl 2 , 20 mM Tris-HCl, pH 7.4) and eluted with a linear salt gradient (0 -0.4 M KCl) at a flow rate of 2 ml min Ϫ1 over the course of 180 min. Peak fractions were concentrated to 5 ml and subjected to size-exclusion chromatography (Sephacryl S-100, GE Healthcare) in buffer B (1 mM dithiothreitol, 2 mM CaCl 2 , 50 mM HEPES, pH 7.4). Final purity was greater than 98%, as judged by SDS-PAGE.
NMR Spectroscopy-Samples for NMR analysis consisted of 15 N-labeled or 13 C/ 15 N-labeled Frq1 bound to 1 eq of unlabeled Pik1-(121-174) (1.0 mM) in 0.3 ml of a 95% H 2 O, 5% [ 2 H]H 2 O solution containing 5 mM sodium acetate and 2 mM CaCl 2 (pH 5.0). Reverse-labeled samples (i.e. 15 N-or 13 C/ 15 N-labeled Pik1-(121-174) bound to 1 eq of unlabeled Frq1) were also prepared for some of the NMR experiments. All NMR experiments were performed at 40°C on a Bruker DRX-500 or DRX-600 spectrometer equipped with a four-channel interface and triple resonance probe with triple axis pulsed field gradients and DRX-600 spectrometer equipped with an Ultrashield Bruker magnet, a three-channel interface, and cryo-probe with z axis pulsed field gradients. The 15 N-1 H HSQC spectra (see The number of complex points and acquisition times were as follows: 256, 180 ms ( 15 N (F 1 )); and 512, 64 ms ( 1 H (F 2 )). The 13 C(F1)-edited, 13 C(F3)-filtered NOESY-HMQC spectra (see Fig. 2C) were recorded on a sample of unlabeled Frq1 protein bound to 13 C-labeled Pik1-(121-174) (51) as well as 13 C-labeled Frq1 bound to unlabeled Pik1-(121-174) (data not shown). Intermolecular NOESY experiments were performed as described previously (52). Stereospecific assignments of chiral methyl groups of valine and leucine were obtained by analyzing 1 H-13 C HSQC experiments performed on a sample that contained 10% 13 C labeling in either Frq1 or Pik1-(121-174) (53). All triple resonance experiments were performed, processed, and analyzed as described (54, 55) on a sample of 13  The triple resonance and NOESY spectra measured above were analyzed to determine secondary and tertiary structure in Frq1-Pik1-(121-174) complex. The chemical shift index (see Ref. 56 for detailed description), 3 J NH␣ coupling constants, and NOE connectivity patterns for each residue were analyzed and provided a measure of the overall secondary structure. Small 3 J NH␣ coupling constants (Ͻ5 Hz), strong NOE connectivities (NN(i,i ϩ 1) and ␣N(i,i ϩ 3)), and positive chemical shift index are characteristic of residues in an ␣-helix. Conversely, large 3 J NH␣ coupling constants (Ͼ8 Hz), strong ␣N(i,i ϩ 1) and weak NN(i,i ϩ 1) NOE connectivities, and negative chemical shift index are characteristic of residues in a ␤-strand. The results of the secondary structure analysis of Frq1-Pik1-(121-174) complex are summarized schematically in Fig. 1. 15 N{ 1 H} NOE data were measured using two-dimensional 15 N, 1 H HSQC-based experiments as described previously (57). Saturation was carried out with a series of 120°1H pulses separated by 5-ms delays applied during the interscan delay (3 s).
Structure Calculation-Backbone and side chain NMR resonances of Frq1-Pik1-(121-174) complex were assigned as described previously (55). Structure calculations were performed using the YASAP protocol within X-PLOR (58), as described previously (59). A total of 2300 interproton distance constraints were obtained as described (55) by analysis of 13 Cedited and 15 N-edited NOESY-HSQC spectra (100 ms mixing time) of 13 C, 15 N-labeled Frq1 bound to unlabeled Pik1-(121-174) and 13 C, 15 N-labeled Pik1-(121-174) bound to unlabeled Frq1. In addition to the NOE-derived distance constraints, the following additional constraints were included in the structure calculation: 18 distance constraints involving Ca 2ϩ bound to loop residues 1, 3, 5, 7, and 12 in each EF-hand motif (EF-2, EF-3, and EF-4); 170 distance constraints for 85 hydrogen bonds; and 222 dihedral angle constraints. Fifty independent structures were calculated, and the 20 structures of lowest energy were selected. The average total and experimental distance energies are 4730 and 61 kcal mol Ϫ1 . The average root mean square deviations (r.m.s.d.) from an idealized geometry for bonds and angles are 0.0068 Å and 2.05°. None of the distance and angle constraints were violated by more than 0.40 Å or 4°, respectively.
Deletion Analysis of Pik1-To construct Pik1-(⌬34 -110), Pik1-(⌬8 -760), and Pik1-(⌬16 -833), DNA fragments of PIK1 carrying these respective internal deletions were generated by PCR with appropriate primers and pRS316-GAL1prom-GFP-Pik1 as the template. The resulting PCR products were incorporated in place of the normal PIK1 open reading frame in pRS316-GAL1prom-GFP-Pik1, which had been linearized by cleavage with MfeI, via homologous recombination-mediated gap-repair (60) in yeast strain BY4743. Plasmid DNA was recovered from the resulting Ura ϩ transformants, amplified in Escherichia coli, and sequenced to verify production of the proper in-frame deletions. The control plasmid, pRS314-GAL1prom-GFP-Pik1, has been described before (17). Standard methods for DNA-mediated transformation, sporulation, tetrad dissection, and other genetic manipulations of yeast cells were used (61). In vitro lipidkinase assays and immunoblot analysis of proteins were done essentially as described before (17).

RESULTS
Preparation and Characterization of the Frq1-Pik1-(121-174) Complex-We showed previously that Frq1 interacts with Pik1 in a localized region (residues 121-174) that is necessary and sufficient for Pik1 activation by Frq1 (21). A stoichiometric complex of Frq1 bound to Pik1-(121-174) at saturating Ca 2ϩ is soluble and stable under NMR conditions. The Frq1-Pik1-(121-174) complex is monomeric in solution with a total molecular mass of ϳ30 kDa, as judged by dynamic light scattering analysis and size-exclusion chromatography. The binding energetics for the Frq1-Pik1-(121-174) complex were quantified using isothermal titration calorimetry, as described previously (21). Analysis of the isothermal titration calorimetry data revealed that complex formation occurs with a stoichiometry of 1:1, a dissociation constant of ϳ100 nM, and enthalpy of ϩ7 kcal/mol, suggesting that binding is largely entropy-driven. Entropically driven binding is consistent with hydrophobic intermolecular interactions and/or a protein conformational change.
NMR Structural Analysis of the Frq1-Pik1-(121-174) Complex-The 1 H-15 N-HSQC NMR spectrum of 15 N-labeled Frq1 exhibited spectral changes that saturated upon the addition of 1 eq of unlabeled Pik1-(121-174) ( Fig. 2A), confirming that Frq1 binds to Pik1-(121-174) in a 1:1 complex under NMR conditions. The 1 H-15 N HSQC spectrum of 15 N-labeled Frq1 in the complex exhibited the expected number of amide resonances (190) with uniform intensity, indicating that the complex is structurally homogeneous and stably folded. Pulsed field gradient NMR diffusion studies (62) confirmed that the complex is monomeric under NMR conditions. Sequence-specific assignments for the amide peaks are indicated in Fig. 2A. More than 96% of the amide resonances were assigned except those of Ser-60, Thr-91, Lys-100, Asn-184, and Leu-185 that could not be assigned because of chemical shift degeneracy and/or undetectable NMR intensities. The amide chemical shifts of many residues of Frq1 in the complex are similar to those of free Frq1 (32), suggesting that the overall main chain structure of Frq1 in the complex (discussed below) is similar to that of free Frq1 (32) and mammalian NCS-1 (23) determined in the absence of target.
A reverse labeled sample containing 15 N-labeled Pik1-(121-174) bound to unlabeled Frq1 allowed us to selectively probe 1 H-15 N HSQC NMR spectra of Pik1-(121-174) in the complex. The NMR spectrum of Pik1-(121-174) exhibited a subset of sharp resonances clustered near the middle of the spectrum at 8.0 ppm and a separate group of broader peaks (Fig. 2B). The sharp resonances were assigned to residues 138 -154 of Pik1-(121-174). The sharpness of the peaks, narrow chemical shift dispersion, and low heteronuclear NOE values (Ͻ0.6, see supplemental Fig. 1) all suggest that these residues are largely unstructured. The remaining peaks exhibited much greater chemical shift dispersion and, higher heteronuclear NOE values and were assigned to residues 125-136 and 156 -169, which form two separate ␣-helices in Pik1-(121-174), as deduced from analyses of J-coupling, chemical shift index, and sequential NOE patterns (see "Experimental Procedures" and supplemental Fig. 1 and Fig. 1B).
Our analysis of the NMR data for the Frq1-Pik1-(121-174) complex (see "Experimental Procedures") permitted the assign- ment of more than 94% of all NMR resonances and NOE data, which then served as the basis for determining the structural constraints in Table 1. Differential isotope labeling of the complex and analysis of isotope-filtered NOE experiments (52,55) enabled the selective probing of 89 intermolecular NOEs involving Frq1 residues located less than 5 Å away from residues of Pik1-(121-174) at the binding interface. Structures derived from the NMR data are illustrated in Figs. 3 and 4 (atomic coordinates are available from RCSB Protein Data bank, code 2JU0). The overall structure is a 1:1 complex of Frq1 and Pik1-(121-174) with overall dimensions of 46 Å (length) by 28 Å (height) by 31 Å (depth). The final NMR-derived structures (20 lowest energy structures out of a total of 50) were superimposed, and the r.m.s.d. relative to the mean structure was calculated to be 0.55 Å for main chain atoms and 1.2 Å for all heavy atoms in regions of regular secondary structure (see supplemental Fig. 2 and Table 1). The average main chain structure of the complex in solution is represented as a ribbon diagram in Fig. 3A.

Structure of Frequenin Bound to PtdIns 4-Kinase
The structure of Pik1-(121-174) in the complex adopts a conformation that contains two ␣-helices (residues 125-136 and 156 -169) connected by a disordered loop (supplemental Fig. 1 and Fig. 1B). The N-terminal helix contains hydrophobic residues (Ala-128, Val-131, Leu-132, and Leu-135) that contact C-terminal residues of Frq1 (Leu-101 Intermolecular Interactions in the Frq1-Pik1-(121-174) Complex-Analysis of isotope-filtered NMR NOESY data selectively probed those residues of Frq1 located less than 5 Å away from residues of Pik1-(121-174) situated at the binding interface (Fig. 4). Representative slices of three-dimensional 13 C(F 1 )-edited, 13 C(F 3 )-filtered NOESY-HMQC spectra of 13 Clabeled Pik1-(121-174) bound to unlabeled Frq1 (Fig. 4A) reveal that Pik1 residues in the C-terminal helix (157-168) form contacts with Frq1 residues of EF-1 and EF-2 in the N-terminal hydrophobic groove. Most striking are hydrophobic contacts involving Ala-157, Ala-159, and Val-161 (Fig. 4B). These contacts and others were confirmed in reverse labeling isotopefiltered NMR experiments performed on 13 C-labeled Frq1 bound to unlabeled Pik1-(121-174). Intermolecular contacts were also observed for N-terminal Pik1 residues (125-136) with EF3 and EF4 of Frq1 (Fig. 4C). However, the intermolecular NOE intensities for the N-terminal Pik1 residues were quite weak by comparison, perhaps because of conformational instability in this region. At lower temperatures and higher salt concentrations, the intermolecular NOEs to the N-terminal helix of Pik1 were not discernible, whereas intermolecular NOEs involving the C-terminal helix of Pik1 (Fig. 4A) were quite strong under all conditions. The sharper NMR linewidths for the C-terminal resonances of Pik1 suggest that the C-terminal helix of Pik1 forms more stable contacts with Frq1 than does the N-terminal helix of Pik1, which seems to be significantly less stable.
The C-terminal helix of Pik1-(121-174) contains hydrophobic residues that make extensive contacts with aromatic and other hydrophobic side chains of Frq1 (Fig. 4B) The N-terminal helix of Pik1-(121-174) interacts with hydrophobic side chains of Frq1 (Fig. 4C). The side chain methyl groups of Val-127 (Pik1) and Val-131 (Pik1) are less than 5 Å away from the ␦1and ␦2-methyl groups of Leu-138 from Frq1. The ␥1-methyl group of Val-131 (Pik1) contacts the ␦1-methyl group of Ile-152 (Frq1) and ␥1-methyl group of Val-125 (Frq1). The side chain methyl groups of Leu-132 (Pik1) contact both the ␦1and ␦2-methyl groups of Leu-101 (Frq1) and ␤-methyl group of Ala-104 (Frq1) (data not shown). The ␦1-methyl of Leu-135 (Pik1) contacts the aromatic ring of Trp-103 (HH2 and HZ3) and ␤-methyl of Ala-104. The ␦2-methyl of Leu-135 (Pik1) contacts the ␥1-methyl of Val-128 (Frq1) (data not shown). These intermolecular interactions, illustrated in Fig. 4C, contribute to the overall binding energy. However, the exchange broadening of NMR resonances associated with the N-terminal helix of Pik1-(121-174) suggests a relatively unstable interaction in this region (Fig. 4C), in contrast to the more stable interaction involving the Pik1 C-terminal helix (Fig. 4B). The intermolecular interactions depicted in Fig. 4, B and C, are believed to represent two distinct and independent binding sites because the two sites are separated spatially and do not interact structurally. Nonetheless, these two sites presumably act synergistically because the ligands with which they interact (the 11-and 14-residue ␣-helices in Pik1) are tethered covalently by a 20-residue spacer.
Catalytic Activity of Pik1 Deletion Mutants-A series of deletion constructs of Pik1 were analyzed to identify regulatory domains ( Table 2). To examine catalytic competency relative to wild-type Pik1, a qualitative immune complex in vitro lipid kinase assay was used (13,17). A deletion construct of Pik1 containing only the lipid kinase catalytic domain, Pik1-(⌬8 -760), exhibits very low basal activity (Fig. 5, lane 3), just barely detectable above a negative kinase-dead control (lane 1). This  OCTOBER 19, 2007 • VOLUME 282 • NUMBER 42

Structure of Frequenin Bound to PtdIns 4-Kinase
level of activity is much lower than that of wild-type Pik1 in the absence of Frq1 (Fig. 5, lane 4). Correspondingly, expression in yeast of the Pik1 catalytic domain alone, Pik1-(⌬8 -760), does not complement the lethality of a pik1⌬ mutation (Fig. 6).
These results indicate that the lipid kinase domain of Pik1 by itself is nonfunctional both in vitro and in vivo, in contrast to the constitutive activation observed when the catalytic domains of many protein kinases are separated from their regulatory domains (64 -67). A deletion construct of Pik1 that lacks only the Frq1-binding site, Pik1-(⌬152-191), does not bind Frq1, as demonstrated before (22). However, deletion of the Frq1-binding site Pik1-(⌬152-191) has almost no effect on basal catalytic activity; compare Pik1-(⌬152-191) to wild-type Pik1 (Fig. 5,  lane 7 versus 5). Hence, activation of Pik1 by Frq1 does not involve relief of an autoinhibitory constraint, contrary to the mechanism by which calmodulin binding activates calmodulindependent protein kinases (65). Instead, the Pik1 catalytic domain by itself is inactive and requires interaction and stabilization by an auxiliary domain located within residues 8 -760. Indeed, a deletion construct of Pik1 that lacks only the lipid kinase unique (LKU) motif (residues 35-110) has almost no activity compared with wild-type Pik1 (Fig. 5, compare lanes 2  and 4), and this deletion construct also does not complement a pik1⌬ mutation (Fig. 6A). Taken together, these results indicate that the LKU motif and the catalytic domain are both essential for functional lipid kinase activity. In summary, it seems that Frq1 binding to Pik1 does not activate kinase activity by removing an autoinhibitory constraint but rather promotes interaction between the N-terminal LKU motif and the C-terminal catalytic domain.

DISCUSSION
We present here the atomic resolution structure of yeast frequenin (Frq1) bound to an N-terminal fragment of its target, a PtdIns 4-kinase isoform (Pik1). The overall main chain topology of Frq1 in the complex is similar to that of free Frq1 in solution (r.m.s.d. ϭ 2.0 Å in EF-hand regions) and other NCS proteins (23,(33)(34)(35)(36)63). In the complex, the first 8 residues at the N terminus and last 10 residues at the C terminus of Frq1 are solvent-exposed and structurally disordered. We propose that the solvent-exposed N-terminal myristoyl group of Frq1 may help recruit Pik1 to membranes where it can encounter its substrate PtdIns. Three Ca 2ϩ are bound to Frq1 at EF2, EF3, and EF4. The four EF-hands form an elongated groove on one side of the protein lined by two distinct hydrophobic patches that interact with two separate ␣-helical segments in Pik1-(121-174) (Fig. 3B). The intermolecular interactions between Frq1 and Pik1-(121-174) are mostly hydrophobic (Fig. 4). Most interestingly, Frq1 binding induces a U-shaped helix-loop-helix structure in Pik1-(121-174) that we propose permits functional interaction between the N-terminal LKU motif and the C-terminal catalytic domain of Pik1 that are both necessary for its lipid kinase activity (Fig. 8).
The sequence similarity of Frq1 and recoverin suggests that a Ca 2ϩ -myristoyl switch (i.e. Ca 2ϩ -induced extrusion of the N-terminal myristoyl group) (68, 69) by Frq1 might promote the targeting of Pik1 to membranes. Indeed, myristoylation of Frq1 enhances Pik1 activation (70). However, Frq1 binding to Pik1 does not require calcium or myristoylation, and Ca 2ϩbinding deficient mutants of Frq1 and mammalian NCS-1 bind and activate Pik1 (20,22). If neither Ca 2ϩ nor myristoylation is essential for Frq1 binding to Pik1, then perhaps Frq1 does not possess a functional Ca 2ϩ -myristoyl switch. Indeed, we found  were induced to sporulate and the resulting tetrads were dissected and germinated on galactose medium. In each panel, the four spore clones (A-D) of 12 representative tetrads are shown. The fact that only the two PIK1 ϩ spores in any given tetrad (panels II-IV) are able to grow demonstrates that, unlike wild-type Pik1 (panel I), none of the mutants is able to support the growth of either of the two pik1⌬ spores and, thus, that removal of the N-terminal segment of Pik1, or of just its LKU motif alone, inactivates the function of Pik1 in vivo.
previously that the N-myristoyl group of Frq1 remains solventexposed regardless of the Ca 2ϩ level (32) and that Frq1 binds to membranes in both the presence and absence of calcium (20). These observations suggest that the myristoyl group of Frq1 may remain extruded even in the Ca 2ϩ -free state, in contrast to the sequestered myristoyl group observed in Ca 2ϩ -free recoverin (39). These considerations likely explain the calcium-insensitive membrane localization displayed in vivo by both Frq1 (70) and NCS-1 (71). Consistent with this view, structural studies of Ca 2ϩ -free Frq1 (32) and NCS-1 (72) indicate that the N-myristoyl group is extruded (32), in stark contrast to the structure of Ca 2ϩ -free recoverin with a sequested myristoyl group. These observations might explain, in part, why retinal recoverin is unable to complement the lethality of a frq1⌬ mutant (20), whereas NCS-1 is able to do so (22). A recent mutagenesis study comparing NCS-1 and recoverin suggested that nonconserved residues of recoverin (Glu-15 to Asn-20) might be responsible for conferring the Ca 2ϩ -myristoyl switch property of recoverin that apparently is not observed for Frq1 and NCS-1 (72).
Nonconserved residues of NCS proteins at the C terminus and immediately following EF3 (Fig. 1A) may be structurally important for explaining target specificity. The nonconserved C-terminal region of Frq1 (residues, 180 -190) has NMR chemical shifts and 3 J NH␣ indicative of an unstructured random coil in the target complex, in contrast to a well defined C-terminal helix seen in free Frq1 (32). The C-terminal helix in free Frq1 makes contact with residues in EF3 and EF4 (Leu-101, Ala-104, Met-121, Ile-152, Phe-169, and Ser-173) that also contact the N-terminal helix of Pik1-(121-174) in the complex (Fig. 4C). Therefore, the N-terminal Pik1 helix appears to substitute for and perhaps displace the C-terminal helix of Frq1, likely leading to the observed C-terminal destabilization in the complex. The corresponding C-terminal helix of KChIP1 is similarly displaced upon its binding to the Kv4.3 channel (36, 37) but not upon its binding to Kv4.2 (35). The C-terminal helix in recoverin forms a stable interaction with EF3 and EF4, enabling the C-terminal helix to perhaps serve as a built-in competitive inhibitor that would presumably block its ability to bind to tar-gets like Pik1 and Kv4.3. This role for its C terminus may provide yet another reason why recoverin is unable to complement a frq1⌬ mutation in yeast and could account for why the C-terminal sequences of NCS proteins are not well conserved (Fig.  1A). Another nonconserved region of Frq1 implicated in target specificity is the stretch between EF3 and EF4 (residues 134 -146). This region of Frq1 adopts a short ␣-helix in the complex that contacts the N-terminal helix of Pik1-(121-174). By contrast, the region between EF3 and EF4 is unstructured in many other NCS proteins (32,34,35,63,73).
Pik1 binding to Frq1 is somewhat analogous to target binding seen in other NCS proteins, including recoverin and KChIP1 (Fig. 7). NCS proteins generally have an N-terminal domain of two EF-hands that form an exposed hydrophobic crevice that interacts with a helical segment of target proteins. In recoverin, the two N-terminal EF-hands form an exposed hydrophobic groove that interacts with a hydrophobic target helix from rhodopsin kinase (RK25) (40) (Fig. 7A). The N-terminal EF-hands of KChIP1 interact with a target-helix derived from the T1 domain of Kv4.2 channels (Fig. 7B) (35) and Kv4.3 channels (36). The orientation of the target helices bound to recoverin and KChIP1 are somewhat similar; the C-terminal end of the target helix is spatially close to the N-terminal helix of EF-1. By contrast, the Pik1 target helix binds to Frq1 in almost the exact opposite orientation (Fig. 7C). The N-terminal end of the Pik1 helix is closest to EF1 (Fig. 7C, green) in Frq1, whereas the C-terminal end of the RK25 target helix is closest to the corresponding region of recoverin. Residues Gly-33 and Asp-37 in Frq1, which are not conserved in recoverin, make important contacts with the Pik1 target helix and presumably assist in imposing the observed orientation of the helix. Thus, the requirement that the helix (in this case, from Pik1) must bind to Frq1 with a polarity opposite to that observed for the helices in other target-NCS family member complexes could clearly contribute to dictating the substrate specificity of frequenins, as compared with other NCS subtypes. Another important structural feature seen in the Frq1-Pik1 interaction is that two helical segments of the target are captured in the complex, whereas in the target complexes characterized for recoverin and KChIP1, only one helix is bound. Therefore, selective substrate recognition by frequenins may be explained by two distinct and unique properties not shared by other NCS proteins as follows: orientation of the bound target helix and the number of target helices bound.
The structure of the Frq1-Pik1-(121-174) complex combined with our deletion and mutational analysis of Pik1 (Table  2) provides insight into the activation mechanism of Pik1 by Frq1 (Fig. 8). Our structure of the Frq1-Pik1 complex reveals that the N-terminal myristoyl group of Frq1 is exposed and thus able to recruit the Frq1-Pik1 complex to membranes where the enzyme can encounter inositol headgroups. Indeed, N-terminal myristoylation of Frq1 enhances Pik1 activation at least 2-fold (70), but myristoylation of Frq1 is not necessary for its activation of Pik1. Therefore, in addition to membrane recruitment, Frq1 must also influence the structure and activity of the lipid kinase catalytic domain in another way. One possibility was that Frq1 might activate the kinase domain indirectly by binding to an autoinhibitory domain in Pik1, analogous to activation of calmodulin-dependent protein kinases (65,66). However, a deletion construct of Pik1 that retains only the catalytic domain, Pik1-(⌬8 -760), has almost no detectable lipid kinase activity compared with full-length Pik1 (Fig. 5), and Pik1-(⌬8 -760) does not complement a pik1⌬ mutation in yeast, even when overexpressed ( Fig. 6 and Table 2). The isolated kinase domain of Pik1 is therefore inactive, in stark contrast to the constitutively active catalytic domains of protein kinases when their negative regulatory domains are removed (64,65,67). Therefore, as a consequence, activation of Pik1 by Frq1 cannot simply involve relief of an autoinhibitory constraint. Instead, we propose on the basis of several facts that Frq1 binding to Pik1 induces an allosteric conformational change that leads to enzyme activation. First, deletion of the LKU motif in Pik1 (res-idues 35-110) essentially ablates Pik1 catalytic activity (Fig. 5 and Table 2), suggesting that the LKU motif (in addition to Frq1 binding) is essential for lipid kinase activity, possibly because the N-terminal LKU motif must interact with and activate the C-terminal lipid kinase domain. Second, consistent with this notion, the LKU motif is highly conserved in both PtdIns 4-kinases and PtdIns 3-kinases (11,14,19). Third, also consistent with this proposal, in structures of PtdIns 3-kinase determined by x-ray crystallography, the LKU motif does seem to interact structurally with the catalytic domain (74 -76).
Our structure of the Frq1-Pik1 complex (Figs. 3 and 4) suggests how Frq1 might modulate the interaction between the LKU motif and the catalytic domain (Fig. 8). Frq1 interacts with two antiparallel ␣-helices of Pik1 (colored magenta in Fig. 8), producing a U-turn structure that causes the N-terminal and C-terminal ends to point in the same direction. We propose that this U-turn structure in Pik1 might orient the N-terminal LKU motif in close proximity to the C-terminal catalytic domain. Given that association between the LKU motif and the catalytic domain has been observed in PtdIns3-kinase (76), and given the findings we report here for the PtdIns 4-kinase Pik1, interaction of the LKU motif and the catalytic domain may be a general feature of all lipid kinases that contain such elements and are required for their catalytic activity. It is certainly plausible, in light of the structural information we obtained and as depicted in Fig. 8, that Frq1 binding to residues 121-174 in Pik1 (which are situated adjacent to the LKU motif) would achieve the goal of bringing the LKU motif in close proximity to the C-terminal catalytic domain by causing the conformational change that promotes this structural outcome. Future structural studies on the full-length Pik1 enzyme are now needed at the atomic level to further test this proposed activation mechanism and to define more rigorously the structural nature of the Pik1 active site.