Immunocytochemical Localization and Crystal Structure of Human Frequenin (Neuronal Calcium Sensor 1)*

Frequenin, a member of a large family of myristoyl-switch calcium-binding proteins, functions as a calcium-ion sensor to modulate synaptic activity and secretion. We show that human frequenin colocalizes with ARF1 GTPase in COS-7 cells and occurs in similar cellular compartments as the phosphatidylinositol-4-OH kinase PI4Kβ, the mammalian homolog of the yeast kinase PIK1. In addition, the crystal structure of unmyristoylated, calcium-bound human frequenin has been determined and refined to 1.9 Å resolution. The overall fold of frequenin resembles those of neurocalcin and the photoreceptor, recoverin, of the same family, with two pairs of calcium-binding EF hands and three bound calcium ions. Despite the similarities, however, frequenin displays significant structural differences. A large conformational shift of the C-terminal region creates a wide hydrophobic crevice at the surface of frequenin. This crevice, which is unique to frequenin and distinct from the myristoyl-binding box of recoverin, may accommodate a yet unknown protein ligand.

Frequenin (Frq), 1 or neuronal calcium-sensor 1, is a member of a family of related calcium-myristoyl-switch proteins that have been proposed to function as calcium-ion sensors. Members of this family include recoverin, GCAP, neurocalcin, visinin, and others (1). Recoverin and GCAP have been implicated in the control of recovery and adaptation in visual signal transduction. In vertebrate rod outer segments, GCAP apparently inhibits guanylate cyclase when the cytosolic concentration of Ca 2ϩ is high in the dark, whereas recoverin may inhibit rhodopsin kinase (2,3). Lowering the cytosolic Ca 2ϩ during light illumination attenuates the inhibitory activities of GCAP and recoverin, leading to an activation of these enzymes.
Frq, on the other hand, has attracted much attention because it may function as a calcium-ion sensor to modulate synaptic activity and secretion (4 -7). Drosophila Frq has been implicated in the facilitation of neurotransmitter release at neuromuscular junctions of third instar larvae (Drosophila melanogaster). Drosophila mutants that overexpress Frq show a facilitated neurotransmitter release that dramatically depend on the frequency of stimulation (4). Similarly, overexpression of a rat homolog of Frq in PC12 cells evokes an increased release of growth hormone in response to agonists like ATP (5). These results are consistent with the idea that Frq activity and regulated secretion are coupled.
More recently, it has been shown that the yeast homolog of Frq functions as a Ca 2ϩ -sensing subunit of the yeast phosphatidylinositol (PtdIns)-4-OH kinase, PIK1, a key enzyme in the phosphoinositide signaling system (6). In Saccharomyces cerevisiae, PIK1 participates in the mating pheromone-signal transduction cascade and regulates secretion at the Golgi. PIK1 is essential in yeast cells for normal secretion, Golgi and vacuole membrane dynamics, and endocytosis (7). Yeast PIK1 ts mutants exhibit severe protein trafficking defects and accumulate morphologically aberrant Golgi membranes (8). The aberrant Golgi morphology is strikingly similar to that found in yeast cells lacking a functional ARF1 GTPase (7). ARF1 has been implicated in multiple membrane trafficking events including the recruitment of the mammalian PIK1 homolog, PI4K␤, to the Golgi membrane (9).
Here we show that human Frq (HuFrq) colocalizes with ARF1 in COS-7 cells and occurs in similar cellular localizations as PI4K␤. In addition, in a further step toward understanding the cellular function of Frq, we report the crystal structure of unmyristoylated Ca 2ϩ -bound human Frq (HuFrq) refined to 1.9 Å resolution. This structure confirms that frequenins belong to the large family of myristoyl-switch Ca 2ϩ -binding proteins and reveals the architecture of the Ca 2ϩ -binding sites. Most importantly, comparative analysis of the HuFrq structure with those of neurocalcin and recoverin highlights a unique wide crevice and a solvent-exposed carboxyl terminus that could be responsible for ligand recognition and account for the broad substrate specificity among members of the family.
(GenBank TM /EBI accession no. AF186409). The restriction sites NcoI and NdeI were used for subcloning into expression vector pET-16b (Promega) to generate expression plasmid pET-HuFrq. All nucleotide sequences were verified by automated sequencing.
The expression plasmid pET-HuFrq was transformed into Escherichia coli strain BL21(DE3) (Novagen). Transformed cells were grown in Luria-Bertani (LB) medium containing ampicillin (100 g/ml) at 37°C. HuFrq expression was induced overnight at an A 600 of 0.8 with 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside and reached average yields of 20 mg/liter. Cells were harvested by centrifugation, resuspended in 20 ml of lysis buffer (50 mM HEPES, pH 7.4; 100 mM KCl; 1 mM EGTA; 1 mM dithiothreitol; 1 mM MgCl 2 ) per 1 liter medium, and lysed in a French pressure cell. Protamine sulfate was added to a final concentration of 0.1% for 10 min and then the lysate was cleared by centrifugation (40,000 ϫ g, 30 min, 4°C). The supernatant was filtered (0.45 m), adjusted to 1 mM CaCl 2 , and applied to a 15-ml phenyl-Sepharose CL4-B column (Amersham Pharmacia Biotech). The column was washed with buffer A (20 mM Tris/HCl, pH 7.9; 1 mM MgCl 2 ; 1 mM dithiothreitol) containing 1 mM CaCl 2 until A 280 was below 0.01, and then the protein was eluted with buffer A containing 2 mM EGTA. The eluate was applied to a 3-ml HiTrapQ column (Amersham Pharmacia Biotech). The column was washed with buffer A containing 60 mM NaCl and 1 mM CaCl 2 , and eluted with 120 mM NaCl in buffer A. The HuFrq-containing fractions were extensively dialyzed against water and concentrated to 10 mg/ml using microconcentrators (Pall Filtron). MALDI-TOF analysis of the purified HuFrq used the linear mode and a 337-nm nitrogen laser (Voyager-DERP BioSpectrometer work station, Perseptive Biosystems).

45
Calcium Binding Assay-Protein samples were resolved by SDSpolyacrylamide gel electrophoresis and transferred to nitrocellulose. The 45 Ca 2ϩ blotting was performed as previously described (10); autoradiographic exposure time was 2 days.
Immunofluorescence Microscopy-The polyclonal anti-HuFrq antibodies were raised in rabbit against purified HuFrq and purified by affinity chromatography on HuFrq-Sepharose 4B according to the manufacturer's instructions (Amersham Pharmacia Biotech); specificity analyses performed by enzyme-linked immunosorbent assay showed full recognition of the immunizing HuFrq but no recognition of either recoverin or neurocalcin (11). COS-7 cells were cultured at 37°C in Dulbecco's modified Eagle's medium (Life Technologies) supplemented with 10% (v/v) fetal calf serum (Life Technologies). They were grown to 60% confluency on poly-L-lysine-coated glass-coverslips. The cells were fixed with 4% (v/v) paraformaldehyde in phosphate-buffered saline for 15 min at room temperature, washed twice with phosphate-buffered saline and blocked with 1% (v/v) goat serum and 1% (w/v) bovine serum albumin in phosphate-buffered saline. Permeabilization used a blocking solution containing 0.3% (v/v) Triton X-100. Successive incubations with primary and secondary antibodies were carried out for 1 h at room temperature. Cells were washed in phosphate-buffered saline, and coverslips were mounted with Fluoromount GC (Southern Technologies). The cells were visualized and confocal images acquired using a confocal laser scanning microscope (Leica TCS NT). Primary antibodies were detected by species-specific cyanine dye-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories): Cy3-conjugated antibodies were used for visualization of anti-HuFrq and anti-PI4K␤ (Upstate Biotechnology) staining, whereas Cy2-and Cy5-conjugated antibodies were used for anti-␥-adaptin (Sigma) and anti-ARF1 (Santa Cruz Biotechnology) staining, respectively.
Crystallization and Data Collection-Crystals were obtained at 4°C using the vapor diffusion technique. Typically, 5 l of the HuFrq solution were mixed with 5 l of a reservoir solution made of 0.1 M sodium cacodylate, pH 6.5, 0.2 M NaAc, 30% polyethylene glycol M r 8,000 (Crystal Screen I, solution 28, Hampton Research). Under rare circumstances, three different crystal forms grew from this solution: needlelike (form A), thin plates (form B), and thick plates (form C). Form A crystals belong to the hexagonal space group P6 1/5 with unit cell dimensions: a ϭ b ϭ 82.2 Å and c ϭ 56 Å and contains one HuFrq molecule per asymmetric unit. Both form B and C crystals belong to the monoclinic space group P2 1 with unit cell dimensions: a ϭ 53.8 Å, b ϭ 55.5 Å, c ϭ 77.7 Å, ␤ ϭ 107.6°, and a ϭ 29.6 Å, b ϭ 105.2 Å, c ϭ 55.2 Å, ␤ ϭ 106.6°, respectively, and contain two HuFrq molecules per asymmetric unit. Crystals selected for data collection were briefly soaked into the reservoir solution supplemented with 10% (v/v) ethylene glycol, flash-cooled at 100 K in the nitrogen gas stream and stored in liquid nitrogen. No single crystal could be selected for form C. Data for forms A and B were collected on beamline ID14-EH2 of ESRF (Grenoble, France). Oscillation images were integrated with DENZO (12) and scaled and merged with SCALA (13). Amplitude factors were generated with TRUNCATE (13). Form A crystals were found to be twinned and the collected data could not be used.
Structure Determination and Refinement-Initial phases for form B crystals were obtained by molecular replacement using the structure of neurocalcin (14) (PDB code 1BJF) as a search model with the AMoRe package (15), giving a correlation coefficient of 36% and an R-factor value of 48% in the 15 to 4 Å resolution range. Rigid-body refinement, performed on each molecule with CNS (16) using data between 20 and 3 Å, gave an R-factor of 49%. For 2% of the reflections against which the model was not refined, R-free was 48%. The model was refined to 1.9 Å resolution using CNS, including bulk solvent and anisotropic B-factor corrections; the resulting 2Fo-Fc and Fo-Fc electron density maps were used to correct the model with the graphics program TURBO-FRODO (17). Solvent molecules automatically added using CNS were carefully examined on the graphics display. The final model comprises residues Asn 5 -Val 190 and Asn 5 -Gly 188 , respectively, for the two molecules in the asymmetric unit. High temperature factors and weak electron density are associated with residues 1-7, 49 -60, and 133-138. The average r.m.s.d. between the two HuFrq molecules is 0.6 Å for 182 C␣ atoms with the largest deviation (1.4 Å) for residue Gln 54 . The stereochemistry of the model was analyzed with PROCHECK (18); no residues were found in the disallowed regions of the Ramachandran plot. Data collection and refinement statistics are summarized in Table I. The coordinates and structure factors of HuFrq have been deposited with the Protein Data Bank (1G8I). Fig. 1B was generated by ALSCRIPT (19) and Figs. 4 -5 with SPOCK (20) and Raster3D (21).

RESULTS AND DISCUSSION
Chromosomal Localization-A search in the HTGS data bank with the HuFrq cDNA sequence revealed that the Hu-Frq gene is located on chromosome 9. A comparison of the publicly available chromosome 9 DNA sequence with the frequenin cDNA sequence showed that the exon-intron organization between Drosophila (4) and HuFrq genes has been conserved. Both open reading frames are interrupted by the same exon-intron borders and each are composed of same 8 exons (Fig. 1A). About 50 kilobases upstream of the first Frq exon are located several STS markers, e.g. DGS1924, A001W37, STSG22304, placing the Frq gene at 9q34.11. To our knowledge, a human disease has not been associated yet with this locus.
Molecular Identification and Immunocytochemical Localization-Previously, mammalian homologs of Drosophila Frq have been cloned (22,23). We have used this information to clone HuFrq from human first strand cDNA (Fig. 1A). The predicted HuFrq sequence contains 190 amino acids (Fig. 1B) with a theoretical monoisotopic mass of 21,865.91 and exhibits four EF hand motifs that represent potential Ca 2ϩ -binding domains. In HuFrq, as in other members of the family, the first of the four EF hand motifs is not likely to be a functional Ca 2ϩ -binding site as it lacks two Ca 2ϩ -coordinating amino acids. The HuFrq N terminus contains the consensus sequence MGXXX(S/T)K for myristoylation (24); hence it could be myristoylated like the rat homolog (5). The HuFrq sequence is 100% homologous to those of rat (5) and mouse (23) and it differs by a single amino acid from that of Xenopus (25). Remarkably, the yeast (6) and HuFrq protein sequences also show a high degree of conversation: 143 of 190 amino acids (75%) are either identical or correspond to conservative replacements (Fig. 1B).
In yeast, Frq has been shown to stimulate the activity of the PtdIns-4-OH kinase PIK1 (6), an enzyme that is essential for normal secretion, Golgi and vacuole membrane dynamics, and endocytosis (7). Xenopus Frq rescues a yeast Frq deletion mutant, indicating that Frq from higher eukaryotes is able to fulfill similar functions like yeast Frq (25). Consistent with the functional role of Frq in yeast are the phenotypes that have been described for Drosophila mutants (4) and for mammalian cells overexpressing Frq (5). In both cases, it appears that evoked secretion is stimulated by Frq (26).
The activity of PI4K␤, the likely mammalian homolog of yeast PIK1 (26), is recruited by the small GTPase ARF1 to the Golgi and contributes to the regulation of Golgi membrane dynamics and Golgi-dependent vesicle formation (9). Accordingly, in immunocytochemical experiments we have compared the immunostaining patterns obtained with anti-PI4K␤, anti-ARF1, and anti-HuFrq antibodies, respectively; overlapping The secondary structural elements are shown in gray with the four EF hands labeled EF1 to EF4. Ca 2ϩ -bound residues are indicated by gray dots and residues that are solvent exposed in the crevice by black dots. The N-terminal myristoylation site is underlined, and the mutated residues are indicated with black dots.
immunostaining reactions were observed (Fig. 2). For comparison, we also included in our investigations experiments with anti-␥-adaptin antibodies, a typical trans-Golgi network (TGN) marker. Paraformaldehyde-fixed COS-7 cells were first incubated with primary antibodies, e.g. polyclonal anti-HuFrq rabbit antibodies, monoclonal anti-␥-adaptin mouse antibodies, polyclonal anti-PI4K␤ rabbit antibodies, and polyclonal anti-ARF1 goat antibodies, respectively. Then, we used secondary Cy2-, Cy3-, or Cy5-labeled antibodies for immunocytofluorescent staining and localization of ␥-adaptin, HuFrq and PI4K␤, and ARF1, respectively, using confocal microscopy (Fig. 2). The anti-HuFrq antibodies revealed a pattern with a crescent of staining on one side of the nucleus and some punctuate staining within the cytoplasm (Fig. 2A). Staining could be eliminated by preincubation of the primary antibody with the immunizing HuFrq protein (not shown), indicating that the observed immunofluorescence is generated by HuFrq-specific antibodies. A similar staining pattern was obtained with ␥-adaptin (Fig. 2B), which in double-labeling experiments colocalized with the HuFrq-immunostaining pattern (Fig. 2C). Previously, ␥-adaptin has been shown to be localized in the TGN and the late endosomes (27). The double-immunostaining patterns also indicate a colocalization for ␥-adaptin and PI4K␤ (Fig. 2, D-F) in agreement with a recent report (28). Finally, we coimmunostained the COS-7 cells with anti-HuFrq and anti-ARF1 antibodies (Fig. 2, G-I). Again, we observed a crescent of staining on one side of the nucleus (presumably the TNG) and some punctuate immunostain extending to the plasma membrane, which colocalized with the HuFrq immunostain (Fig.  2I). The results indicate similar subcellular distributions for HuFrq, ARF1, and PI4K␤. The colocalization is consistent with the proposal that frequenin proteins modulate PtdIns-4-OH kinase activity both in yeast and in mammalian cells and thus may have similar regulatory functions in secretion and Golgi membrane dynamics.
Calcium-binding Properties-We mutated EF hands EF2, EF3, and EF4 of HuFrq together or in pairwise combinations utilizing in vitro mutagenesis. In all four cases, we mutated the amino acid residues at the -X position as previously described for Drosophila Frq (Fig. 1B) (4). Accordingly, we generated four HuFrq mutants: E81V/T117A/T165A (Frq 2,3,4 ), E81V/T117A (Frq 2,3 ), E81V/T165A (Frq 2,4 ), and T117A/T165A (Frq 3,4 ). Bacterial lysates containing approximately equal amounts of each HuFrq mutant were blotted onto nitrocellulose. The blot was incubated with 45 Ca 2ϩ to investigate the Ca 2ϩ -binding capacity of the HuFrq mutants in comparison to wild-type HuFrq (Fig.  3). The results showed that wild-type HuFrq yielded the highest 45 Ca 2ϩ signal. We noted for the recombinant HuFrq mutants with pairwise mutations attenuated 45 Ca 2ϩ -signals of comparably reduced intensity. The pairwise HuFrq mutants each contained a single intact EF hand (EF2 in Frq 3,4 , EF3 in Frq 2,4 , and EF4 in Frq 2,3 ), yet they bound Ca 2ϩ with high affinity; this suggests that EF hands EF2, EF3, and EF4 not only are functional in HuFrq but also are independent from each other. Previously, it was shown that single mutations in yeast Frq1 EF hands did not display a temperature-sensitive phenotype like the quadruple mutant in the frq1-I ts allele, consistent with our observations. By contrast, the triple mutant Frq 2,3,4 did not bind 45 Ca 2ϩ to a significant extent; hence in HuFrq, EF hand 1 does not constitute a high affinity Ca 2ϩbinding site in HuFrq, as predicted earlier from sequence analysis.
Overall Structure-The crystal structure of HuFrq was solved by the molecular replacement method using neurocalcin (14) as a search model and was refined to 1.9 Å resolution. The structure consists of residues Asn 5 -Val 190 with good stereochemistry; clear electron density maps could be observed for all structural elements (Fig. 4A). As predicted, HuFrq shares the typical ␣-helical fold found in homologous proteins, with overall dimensions of 35 ϫ 60 ϫ 40 Å. HuFrq contains 10 helices labeled A to J (Fig. 4B). Consistent with the recently proposed NMR-derived model of yeast Frq (29), the four EF hands, which all present the typical architectural short ␤-strand, come in two pairs: an N-terminal pair (EF1, EF2) and a C-terminal pair (EF3, EF4), which are connected by the hinge loop 93-97 and related by an approximate 2-fold axis. The hinge loop conformation positions the four EF hands on one side of the molecule in a tandem linear array. This structural arrangement is similar to the ones seen in other members of the myristoyl-switch Ca 2ϩ -binding proteins, e.g. neurocalcin, recoverin, and GCAP (1). It is, however, different from the dumbbell arrangement found in e.g. calmodulin and tropinin C (30). The HuFrq structure contains three calcium ions bound to EF hands EF2, EF3, and EF4. The Ca 2ϩ coordination in HuFrq is virtually identical to recoverin and neurocalcin. The side chains of residues x, y, and z in the 12-residue loop each provide an oxygen atom, as does the main chain carbonyl of residue y. Two additional oxygen atoms come from the conserved glutamic side chain of residue z and a water molecule.
Structural Comparison with Homologous Proteins-For residues 8 -175, the HuFrq overall fold is similar to that of neurocalcin (14) with an r.m.s.d. value of 1.3 Å for 164 C␣ atoms (Fig. 4C). Extended comparison with the structure of recoverin (31) yields a r.m.s. deviation of 1.7 Å for only 123 C␣ atoms. These values indicate that the HuFrq structure is more closely related to that of neurocalcin. This reflects the high sequence identity existing between the two proteins, which is 61% for the whole molecule, compared with only 41% between HuFrq and recoverin. However, whereas neurocalcin dimerizes in solution (14), no dimeric assembly could be observed for HuFrq in concentrated solution or in the crystals (data not shown); this suggests that the distinct oligomeric assembly of these proteins may be related to their biological functions.
Novel Hydrophobic Crevice-The dominant and striking feature of the HuFrq structure is the large positional shift of the C-terminal helix J compared with its position in the recoverin and neurocalcin structures. Indeed, helix J has moved by 45°as a hinge-type rigid-body motion, with the residue pair Asp 176 -Pro 177 acting as a pivot point, to adopt an original position that is well ordered in our structure (Fig. 5, A and C). As a result,  4. Quality of the map and overall fold of HuFrq. A, stereo view of the 1.9 Å resolution omit 2Fo-Fc averaged electron density map, contoured at 1, showing part of helix J. The coordinates of this region were omitted and the protein coordinates refined by simulated annealing before the phase calculation. B, stereo ribbon diagram of HuFrq with bound Na ϩ and Ca 2ϩ ions. Secondary structure elements are indicated as in Fig. 1. Labels Ca2, Ca3, and Ca4 refer to the Ca 2ϩ ions bound to EF hands EF2, EF3, and EF4, respectively. Functionally important side-chain residues in the vicinity of helix J are shown as blue/orange bonds with red oxygen and blue nitrogen atoms. Hydrogen bonds are shown as dotted lines. C, stereo overlay of HuFrq (yellow/ cyan) and neurocalcin (orange/green) oriented as in B. Helix J is highlighted in cyan for HuFrq and green for neurocalcin. the helix is tightly packed against loop ␣E-␣F where it establishes numerous polar and hydrophobic interactions, consistent with an average mobility similar to that of other structural elements in the molecule. As a consequence of helix J motion, a large hydrophobic crevice, with dimensions 30 ϫ 15 ϫ 15 Å, is unmasked on the HuFrq face that is opposite to the four EF hands (Fig. 5A). The crevice is made of 46 residues provided by 8 helices, of which helices D, E, and F contribute to the floor of the crevice, helices C, G on one side and helix J on the other side contribute to the walls, and helices H and B contribute to the top and bottom, respectively. Hence, a quarter of the HuFrq molecule contributes to the structure of the crevice.
The major structural significance of the helix J repositioning for ligand binding is obvious from the complete exposure of the hydrophobic surface region. This new conformation dramatically alters the solvent-accessible molecular surface of the molecule when compared with those of neurocalcin or recoverin. In HuFrq, a large patch of hydrophobic residues line the crevice and still are fully accessible to the solvent (Fig. 5A), whereas they are sequestered into the neurocalcin and recoverin protein cores (Fig. 5, B and C). Actually, fragments of the polyethylene glycol used for crystallization are found well ordered within the crevice, where they could mimic an interacting ligand/membrane partner and shield the crevice from the solvent. Presence of the detergent octyl-glucoside was a requisite to solubilize and reduce aggregation of yeast Frq (29). In HuFrq, the unique disordered surface loop, loop 133-137, along with the C-terminal part of helix C that also presents high mobility, are located at the periphery of the crevice and may also play a role upon ligand/membrane recognition. Is this ␣-helix conformational shift specific to HuFrq? To our knowledge, such a large positional shift in the C-terminal portion is novel and unique to HuFrq compared with other members of the family and exemplifies how a structurally related fold can exhibit a distinct binding site. Within the crystal asymmetric unit, a sodium ion is bound in the N-cap of helix J and connects residue Ala 175 in the ␣I-␣J loop that precedes helix J of one molecule to residues Thr 17 , Thr 20 , and Phe 22 of the second molecule via the carbonyl atoms; however, there is no evidence for an active role of this ion in the ␣-helical shift. Instead, sequence differences in the C-terminal ends of proteins of the family may explain the conformational features specific to the HuFrq structure. Indeed, seven residues are conserved in the C-terminal region of HuFrq compared with other members: Ala 182 , Asp 187 , Gly 188 , and Leu 189 , which are exposed at the periphery of the crevice and may have a role for ligand recognition, and Ser 184 , Tyr 186 , and Asp 187 , which are located on the opposite face where they tightly anchor helix J to the protein core (Fig. 4B). A recent report on a yeast Frq model, designed from a combination of partial NMR data and homology modeling based on the recoverin structure, might suggest a similar location of helix J in the yeast homolog (29). Hence, the fact that Drosophila Frq possesses a glycine doublet in place of the conserved residue pair Tyr 186 -Asp 187 appears most surprising because these substitutions would be expected to destabilize helix J. Finally, a Kv channel-interacting protein has recently been identified as a novel member of the calcium-binding protein family (32), and its C-terminal sequence presents high similarity with the C-terminal region of HuFrq. This suggests that a hydrophobic crevice similar to that found in HuFrq may exist and be involved in the binding of the cytoplasmic N termini of Kv4 ␣-subunits.
The Calcium Myristoyl Switch-HuFrq may share the molecular mechanisms of calcium-myristoyl switch proteins as seen from the solution structures of recoverin, where the myristoyl group is sequestered in a deep hydrophobic box in the Ca 2ϩ -free state (33) and ejected into the solvent to interact with a lipid bilayer membrane in the Ca 2ϩ -bound state (34). HuFrq possesses two discrete conformations at the hinge point Lys 7 -Leu 8 in the N-terminal region, which therefore is a flexible arm. Residues Gly 42 and Gly 96 are the two hinge points that distinguish between the Ca 2ϩ -free and Ca 2ϩ -bound conformations in recoverin, and these two residues are conserved in HuFrq. Of the five helices that participate in the formation of the myristoyl box in recoverin, four (helices B, C, E, and F), line the large crevice in HuFrq. In addition, most of the hydrophobic residues that are recruited to accommodate the myristoyl group in recoverin are conserved in HuFrq and are solventexposed at the crevice surface. In HuFrq, residue Gly 96 , which FIG. 5. Structural differences between HuFrq and homologous proteins. Molecular surface of HuFrq (A), viewed down the large hydrophobic crevice (orange) and oriented as in Fig. 2., neurocalcin (B) and recoverin (C) (same orientation). Secondary structure elements are labeled. The C-terminal helices, helix J in HuFrq (cyan) and in neurocalcin (green) and helices J and K in recoverin (green) are displayed. is one of the two hinge points in recoverin, establishes van der Waals contacts with helix J, suggesting that a conformational shift at this position might modify the position of this helix. Would HuFrq use the same structural switch as seen in recoverin, the shape and/or size of its hydrophobic crevice would be dramatically altered, a modification that could reflect a regulatory mechanism. Further biochemical experiments must await production of myristoylated HuFrq to test this hypothesis, although recent studies suggest that the presence of a N-myristoyl group in yeast Frq does not affect its overall structure whether Ca 2ϩ is present or not (29).
Rat Frq, as opposed to recoverin, neurocalcin, and hippocalcin, showed a Ca 2ϩ -independent interaction with membranes (35), suggesting that the presence of the hydrophobic crevice, so far a feature unique to the Frq, may account for this difference in calcium dependence. As well, yeast Frq binds to and activates PIK1 in a Ca 2ϩ -independent manner (6), a feature also consistent with our proposal of the hydrophobic crevice as a functionally important binding site. In contrast, the Ca 2ϩ -enhanced interaction of yeast Frq with membranes (29) would suggest that in Ca 2ϩ -free yeast Frq the shape of the hydrophobic crevice might be altered.
In summary, we have cloned HuFrq from human cDNA and analyzed its cellular localization in COS cells. HuFrq was overexpressed in E. coli, purified to homogeneity, and crystallized in the Ca 2ϩ -bound state. This Ca 2ϩ -bound HuFrq structure is a critical step toward identification of a physiological protein partner using biological experiments. Further crystallographic investigations will help identify the interactions involved in complex formation and conformation.