Inositol 1,4,5-trisphosphate 3-kinase A associates with F-actin and dendritic spines via its N terminus.

The consequences of the rapid 3-phosphorylation of inositol 1,4,5-trisphosphate (IP(3)) to produce inositol 1,3,4,5-tetrakisphosphate (IP(4)) via the action of IP(3) 3-kinases involve the control of calcium signals. Using green fluorescent protein constructs of full-length and truncated IP(3) 3-kinase isoform A expressed in HeLa cells, COS-7 cells, and primary neuronal cultures, we have defined a novel N-terminal 66-amino acid F-actin-binding region that localizes the kinase to dendritic spines. The region is necessary and sufficient for binding F-actin and consists of a proline-rich stretch followed by a predicted alpha-helix. We also localized endogenous IP(3) 3-kinase A to the dendritic spines of pyramidal neurons in primary hippocampal cultures, where it is co-localized postsynaptically with calcium/calmodulin-dependent protein kinase II. Our experiments suggest a link between inositol phosphate metabolism, calcium signaling, and the actin cytoskeleton in dendritic spines. The phosphorylation of IP(3) in dendritic spines to produce IP(4) is likely to be important for modulating the compartmentalization of calcium at synapses.

The consequences of the rapid 3-phosphorylation of inositol 1,4,5-trisphosphate (IP 3 ) to produce inositol 1,3,4,5-tetrakisphosphate (IP 4 ) via the action of IP 3 3-kinases involve the control of calcium signals. Using green fluorescent protein constructs of full-length and truncated IP 3

3-kinase isoform A expressed in HeLa cells, COS-7 cells, and primary neuronal cultures, we have defined a novel N-terminal 66-amino acid F-actin-binding region that localizes the kinase to dendritic spines.
The region is necessary and sufficient for binding Factin and consists of a proline-rich stretch followed by a predicted ␣-helix. We also localized endogenous IP 3 3-kinase A to the dendritic spines of pyramidal neurons in primary hippocampal cultures, where it is co-localized postsynaptically with calcium/calmodulin-dependent protein kinase II. Our experiments suggest a link between inositol phosphate metabolism, calcium signaling, and the actin cytoskeleton in dendritic spines. The phosphorylation of IP 3 in dendritic spines to produce IP 4 is likely to be important for modulating the compartmentalization of calcium at synapses. Inositol 1,4,5-trisphosphate (IP 3 ) 1 signals in neurons are terminated predominantly via the action of calcium-stimulated IP 3 3-kinases (IP3kins) (1). These enzymes produce 1,3,4,5tetrakisphosphate (IP 4 ), which does not gate IP 3 receptor calcium channels and may have signaling properties of its own (2)(3)(4). Brain has more IP3kin activity than other tissues (5), and this is due to the concentration of the A isoform in the dendrites of principal neurons, such as cerebellar Purkinje cells, hippocampal CA1 pyramidal cells, and dentate gyrus granule cells (6 -8). Targeted disruption of IP3kin A isoform (IP3kinA) produces mice with enhanced long term potentiation but normal spatial learning (9). By contrast, injecting IP 4 into pyramidal neurons prior to tetanic stimulation strengthens the long term potentiation via a mechanism that involves the enhancement of voltage-activated calcium channels by IP 4 (10).
IP3kins are members of the inositol polyphosphate kinase family, which consists of proteins sharing a conserved C-terminal catalytic region (4,11). Animal IP3kins are the only inositol polyphosphate kinases thought to be regulated by calcium, both via direct interaction with calmodulin and by phosphorylation by calmodulin-dependent kinase II (CaMKII) (12)(13)(14). The B isoform of IP3kin, which is expressed in most cells and in glia in brain (5,15), is localized to internal membranes of the endoplasmic reticulum and Golgi apparatus, where it is poised to regulate IP 3 concentrations near calcium stores (16). In contrast, nothing is known about targeting mechanisms for isoform A in neurons.
Based on purification of the bovine and rat brain enzyme, IP3kinA activity and protein have been described as cytosolic (17)(18)(19), in contrast to type I 5-phosphatase that appears to be mostly membrane-bound (20). Immunocytochemical studies employing antibodies against the purified rat brain enzyme revealed IP3kinA to be very concentrated on membranes and throughout the cytosolic matrix of dendritic spines (7,8). In this study, we report that the targeting of IP3kinA to the actin cytoskeleton in transfected HeLa cells and COS-7 cells and of endogenous IP3kinA to the dendritic spines of neurons is determined by the 66 N-terminal amino acids. We also observed a co-localization of IP3kinA and CaMKII isoforms in synaptically mature hippocampal cultures. This suggests that the association of F-actin and IP3kinA in hippocampus subserves a neuronal specific function of modulating highly localized intracellular calcium signals.

EXPERIMENTAL PROCEDURES
Molecular Biology-Clone CP16 (21), containing the full-length rat cDNA for IP3kinA, was used as a polymerase chain reaction template, and various primer sets were designed to amplify parts of the cDNA that were cloned into pEGFPC3 and -N1 vectors (CLONTECH) for expression as green fluorescent fusion proteins. All polymerase chain reaction was carried out as described previously (13), using Pfu Turbo polymerase (Stratagene) in the presence of 10% Me 2 SO. The full open reading frame (aa 1-489) was amplified and cloned into pEGFPC3 with the following primer set: full sense (GAGAAAGTCCATGACCCTGCC-CGGACACCCGAC) and full antisense (GAGAGGATCCTCATCTCTC-AGCCAGGTTGGGC). The underlined regions depict HindIII and BamHI sites incorporated into the primers. All other primers contained the same 5Ј GAGA sequence, HindIII/BamHI restriction sites, and as required contained added or deleted start or stop codons or extra nucleotides to preserve reading frame, according to the manufacturer's instructions. These primer sequences are as follows: (restriction sites and added start or stop codons not shown): aa 66 -489, 66.sense (GT-CCCTAACGGGCTCCCG), and full antisense, aa 1-33, full sense and 33.antisense (TCACTCCCCGACGCTCCGGCG); aa 34 -66, 34.sense (CTGCGCCGCTCTTCGAAGCG) and 66.antisense (GACTTGTCCCCC-ACGCCGCTT); aa 1-66, full sense and 66.antisense; aa 34 -459, 34.sense and full antisense; aa 109 -489, 109.sense (TCGCACCTGCA-GCAGCCA) and full antisense; aa 155-489, 155.sense (ACCAGCGAA-GACGTGGGGCAG) and full antisense. Each of these constructs was tested in both the N-terminal and C-terminal GFP vectors to control for possible GFP-induced artifacts. Two of these constructs aa 1-489 and aa 66 -489 were also cloned into EcoRII/SalI sites of the pCMS-EGFP vector (CLONTECH) for expression without the GFP fusion attached. The open reading frame for rat ␥-actin was amplified by polymerase chain reaction from a cDNA library made from hippocampal cultures (a gift of Dr. Hilmar Bading, Laboratory of Molecular Biology, Cambridge) and cloned into pEGFPC3 at the HindIII/BamHI sites. All constructs were tested by restriction enzyme digest before use, and some (those used in the figures) were sequenced across the multicloning site to confirm identity and reading frame.
Secondary Protein Structure Prediction-Three Internet-based programs (nnPredict, SOPM, and PSA) were used to predict the secondary structure of the actin-association region of IP3kinA. These can be found at the BCM search launcher web site: searchlauncher.bcm.tmc.edu/.
F-actin Co-sedimentation-The 1-66 actin association region (N66) of IP3kinA fused to the bacterial protein NusA was produced by subcloning into the pET43A vector (Novagen) at the BamHI/SalI sites. This vector also incorporates His 6 before and after the multicloning site. A 1:50 dilution of Escherichia coli strain Rosetta BL21(DE3) (Novagen) harboring the construct was grown for 3 h at 37°C followed by cooling to 18°C and overnight induction in the presence of 1 mM isopropyl-1thio-␤-D-galactopyranoside. The following day cells were harvested by centrifugation and disrupted in BugBuster reagent (Novagen) in the presence of a protease inhibitor mixture, according to the manufacturer's instructions. The cleared, soluble extract was mixed batchwise with Talon beads (CLONTECH), loaded into a column, and eluted with imidazole. Fractions containing purified protein were pooled and dialyzed overnight against 20 mM Tris, pH 8.0, filtered, and then frozen as aliquots at Ϫ20°C before use in co-sedimentation experiments. Actin co-sedimentation was carried out in actin binding buffer (20 mM Tris, pH 8.0, 100 mM NaCl, 2 mM MgCl 2 , 1 mM ATP, 0.5 mM dithiothreitol, 1 mM EGTA, and 0.1 mM CaCl 2 ), using F-actin purified from skeletal muscle acetone powder (a gift of John Kendrick-Jones, Laboratory of Molecular Biology, Cambridge, UK) or from human platelets (Cytoskeleton, Denver, CO). NusA-N66 or NusA alone was mixed with various concentrations of F-actin and incubated for 30 min at 30°C before centrifugation at 200,000 ϫ g for 1 h in an Airfuge (Beckman). Pellets were resuspended in the starting assay volume, and equal volumes of supernatants and pellets were loaded onto 10% SDS-polyacrylamide electrophoresis (SDS-PAGE) gels; protein was visualized by Coomassie Blue.
The F-actin binding curve was constructed by varying the concentration of skeletal muscle actin between 1.0 and 20 M in the presence of a constant amount of NusA-N66 fusion protein (2.65 M). Over this range of actin concentrations, the portion appearing in the pellet as F-actin varied between 39% at the lowest concentration and 79% at the highest. Because the effective F-actin concentration changes in this way, at each concentration of F-actin, we multiplied the concentration of actin added to the assay by the portion appearing in the pellet as F-actin to obtain the apparent F-actin concentration depicted on the binding curve. The risk of nonspecific trapping of protein in the pellet increases at high F-actin concentrations; however, even at the highest F-actin concentrations used, trapping of the truncated (non-F-actin binding) fusion protein fragments was negligible. Gels were dried and scanned, and then the density of the gel bands was determined using the gel-plotting macro of NIH image software (version 1.62). Only the largest (full-length) gel band for NusA-N66 was used for quantification. Curve fitting was done with KaleidaGraph software using the general form: y ϭ m1⅐(M0 ϩ m2).
Cell Culture and Transfection-Hippocampal cultures were prepared from neonatal rats (age 12-24 h), as described previously (22). Hippocampi were dissociated with papain (Worthington), grown for 2 days in Neurobasal media with B27 supplement (Life Technologies, Inc.) and 10% horse serum, and then maintained thereafter in serum-free media for up to 5 weeks. Cerebellar granule cells were prepared by trypsin digestion of postnatal day 6 -8 rat cerebella and then maintained as described for hippocampal cells, except that the KCl concentration in the media was increased to 25 mM and cytosine arabinoside (Sigma) was present between 24 and 72 h after plating. Transfection of neurons was achieved with a modified calcium-phosphate technique (23) that used a 1-h exposure of the cells to the DNA precipitate. Neurons were examined between 1 and 14 days after transfection. HeLa cells were maintained in DMEM plus 10% fetal calf serum and transfected by the same calcium-phosphate technique as for neurons, except that the exposure to the precipitate was for 8 h, with a 2-min glycerol shock at the end. COS-7 cells were maintained in DMEM plus 10% fetal calf serum and transfected with the Fugene reagent (Roche Molecular Biochemicals). HeLa and COS-7 cells were examined 1 day after transfection.
Immunocytochemistry and Confocal Microscopy-All steps were carried out at room temperature. Hippocampal cells grown on poly-Dlysine-coated coverslips for 5-35 days were washed twice with PBS containing 0.5 mM Ca 2ϩ and 0.68 mM Mg 2 followed by incubation in 4% freshly depolymerized paraformaldehyde, 0.1 M sodium phosphate, pH 7.4, for 30 min. All subsequent steps were carried out in PBS. Cells were permeabilized in 0.1% Triton X-100 for 5 min, washed briefly, and blocked for 1 h in 2% normal donkey serum, 2% fish skin gelatin (Sigma). Coverslips were then incubated for 1 h in various antibodies diluted in 2% donkey serum. Visualization was with affinity-purified fluorescein isothiocyanate or Cy3-coupled donkey anti-rabbit or antimouse Fab fragments (Jackson ImmunoResearch, West Grove, PA). F-actin was visualized with Oregon Green 488-or Alexa 568-phalloidin (Molecular Probes, Eugene, OR). Coverslips were mounted on slides using Prolong antifade (Molecular Probes). Images of 1.5-1.8-m confocal sections were captured on a Zeiss LSM 510 confocal microscope using a 63ϫ objective. Before images of double-labeled cells were collected, the microscope pinholes were aligned with reference to a transfected neuron expressing GFP-IP3kinA and then double-stained with rabbit anti-IP3kinA (fluorescein isothiocyanate channel) and mouse anti-GFP (Cy-3 channel). Images were processed with Adobe Photoshop software (Adobe, Mountain View, CA).
Immunoprecipitation-All steps were carried out at 4°C. Adult rat forebrain was frozen on dry ice and then homogenized in ice-cold 20 mM Tris, pH 7.4, 100 mM KCl, 0.5% Triton X-100, 0.1 mM dithiothreitol, 50 mM NaF, 0.1 mM Na 3 VO 4 containing phenylmethylsulfonyl fluoride and benzamidine (each 1 mM), leupeptin, chymostatin, and antipain (each 1 M). The homogenate was extracted for 30 min on a rotating wheel, followed by centrifugation at 48,000 ϫ g for 1 h at 4°C. The extract (supernatant) was frozen at Ϫ70°C. In some experiments, the thawed extract was centrifuged an additional time at 200,000 ϫ g for 1 h before being used for immunoprecipitation. Immunoprecipitation for one gel lane consisted of 150 l of extract that was precleared for 1-2 h with 20 l of packed protein G-Sepharose (Amersham Pharmacia Biotech) before overnight incubation with 3 l of IP3kinA antiserum or with 3 l of normal rabbit serum. The next day beads were washed once in brain homogenization buffer without inhibitors containing 5 mg/ml bovine serum albumin, then twice in the same buffer without albumin, and then once in Tris-buffered saline before elution of beads by boiling in SDS-PAGE sample buffer. Samples were separated on 10% acrylamide minigels and then transferred to nitrocellulose. Following overnight incubation in primary antibodies, proteins were visualized by chemiluminescence (25). In one experiment, the protein profiles obtained after immunoprecipitation with either IP3kinA antiserum or control antiserum were compared by staining with Coomassie Blue. These appeared identical except for a 53-kDa band appearing in the IP3kinA lane only.
Antibodies and Drugs-The polyclonal antiserum made in rabbit against purified rat brain IP3kinA was described previously (18) and used for immunohistochemistry and Western blots (6,8); it was used at 1:1000 for immunofluorescence and 1:1500 for Western blots. Mouse monoclonal antibodies against the ␣-subunit of CaMKII (catalog number MAB3119) and against synapsin (catalog number MAB355) were from Chemicon. Mouse monoclonal antibody against the ␤-subunit of CaMKII (Clone CBb-2) was from Zymed Laboratories Inc. Monoclonal antibody against actin (clone AC-40) was from Sigma. Latrunculin A (LTA) (Molecular Probes) was made as a 5 mM stock in Me 2 SO and diluted 1:1000 into the culture media.

RESULTS
Determination of an Actin Association Region in IP3kinAtransfected HeLa Cells-Our initial experiments to examine the targeting of IP3kinA, a neural specific protein, utilized HeLa cells as a more convenient vehicle for localizing the multiple constructs generated. Full-length IP3kinA (amino acids 1-459) fused N-terminal to GFP appeared to associate with the plasma membrane and cytoskeleton of HeLa cells (Fig. 1A). This localization was not an artifact of the GFP tagging, since the GFP could also be placed C-terminal to GFP (not shown); furthermore, untagged, full-length IP3kinA, when expressed in HeLa cells and stained with a polyclonal antibody to IP3kinA, produced a similar staining pattern (not shown). Double staining with fluorescent phalloidin revealed a high degree of colocalization with F-actin (Fig. 1B).
A series of truncated IP3kinA GFP constructs were made and expressed in HeLa cells. We first tested a 66 -459-amino acid construct, which corresponds to a clone called C5 that was previously shown to produce a fully Ca 2ϩ /calmodulin-sensitive enzyme in bacteria (21). This construct, when fused to GFP, appeared completely cytosolic when transfected into HeLa cells (not shown), suggesting that the first 66 amino acids are necessary for targeting to the cytoskeleton. Further studies established that the first 66 amino acids were sufficient to target GFP to F-actin (Fig. 1, C and D). The cytosolic localization of construct 34 -459 (Fig. 1E) suggested that the first 66 amino acids were necessary for co-localization with F-actin. Indeed, when the stretch of 66 amino acids was cut in half, neither 1-33 ( Fig. 1G) nor 34 -66 (Fig. 1I) could efficiently target GFP to F-actin.
Although in all cases placing the GFP at either end of the construct produced similar results, some subtle differences were observed. First, the 1-33 construct (but never the 34 -66 constructs) sometimes appeared to localize partially with Factin in cells with very high expression levels, but this was only observed when the GFP was placed C-terminal to 1-33. Second, constructs with the GFP at the C terminus had easier access to the nucleus; construct 66 -459 with GFP placed at the N terminus was excluded from the nucleus but entered the nucleus when GFP was placed at the C terminus. Constructs 109 -459 and shorter entered the nucleus freely, regardless of the site of GFP tagging (not shown). Third, in a subset of cells with very high expression levels, the presence of the actin association region appeared to stabilize or promote the formation of F-actin, as judged by our ability to pick out some of the transfected cells on the basis of enhanced phalloidin staining. This phenomenon was explored further in the transfected hippocampal cultures (see below and Fig. 4). The amino acid sequence of the actin-targeting region was examined through the use of computer programs that predict secondary structure (Fig. 1,  bottom). Three programs agreed that the stretch 31-49 has a high likelihood of forming an ␣-helix. This contrasts to the most N-terminal stretch of 27 amino acids, which contains 6 prolines.
Detergent Extraction of Living Cells-Another criterion for an association with the actin cytoskeleton is resistance to extraction with the detergent Triton X-100. We therefore expressed the GFP-(1-66) and GFP-(66 -459) constructs in HeLa and COS-7 cells, and we then extracted them with 1% Triton X-100 for 5 min at room temperature before fixation and staining with fluorescent phalloidin. In some cases, cells were pretreated with the actin-disrupting drug LTA (5 M) for 15 min before extraction. The results of some of these experiments are depicted in Fig. 2. In both HeLa and COS cells, the 1-66 construct appeared to be highly resistant to detergent extraction (Fig. 2C), whereas the 66 -459 construct was washed away so efficiently that it was impossible (HeLa cells) or difficult (COS cells; Fig. 2G) to determine which cells had been transfected with 66 -459. Moreover, when HeLa cells expressing the 1-66 construct were pretreated with LTA for 15 min prior to extraction, no green fluorescent cells remained (not shown). These experiments demonstrate that the 1-66 region is likely to associate with F-actin directly and that disruption of F-actin is sufficient to liberate a cytosolic (Triton X-100-extractable) protein.
Co-sedimentation with F-actin-We next determined if the first 66 amino acids of IP3kinA bound F-actin directly in a purified system. The actin association region was difficult to express in bacteria. Our initial attempts to produce His 6tagged 1-459 or 1-66 using the pQE-30 vector in the XL-10 gold strain of E. coli produced no detectable fusion protein in either the soluble or inclusion body fractions. To overcome this problem, we fused the 1-66 region (N66) downstream of the solubility-enhancing NusA protein in the PET43A vector and expressed the construct in the BL21(DE3) strain carrying a plasmid coding for rare tRNAs. This vector also incorporates one His 6 sequence between NusA and N66 and another His 6 after N66 to facilitate purification. Although this strategy resulted in substantial full-length NusA-N66 fusion protein purified from the bacterial extracts (about 47% of the total purified in our best preparations), the remainder of the purified protein was truncated (Fig. 3A, lanes 5 and 7). Our co-sedimentation experiments (Fig. 3A) confirmed that only the highest molecular weight band (containing the full N66 stretch) was able to bind F-actin (Fig. 3A, lanes 6 and 8) and also that NusA itself was not an F-actin-binding protein (lanes 2 and 4). Because skeletal muscle actin is known to have different properties than non-muscle actin when expressed in neurons (26), we tested both skeletal muscle and non-muscle (platelet) actin; the results with either type of F-actin appeared identical, at least qualitatively (lanes 6 and 8).
In order to estimate the affinity and stoichiometry of our fusion protein for F-actin, we attempted to carry out binding assays whereby we varied the concentration of NusA-N66 against a fixed concentration of skeletal muscle F-actin, used at just above the critical concentration (ϳ1 M F-actin). However, because so much of the NusA-N66 protein was truncated (and unable to bind F-actin), we were unable to raise the concentration high enough in the assay to saturate the F-actin-binding sites (2.6 M was the maximal concentration of apparently full-length NusA-N66 achieved in our assays). Then we tried titrating various F-actin concentrations against a maximal concentration of NusA-N66 (2.6 M). F-actin binding assays done this way are theoretically possible, but there are a number of caveats that complicate the interpretation of the data (see below). Nevertheless, it was possible to generate a curve that resembled saturable binding and allowed us to estimate a rough stoichiometry of between 2 and 4 mol of actin per mol of fusion protein and to estimate the K d to be 2.7 M (Fig. 3B). At least two caveats apply. First, the lowest point on the curve was obtained near or below the critical concentration (1 M actin was added to the assay, and only 39% appeared in the pellet as F-actin). Second, even at the highest F-actin concentration tested (20 M), some of the full-length NusA-N66 fusion protein (0.98 M) remained in the unbound fraction when theoretically all of the fusion protein should pellet with the actin if the actin concentration is raised high enough. Perhaps some the apparently full-length fusion protein was subtly modified so that it could no longer bind F-actin. The most reliable data on our curve lie in the points between 1 and 5 M F-actin, where the average stoichiometry was 2.3 mol of F-actin per mol of fusion protein. These data serve as a rough guide for future studies using full-length, untruncated fusion protein in more conventional binding assays.
Expression of GFP-IP3kinA in Cultured Hippocampal Cells-Primary cultures of hippocampal cells (6 -16 days in vitro) were transfected using a calcium phosphate procedure, fixed, and examined between 18 h and 14 days later. GFP-IP3kinA assumed the morphology of the F-actin in each of the various cell types in which it was expressed. The highest levels of expression were found in astrocytes, where fluorescence occurred mainly on spikey structures, stress fibers, and also cortical actin (Fig. 4). Transfected astrocytes also produced the clearest confirmation of the observations made in HeLa and COS-7 cells that the 1-66 region of IP3kinA sometimes stabilizes or promotes actin polymers. In Fig. 4C, phalloidin staining for F-actin is clearly increased compared with neighboring untransfected astrocytes. In the same experiment, the cultures were also transfected with GFP fused to ␥-actin. Surprisingly, cells overexpressing actin did not show increased phalloidin staining compared with their neighbors (Fig. 4, B and D), thus serving as controls for the specificity of the IP3kinA F-actinpromoting effect. It is important to note that most of the transfected non-astrocytic cells did not exhibit enhanced staining for F-actin and that there was a clear correlation between a cell having very high GFP-IP3kinA fluorescence and its having enhanced levels of F-actin.
In our initial attempts to transfect neurons, we used short transfection times (20 -48 h) and young neurons (3-9 days in culture). In these experiments, we found that (as was the case in HeLa cells, COS-7 cells, and astrocytes) the GFP-IP3kinA fluorescence was co-localized with F-actin, as evidenced by virtually complete co-localization with fluorescent phalloidin. Fluorescence was observed in a wide variety of F-actin-rich structures: some spikey, some string-like, and some punctate. The great variety of cell types and structures associated with GFP-IP3kinA in young cultures made it difficult to generalize about which structures were labeled, but they no doubt in- cluded microspikes, growth cones, and dendritic filopodia.
While our "acute" transfection strategy worked well when the cultures were young and had not yet established synapses, we encountered great difficulty transfecting the neurons in the cultures once they had matured. If we transfected cells older than about 14 days in culture, the main cells transfected were glia and aspiny neurons. This presented a considerable problem, since IP3kinA is expressed very late in development, during postnatal weeks 2 and 3 and is concentrated mainly in the dendritic spines of mature pyramidal neurons (6,8). With "acute" transfection of 3-week-old cultures, the vast majority of our GFP-IP3kinA expression was ectopic and only rarely did GFP-IP3kinA appear in a cell type that would normally express IP3kinA in vivo.
We later adopted a strategy (26) whereby we transfected the neurons during the 1st week in culture and then waited much longer (5-14 days) before examining the cells. Most transfected neurons appeared to continue to express the GFP constructs for at least 2 weeks, even after synaptic connections had established, and the levels of green fluorescence in transfected neurons tended to be lower than for astrocytes. In cultures 2 weeks old or older, scores of punctate-labeled transfected pyramidal neurons were present on each coverslip. In these cells, GFP-IP3kinA was observed in a pattern suggestive of dendritic spines (Fig. 5, A and B). Likewise, in synaptically mature cultures of cerebellar granule neurons (29 days in vitro), GFP-IP3kinA appeared to be associated with synapses, as judged by their apposition to axons of nearby green cells (Fig. 5C, arrow,  Ax). As in HeLa cells and COS-7 cells, truncation of the 66amino acid F-actin binding sequence produced a cytosolic protein (Fig. 5D). The GFP-(1-66) construct expressed in hippocampal neurons also concentrated in actin-rich structures such as putative spines (Fig. 5, E and F).
Co-localization of Endogenous IP3kinA with F-actin at the Synapse-Although IP3kinA has been mapped in hippocampus at both the light and ultrastructural levels (6 -8), there are no reports of its localization in primary hippocampal cultures. In our initial attempts to visualize the IP3kinA in 1-2-week-old cultures using an antibody, we observed no specific staining. IP3kin activity in rat brain increases ϳ10-fold during the 1st 3 postnatal weeks (27), due to a massive increase in IP3kinA expression between weeks 1 and 3. When we grew our cultures for 2.5 weeks or longer, we could observe a subset of neurons that stained intensely with the IP3kinA antibody; this suggested that the developmental expression of the IP3kinA in hippocampal cultures recapitulated expression in vivo. IP3kinA immunoreactivity was confined mainly to very large spiny neurons, most of which had a pyramidal cell soma. Many other neurons and all glia were unlabeled, suggesting that the antibody recognized isoform A selectively. The staining was highly punctate (Fig. 7, A and C) and covered the whole somatodendritic tree; additionally, some cells exhibited a lesser cytosolic staining that usually included the nucleus.
The punctate staining was associated with F-actin, as judged by double-staining with phalloidin (Fig. 6A, yellow). When the cultures were treated for 5 min with the actin-disrupting drug LTA (5 M), the phalloidin staining was greatly reduced relative to controls, and the punctate F-actin staining pattern was largely abolished (Fig. 6B, green). Likewise, following a 5-min LTA treatment, IP3kinA staining had lost its punctate pattern and appeared more cytosolic (Fig. 6B, red). Some LTA-resistant patches, usually located on distal dendrites, remained after the LTA treatment, and they corresponded to patches of IP3kinA more resistant to disruption. These experiments suggested that the punctate pattern of IP3kinA staining is due chiefly to its ability to bind F-actin directly.
Dendritic spines are actin-rich structures in excitatory neurons abutting presynaptic terminals. The punctate pattern we observed for IP3kinA in our cultures suggested that, similar to the situation in brain, endogenous IP3kinA in cultures was concentrated in dendritic spines, which corresponded to the punctate, actin-rich structures. Accordingly, we double-stained with the presynaptic marker synapsin (Fig. 6C, green). A substantial portion of the punctate structures (Fig. 6C, red) indeed localized to bona fide dendritic spines, i.e. they were adjacent to but not co-localized with synapsin (Fig. 6C, arrows). When we treated the cultures with LTA for 5 min, almost all of the IP3kinA immunoreactivity lost its punctate pattern (Fig. 6D,  red) and had moved away from the presynaptic terminals. Previous studies have demonstrated that LTA treatment causes spine morphology to collapse and F-actin and its associated binding proteins to disperse; however, LTA does not disrupt the core postsynaptic protein complex consisting of, among other things, PDZ-containing proteins and the NMDA  1, 2, 5, and 6) or non-muscle actin from platelets (N/M, lanes 3, 4, 7, and 8) was added to the fusion proteins, incubated in F-actin stabilization buffer, and then pelleted at 200,000 ϫ g. Pellets (P) were resuspended in the starting assay volume, and equal volumes of supernatants (S) and pellets were analyzed by SDS-PAGE followed by staining with Coomassie Blue. Co-sedimentation is observed with full-length NusA-N66 (lanes 5 and 7, pellets) but not with truncated fusion protein (lanes 5 and 7, smaller bands) or with NusA protein alone (lanes 1-4), and no difference in co-sedimentation properties is observed between the two kinds of actin. B, binding curve obtained by varying the concentration of F-actin in the presence of a constant concentration of 2.65 M fulllength NusA-N66. subtype of glutamate receptor (24,28). The almost completely cytosolic, non-punctate pattern observed for IP3kinA following a short LTA treatment argues that F-actin (as apposed to, for example, calmodulin or PDZ proteins) is the major, if not sole, element anchoring IP3kinA near the synapse.
Co-localization of IP3kinA and CaMKII Isoforms-Previous biochemical results obtained in stimulated rat brain cortical slices established an important role for CaMKII in activating IP3kinA via its phosphorylation at threonine 317 in the sequence (12)(13)(14). We therefore co-localized endogenous IP3kinA and CaMKII isoforms in synaptically mature hippocampal cultures (Fig. 7). A monoclonal antibody specific for the ␣ isoform stained most neurons in the cultures but not glia. A substantial portion of the immunoreactivity appeared to be cytosolic and stained axons as well as dendrites but, unlike IP3kinA, was excluded from the nucleus (Fig. 7, A versus C). Every cell that stained prominently for IP3kinA was also positive for CaMKII␣. However, we observed a great deal of cell-to-cell and culture-to-culture variation in the degree of punctate CaMKII␣ staining. This is probably due to the fact that this protein is known to translocate from the cytosol to the synapse in response to synaptic activity (29,30), and thus we were observing differences in the spontaneous synaptic activity of different neurons. The neuron depicted in Fig. 7, A and C, represents one end of a continuum of patterns observed. In this cell, IP3kinA is highly concentrated in putative dendritic spines, whereas CaMKII␣ appears almost completely cytosolic. Nevertheless, high power images (Fig. 7, B and D) indicated that some IP3kinA spines also contained CaMKII␣. Overall, roughly half of the IP3kinA-positive punctate structures also contained punctate staining for CaMKII␣.
By contrast to the mainly cytosolic CaMKII␣, CaMKII␤, like IP3KinA, is thought to be associated with F-actin (31). Indeed, when we double-stained for IP3kinA and CaMKII␤, we observed many examples of neurons in which the two proteins appeared highly co-localized. One such example is depicted in Fig. 7, E and G. As was the case with the ␣ isoform, the ␤ isoform also exhibited heterogeneity in its degree of punctate staining. Despite its proposed ability to bind F-actin, some CaMKII␤ in some cells appeared cytosolic. This may be because this protein occurs in at least four splice forms (32), and some of the splicing occurs in the suggested actin-binding region (31). In summary, IP3kinA was usually found to be co-localized with CaMKII, its endogenous activator, and this co-localization occurred more often with the ␤ than the ␣ isoform.
Co-immunoprecipitation of IP3kinA with Actin and CaMKII-To determine if IP3kinA could interact directly with either of the CaMKII isoforms, we performed immunoprecipitation experiments using detergent extracts from rat forebrain. In our initial experiments, we immunoprecipitated IP3kinA from a 0.5% Triton X-100 extract that had been centrifuged at 48,000 ϫ g at 4°C. In these experiments, we indeed observed an association of IP3kinA with actin and both CaMKII isoforms (Fig. 8, left lanes 3). We wondered if this result could be explained by an incomplete removal of F-actin fragments from the extract. We therefore subjected the extract to further centrifugation at 200,000 ϫ g before immunoprecipitation with the IP3kinA antiserum. It was important to establish that the 200,000 ϫ g supernatant still contained IP3kinA, so we blotted this extract (and the starting 48,000 ϫ g extract) with the IP3kin antibody (Fig. 8, top). Although removal of the remaining F-actin by centrifugation depleted IP3kinA levels roughly 50%, substantial IP3kinA protein remained soluble and thus available for immunoprecipitation. Substantial amounts of G-actin and both CaMKII isoforms were also present in the post-200,000 ϫ g supernatant (Fig. 8, right lanes 1). When we carried out a direct comparison of co-immunoprecipitation from the medium speed versus high speed extracts, the apparent association of IP3kinA with actin and CaMKII isoforms disappeared (Fig. 8,  lanes 3). The most likely interpretation of these data is that IP3kinA does not directly associate with CaMKII␣, or CaMKII␤, and that any association with these proteins occurs only as part of a larger F-actin-containing complex. DISCUSSION This is the first report establishing a link between the N terminus of IP3kinA and F-actin, which in brain is highly concentrated in dendritic spines. The data presented here stress the importance and novel function of an N-terminal amino acid sequence in the localization of a major IP 3 -metabolizing enzyme and contrast to the quite different multiple targeting mechanisms that have been reported (20) for the family of inositol polyphosphate-5-phosphatases. The targeting sequence of IP3kinA consists of an N-terminal proline-rich stretch followed by an 18-amino acid predicted ␣-helix that binds F-actin with high affinity. The N-terminal region of IP3kin isoform B has practically no sequence identity with the F-actin-binding region of IP3kinA. However, 5 of the first 25 amino acids in isoform B are prolines, followed by a 13-residue predicted ␣-helix. Recently, we expressed the first 66 amino acids of IP3kin isoform B as a GFP fusion protein and found it to be cytosolic and Triton X-100-extractable. 2 The recently cloned isoform C, which is not expressed in brain or regulated by CaMKII (33), also shows no sequence similarity to isoform A in the N-terminal region. Thus we suggest that association with F-actin occurs selectively in isoform A, presumably to subserve a neuronal specific function of modulating highly localized intracellular calcium signals.
Proline-rich stretches are present in many actin-binding proteins (34) and often promote or stabilize the formation of Factin. Although our GFP fusion proteins required the full 66amino acid N-terminal amino acids to produce complete colocalization with F-actin, we did observe that merely the first 33 amino acids (containing the proline-rich region) could sometimes confer a lesser degree of actin association when fused upstream (but not downstream) of GFP. This might be explained by an ability of GFP to partially substitute for the ␣-helix stretch. The 34 -66 region alone fused to GFP showed no tendency to associate with F-actin. Proline-rich stretches are also the binding sites for Src-homology 3 (SH3) domains (35). The minimal core requirement for SH3 domain binding is the PXXP motif, two of which occur in the IP3kinA F-actinbinding region. Perhaps the targeting to F-actin localizes IP3kinA near some type of synaptic scaffolding, and it is feasible that SH3 domain-containing proteins and F-actin might compete for the same binding site on IP3kinA.
The ability of IP3kinA to bind F-actin may explain the discrepancy between biochemical studies, reporting IP3kin activity and protein to be cytosolic (18,19), and electron microscopy studies, which demonstrate an enrichment in both the cytosolic matrix and membranes of dendritic spines (7,8). Relevant here is the previous, extensive evidence that calpains cleave the 53-kDa full-length kinase to produce a ladder of catalytically active, calcium-regulated soluble fragments (18,19). Our results show that truncation of merely the first 33 amino acids (which would be liberate a protein of predicted size of 49.6 kDa) destroys actin binding and produces a cytosolic protein that probably has access to the nucleus. Thus, most or all known calpain-generated fragments would be cytosolic. A recent proteomic cataloguing of proteins associated with the postsynaptic 2 M. J. Schell and R. F. Irvine, unpublished observations. FIG. 8. Co-immunoprecipitation of IP3kinA with CaMKII is observed from medium-speed but not high speed brain extracts. Rat brain was homogenized, extracted in detergent 0.5% Triton X-100 at 4°C, and then centrifuged at 45,000 ϫ g as described under "Experimental Procedures." The supernatant from this centrifugation was used for experiments depicted in the left column, whereas in the right column the extract has been subjected to an additional spin at 200,000 ϫ g before use. The top lanes depict gel bands representing the relative amounts of IP3kinA present in each extract, as determined by Western blot of identical exposure times. The bottom lanes are from an immunoprecipitation (IP) experiment where lane 1 represents 10% of the input extract (15 l), lane 2 is immunoprecipitation using nonimmune rabbit serum, and lane 3 is immunoprecipitation using rabbit antiserum against IP3kinA. These lanes were probed with monoclonal antibodies against actin (2nd row), CaMKII␣ (3rd row), or CaMKII␤ (bottom row), and exposures of the left and right columns were for identical times. density did not report the presence of IP3kinA (36). Taken together, this suggests that IP3kinA, although highly enriched at synaptic sites in dendrites, is only dynamically or loosely associated with membranes and cytoskeleton. Mild detergent (such as the 0.1% Chaps used during its purification (19)), rapid actin disruption by the drug LTA, or limited proteolysis are sufficient to generate a cytosolic protein.
Although IP3kinA is known to be concentrated in dendritic spines in brain (7,8), this is the first report of the localization of endogenous IP3kinA in cultured hippocampal neurons. To visualize endogenous IP3kinA in our cultures, it was important to grow the cells for longer than 2 weeks, a time when synaptic junctions have begun to stabilize. A previous study using hippocampal neurons made from IP3kinA knockout mice reported that the global calcium signals of these cells were indistinguishable from cultures made from control animals (9). Their cultures were grown for 7-10 days before use, and one reason for the lack of effect on the calcium signal in the knockouts may simply have been that the cultures weren't grown for a time sufficient for IP3kinA to be expressed fully in the controls.
IP3kins have a variety of cellular functions (4), one of which is to remove IP 3 by a calcium-regulated reaction which would be of particular significance in dendritic spines. A downstream target for the IP 4 produced by IP3kinA, if any, is unclear; however, the activation of voltage-gated calcium channels by IP 4 has been implicated in a number of studies (10,37,38). IP 4 produced in a spine would be expected to shift the effect of a receptor's signal away from calcium stores and toward calcium entry. Recently, Penner and colleagues (39) have demonstrated a mechanism whereby IP 4 inhibits the IP 3 5-phosphatase, raising IP 3 levels and enhancing Ca 2ϩ entry. If such a protection mechanism occurred in dendritic spines, IP 4 could act as a kind of memory of recent activity in a spine and thereby enhance the effects of subsequent pulses of IP 3 .
Another putative target for IP 4 is the Ras/Rap GTPase-activating protein GAP1 IP4BP (recently renamed RASA3), a protein purified and cloned by virtue of is high affinity and selectivity for binding IP 4 (40). GAP1 IP4BP is expressed in most or all neurons (41) and is localized to the plasma membrane of cells by virtue of its ability to bind inositol lipids (42). It is not known if GAP1 IP4BP is present in dendritic spines, but if it were near sites of IP 4 generation, the predicted result would be to stimulate the GTPase activity of synaptic Ras proteins (43).
In dendritic spines, calcium stores play key roles in modulating synaptic plasticity (44,45). For example, calcium released selectively from the spine apparatus of Purkinje cells is crucial for the establishment of long term depression (46). The presence of IP3kinA near these structures, where it co-localized with CaMKII, its activator, suggests that IP3kinA functions to modulate the spatial and temporal dynamics of spine calcium. CaMKII␤ functions as an F-actin targeting molecule that localizes CaMKII ␣/␤ hetero-oligomers to spines (31). Consistent with this, we observed a higher degree of co-localization with ␤ than with ␣.
F-actin motility and turnover in spines is dynamic (47,48), and thus many mechanisms to regulate the localization and activity of IP3kinA can be envisioned. For example, the rapid depolymerization of F-actin in response to glutamate receptor activation (49) would be expected to cause IP3kinA to move away from synaptic sites, similar to what we observed by depolymerizing F-actin with LTA. Such a synaptic de-localization of IP3kinA would allow subsequent bursts of IP 3 to persist longer, and this could be envisioned as a type of synapseselective memory mechanism. A similar enhancing effect on calcium signals would be expected if calcium-activated proteases cleaved IP3kinA, allowing the kinase to diffuse out of the spine. Indeed, calpain inhibitors can block long term potentiation (50). In conclusion, the role of IP3kin in a dendritic spine is likely to involve the restriction of IP 3 -generated calcium signals to individual synapses. Moreover, the rapid, localized synthesis of IP 4 may in turn have complex effects on cytosolic calcium and on the molecular mechanisms that control learning and memory.