Three-dimensional reconstructions of calcium/calmodulin-dependent (CaM) kinase IIalpha and truncated CaM kinase IIalpha reveal a unique organization for its structural core and functional domains.

Studies of the structural organization of calcium/ calmodulin-dependent protein kinase IIalpha (CaM KIIalpha) and truncated CaM KIIalpha by three-dimensional electron microscopy and protein engineering show that the structures consist of 12 subunits that are organized in two stacked hexameric rings with 622 symmetry. The body of CaM KIIalpha is gear-shaped, consisting of six slanted flanges, and has six foot-like processes attached by narrow appendages to both ends of the flanges. Truncated CaM KIIalpha that lacks functional domains has a structure that is very similar to the body of CaM KIIalpha. Thus, the functional domains reside in the foot-like processes, and the association domain comprises the gear-shaped core. The ribbon diagram of the bilobate structure of CaM KI fits nicely in the envelope of the foot-like component and indicates that the crevice between the two lobes comprising the functional domains is near the middle portion of the foot. The clustering of the functional domains provides a favorable arrangement for the autophosphorylation reaction, and the unusual arrangement of the catalytic domain on extended tethers appears to be significant for the remarkable functional diversity of CaM KIIalpha in cellular regulation.

Studies of the structural organization of calcium/ calmodulin-dependent protein kinase II␣ (CaM KII␣) and truncated CaM KII␣ by three-dimensional electron microscopy and protein engineering show that the structures consist of 12 subunits that are organized in two stacked hexameric rings with 622 symmetry. The body of CaM KII␣ is gear-shaped, consisting of six slanted flanges, and has six foot-like processes attached by narrow appendages to both ends of the flanges. Truncated CaM KII␣ that lacks functional domains has a structure that is very similar to the body of CaM KII␣. Thus, the functional domains reside in the foot-like processes, and the association domain comprises the gearshaped core. The ribbon diagram of the bilobate structure of CaM KI fits nicely in the envelope of the foot-like component and indicates that the crevice between the two lobes comprising the functional domains is near the middle portion of the foot. The clustering of the functional domains provides a favorable arrangement for the autophosphorylation reaction, and the unusual arrangement of the catalytic domain on extended tethers appears to be significant for the remarkable functional diversity of CaM KII␣ in cellular regulation.
Calcium/calmodulin-dependent protein kinase (CaM K) 1 II phosphorylates Ser and Thr residues in numerous proteins. Its relatively high concentration in brain tissue and its lack of specificity indicate that it has an important role as a kinase in many aspects of neuronal function. Its substrates are involved in neurotransmitter synthesis and release; carbohydrate, lipid, and amino acid metabolism; transcriptional, translational, and cytoskeletal regulation; calcium homeostasis; and receptor and channel function (for reviews, see Refs. 1 and 2).
At least four distinct genes that encode isozymes of CaM KII are selectively expressed in different tissues (2). Expression of the ␣ and ␤ isoforms is restricted primarily to the nervous system, and the ␣ isoform is found only in neurons. The abundance of ␣ and ␤ isoforms is both anatomically and developmentally regulated in the nervous system (3,4), and it has been proposed that the subunit composition of holoenzymes may influence the targeting of the enzyme to distinct subcellular sites, such as the postsynaptic density (5)(6)(7). For example, recent investigations have shown that the ␤ isoform is preferentially bound to F-actin in dendritic spines and in the cell cortex and that the ␣ isoform is targeted to these locations when it is co-expressed with the ␤ isoform (8).
Studies have implicated the ␣ isoform of CaM KII in neuronal cell function. Electrophysiological and behavioral tests in mice carrying a null mutation for the ␣ isoform showed that this enzyme has a major role in control of neuronal excitability (9) and spatial learning (10). Indeed, this isoform is thought to have a key role in the long lasting synaptic enhancement denoted long term potentiation (11)(12)(13).
CaM KII is dependent on Ca 2ϩ /calmodulin for activation, and its autophosphorylation has an important role in regulating its activity. Autophosphorylation involves Thr-286 in the autoregulatory domain of the enzyme that results in a Ca 2ϩ /calmodulinindependent activity (14,15) and a major increase in Ca 2ϩ /calmodulin binding affinity (16 -18). These phenomena may provide for the unique functions of CaM KII for signal processing in neurons (17, 19 -22). In this regard, a point mutation at Thr-286 blocked the autophosphorylation of CaM KII␣, and the mice containing this mutation exhibited deficits in long term potentiation and spatial learning similar to those exhibited by mice bearing the null mutation of this isoform (23).
The deduced amino acid sequences of the ␣ and ␤ isoforms show that they have extensive homology; the primary difference between them is the addition of two variable inserts in the ␤ subunit C-terminal to the calmodulin-binding domain (24). Enzymatic studies of the ␣ and ␤ isozymes of CaM KII indicate that they have similar functions and show little difference in substrate specificity (1). Initial biophysical studies indicated that CaM KII isolated from either rat forebrain or cerebellum is a multisubunit complex that contains varying ratios of the ␣ and ␤ subunits. The forebrain (M r 650,000) and cerebellar (M r 508,000) enzymes consist predominantly of the ␣ and ␤ isoform, respectively (6,25), and electron microscopy studies support this distribution (26). The images produced by rotary shadowing indicated that the holoenzyme consists of 8 -10 spokes radiating from a central core 100 Å in diameter to give an overall dimension of 330 Å. The distal ends of the spokes had a bulbous shape, and these features were described as "lollipops." It has been proposed that these lollipops comprise the regula-tory/catalytic region because they were reported to bind Ca 2ϩ / calmodulin. It was proposed from visual analysis of the individual images that CaM KII isolated from the cerebellum and forebrain is organized as octamers and decamers, respectively. In addition, immune electron microscopy studies employing ␣-specific antibodies indicated that only the decamers were labeled and, consequently, that the two oligomeric complexes are comprised of either ␣ or ␤ subunits (26). However, this proposal is at variance with other studies that indicated that the holoenzyme is comprised of a mixture of both subunits (6,27). It has been proposed that stain images of the homo-oligomeric skeletal muscle CaM KII consist of hexameric rings 100 Å in diameter, but these images did not display the lollipops. From electron microscopy studies and biophysical data, it was proposed that the structure is a dodecamer consisting of two stacked hexameric rings (28).
Even though the studies of Kanaseki et al. (26) indicate that there might be significant structural differences between these isoforms, it is not possible to relate these differences to the three-dimensional structure of the complexes. In this regard, the proposed binding of calmodulin at the distal ends of the lollipops was not supported by a difference map analysis, and the noisy particle images are difficult to interpret and do not permit an unambiguous location of calmodulin. Moreover, the proposed distribution of the two forms in specific regions of the brain is uncertain because the distribution may be influenced by the relative affinity of the two forms for the carbon support.
Unfortunately, the functional diversity of CaM KII cannot be related to its structure. As part of our initial structural studies of the CaM KII isoforms, we have determined the three-dimensional structures of the CaM KII␣ and truncated CaM KII␣ (residues 315-478) complexes employing three-dimensional electron microscopy. The latter subunit lacking the catalytic and regulatory domains was previously shown to self-associate into an oligomeric complex (29). Comparisons of the reconstructions reveal a unique organization of the structural core and functional domains of CaM KII␣ that appear to contribute to its remarkable functional attributes.

EXPERIMENTAL PROCEDURES
Protein Preparations-The cDNA of the coding sequence of the rat CaM KII␣ subunit was cloned into the pBakPak baculovirus expression vector (CLONTECH). Recombinant baculovirus was prepared and recombinant enzyme was purified as described previously (18,29). CaM KII␣ prepared in this manner is an oligomer with a sedimentation coefficient of 17 S and a Stokes radius of 85 Å (29). The specific activity of CaM KII␣ preparations utilized in the present study was approximately 19 mol/min/mg, and the protein was Ͼ90% pure when analyzed by SDS-polyacrylamide gel electrophoresis.
To construct the association domain, a polymerase chain reaction fragment was produced beginning at amino acid 315 and extending to the native stop codon of the ␣ subunit (amino acid 478). This fragment was cloned into the Hta version of the Bac-to-Bac vector (Life Technologies, Inc.), and recombinant virus was produced as described (29). The truncated ␣ protein is fused to an N-terminal His tag and linker sequence that increases the predicted molecular mass to 22,189 from 18,477 Da. The proteins isolated from infected Sf21 cells were concentrated by ammonium sulfate precipitation (29) and purified by Ni 2ϩ -NTA affinity chromatography as described by the manufacturer (Qiagen). The His-tagged fusion protein was eluted from the Ni 2ϩ -NTA resin with 100 mM imidazole and used immediately for imaging. The oligomer purified by this means had a sedimentaion coefficient of 10 S and a Stokes radius of 49 Å (29). The Ni 2ϩ -NTA purification provided protein that was Ͼ90% pure as analyzed by SDS-polyacrylamide gel electrophoresis and was at suitable concentrations (15 g/ml) for electron microscopy imaging.
Electron Microscopy-A 6-l sample of CaM KII␣ (15 g/ml) in 10% (w/v) glycerol, 50 mM Hepes, 1 mM dithiothreitol, 1 mM EDTA, 500 mM KCl (pH 7.4) was added to the carbon side of a freshly prepared carboncoated Formvar grid, and the excess was removed by wicking with bibulous paper. The bound particles were washed 3 times with 6 l of 0.25% methylamine tungstate stain containing 10 g/ml bacitracin as a wetting agent; the excess stain solution was removed by wicking with bibulous paper. Stain images were acquired at ϫ 50,000 with a JEOL JEM 1200 electron microscope operating at 100 kV using conventional irradiation procedures. The 50°tilted images were recorded with an underfocus of ϳ1 m at the center of the stage, and the corresponding untilted images were obtained at ϳ0.5 m underfocus. Non-tilted images for the three-dimensional projection alignment method and iterative reconstruction were obtained by applying the preparation at 15 g/ml to the carbon grid and recorded at ϳ1 m underfocus as described above. tCaM KII␣ in 20 mM Tris-HCl, 100 mM KCl, 100 mM imidazole, 10 mM ␤-mercaptoethanol, 10% (w/v) glycerol (pH 8.5) was added to the carbon film as described above, and the non-tilted images were recorded at an underfocus of ϳ0.5 m as described previously. For cryo-electron microscopy, a 3-l sample of CaM KII␣ (70 g/ml) in 50 mM HEPES, 500 mM KCl, 1 mM EDTA, 1 mM EGTA, 1 mM dithiothreitol (pH 7.4) was deposited on a glow-discharged carbon-coated holey grid, and the excess was removed by blotting with filter paper. The grid was rapidly cooled in liquid ethane, specimens were kept below Ϫ170°C in a Gatan cold holder, and the images were recorded at ϫ 50,000 with ϳ1.7 m underfocus at an exposure of ϳ9 e/Å 2 using Kodak SO 163 film.
Digitization and Particle Extraction-Micrographs were digitized using an Eikonix 1412 scanner with a 12-bit dynamic range and a pixel size of 5.7 Å on the specimen scale. Power spectra from the micrograph were analyzed for astigmatism and drift. Micrographs showing significant frost, astigmatism, or drift were rejected. Particles were selected in 64 ϫ 64-pixel boxes using the SUPRIM software package (30). The particle data sets consisted of 5166 CaM KII␣ (stain), 4758 CaM KII␣ (frozen hydrated), and 2914 tCaM KII␣ (stain) particle images. These latter images were digitized with a Zeiss SCAI scanner at a pixel size of 2.8 Å on the specimen scale.
Image Alignment and Random Conical Tilt Reconstruction-Average images of the characteristic views were obtained by extracting particles representative of the 6-fold and the two 2-fold views and subjecting them to a reference-free alignment using the SPIDER software (31). A random conical tilt (RCT) reconstruction was performed using the 50°-tilted images corresponding to the 6-fold view by the weighted back projection of a random conical tilt series (32) using the SUPRIM image processing system (30). Symmetry was not applied to the reconstruction.
Stain dimensional projection alignment and iterative reconstruction was employed to compute the structure from untilted stain and ice images. A set of quasi-uniformly distributed reference projections (34) using the non-symmetrized RCT reconstruction as the initial model was generated within ϭ 0 -90°and ⌽ ϭ 0 -360°with angular steps of 2°. After the initial angular alignment, a reference-based translation alignment of the boxed images was performed by cross-correlating the particle images with the projection of the preliminary structure to the orientation of the particle image (35). The newly computed structure that appeared to have 622 symmetry was used in the subsequent alignment and iterative reconstructions. The resulting stain structure of CaM KII␣ was used as the model to generate the reference projections for the initial frozen hydrated reconstruction. tCaM KII␣ projection images were generated from the stain CaM KII␣ model to which a threedimensional mask was applied to remove the exterior foot processes. Two passes of refinement achieved a stable resolution value for the reconstructions. The resolution values of the reconstructions were measured by the Fourier shell correlation with a Fourier ring criterion of 0.67 (36). The structures were Fermi-filtered with a temperature parameter of 0.01 (37) to the resolution value of the reconstruction, and the frozen hydrated reconstruction was corrected for the contrast transfer function of the electron microscope (38). The solid shaded structures were thresholded to a volume that corresponds to their approximate molecular weight, and the images were rendered using the Explorer software (NAG Inc., Downer's Grove, IL). The wire frame presentation of tCaM KII␣ was obtained using SYNU software (39).

Images of CaM KII␣ in Stain and Ice and of tCaM KII␣-A
gallery of the molecular images shows the 6-, 2-, and 2-fold views and their corresponding average images for CaM KII␣ in stain (Fig. 1A) and ice (Fig. 1B) and for tCaM KII␣ (Fig. 1C). The 6-fold views have the shape of a hexagonal ring consisting of six blobs of higher density. The two 2-fold side views of the molecule consist of two side-by-side rings or a single ring similar to that in the 6-fold view (Fig. 1). It is apparent that the images are noisy and of low contrast, making it difficult to distinguish between the views for the purpose of selecting the 6-fold views of CaM KII␣ for the initial stain RCT reconstruction. The second row of images was enhanced by applying a Butterworth filter (40) to the corresponding "raw" images above. Consequently, the arrangement of the peripheral density adjacent to the higher contrast central image becomes more apparent, making it possible to distinguish between the ring-like 6-and 2-fold views. The 6-fold and 2-fold views have peripheral density that forms a ring or brackets above and below the central ring, respectively (Fig. 1A). The tCaM KII␣ also exhibits similarly shaped images of central density as CaM KII␣ but lacks the peripheral density (Fig. 1C). Accordingly, the RCT structure was used as a model to align the non-tilted stain specimen, and the corresponding structure was similarly employed as a model to compute the structure of the frozen hydrated specimen. Finally, the peripheral density associated with the stain structure was truncated with a three-dimensional mask to give a core structural model to align the images of tCaM KII␣. The obtainment of these structures using an initial stain RCT model to align non-tilted stain and ice images (33) further extends the utility of this approach in three-dimen- sional electron microscopy (41)(42)(43)(44).
Three-dimensional Reconstructions-The stain and ice structures, their projections (Fig. 2), and their corresponding average images (Fig. 1) show good concordance, indicating that the reconstructions are reliable. Even though the non-symmetrized RCT structure is noisy because of the limited number of 6-fold views in the data set ( Fig. 2A), this model produced a structure that appeared 622 symmetric, and consequently, 622 symmetry was imposed on the model-based reconstructions (Fig. 2, B-D).
The CaM KII␣ structures consist of a gear-shaped body that is approximately 140 Å in diameter with a height of 100 Å. There are six openings on the 2-fold axis that lead into a central cavity that traverses the length of the core. The six slanted cogs of the gear (flanges) have 12 foot-like processes attached by a narrow appendage (six each to the outer edge at the top and bottom of the flanges). The heel of the foot is tilted approximately 30°away from the 6-fold axis of the structure so that the foot resides peripheral to the core. The appendages double the height of the core to 200 Å. These appendages are responsible for the ring of density surrounding the central core and extend the diameter of the image in 6-fold view to ϳ200 Å (Fig. 1, A  and B). The lollipop extensions were reported to form a ring with a diameter of 330 Å in images of CaM KII isolated from the brain (26). The significant size difference of the length of these extensions may be related to differences in the structures of the wild type and ␣ isoform. The protein density distributions in the projections of the CaM KII␣ central core and tCaM KII␣ are similar and indicate that the structures consist of 12 subunits arranged in two sets of six subunits that form stacked hexagonally shaped rings in agreement with the structure proposed for the skeletal muscle complex (28).
The structure of tCaM KII␣ (M r 221,000), which lacks the N-terminal catalytic and calmodulin-binding domains (Fig. 3), is very similar to the central core structure of CaM KII␣ (compare Fig. 2, B and D). Consequently, we conclude that the core structure of CaM KII␣ comprises the association domain and that the foot processes comprise the catalytic domain (Fig. 3). The foot-like extensions at the top and bottom of the CaM KII␣ core indicate that the subunits are arranged in a tail-to-tail fashion in both structures (Fig. 3).
A gallery of 1-pixel thick (5.7 Å) slices cut normal to the 6-fold axis of the structures shows their protein density distribution (Fig. 4). The six triangular to rod-shaped components that comprise the slices of the strands in the core structure have their highest density and connectivity in slices that reside about halfway toward the middle of the core, and the strands rotate approximately 60°from the top to the bottom of the complex. At an increased threshold the lower density middle slice disappears, indicating that two subunits are juxtapositioned near the middle of the core (data not shown). We propose that the CaM KII␣ dodecamer may form as a hexamer of dimers.
The structures of CaM KII␣ and tCaM KII␣ are consistent with the physicochemical characterization of these complexes (29). Gel filtration chromatography, SDS-polyacrylamide gel electrophoresis, and sedimentation velocity studies showed that the subunits form stable complexes of 600 kDa (CaM KII␣) and 204 kDa (tCaM KII␣) with subunit composition of approximately 11 and 9 for the intact structure and its core, respectively. Our reconstructions indicate that both complexes are dodecameric (Fig. 2), leading to calculated M r of 649,000 (CaM KII␣) and 221,000 (tCaM KII␣) from their residue composition. The Stokes radii of tCaM KII␣ and CaM KII␣ are 49 and 85 Å, respectively, showing that the intact subunit forms a complex that is significantly larger than that formed by the truncated subunit. The size difference between tCaM KII␣ (100 Å) and CaM KII␣ (200 Å) is consistent with the different values of the Stokes radii for the two structures (Fig. 2). DISCUSSION This study and previous studies (29) show that the C-terminal residues 315-478 (Fig. 3) are responsible for the oligomer formation of CaM KII␣, and this study shows that the structure of tCaM KII␣ forms the core of CaM KII␣ (Figs. 2 and 5A) and that the N-terminal catalytic domain (residues 1-309, Fig. 3) is associated with the foot-like extensions of the core. The sequence identity between the catalytic and calmodulin domains of CaM KI and CaM KII␣ is 45%, indicating that this portion of the two structures is likely to be similar (45). Indeed, a com- FIG. 3. Diagrammatic representation of the domains of CaM KII␣. tCaM KII␣ lacks the N-terminal catalytic and calmodulin domains. It comprises the C-terminal association domain (residues 315-478), which forms the core complex (Fig. 2D) of CaM KII␣. The Nterminal residues (1-309) comprise the functional domain and reside in the foot-like extensions associated with the core (Fig. 2, B and C).
FIG. 4. The protein density distribution in slices of the reconstructions. The 18 slices (5.7 Å thick) are from the top half of the structure (the 2-fold symmetrically related bottom half is not shown) and show that the highest protein density and greatest connectivity between strands occur about halfway toward the middle of the core. The strands rotate clockwise 30°from the top to the middle of the core. parison of the low resolution filtered x-ray structure of CaM KI and the foot-like extension of CaM KII␣ shows that they have a similar morphology (Fig. 5B). Because it was noted previously that all eukaryotic protein kinases share a conserved catalytic core that has a characteristic bilobate architecture with the active site situated at the interface between the two lobes (45), we propose that the active site is located near the middle portion of the foot, which also accommodates the catalytic and calmodulin-binding domains (Fig. 5B). A three-dimensional structure of CaM KII␣ with calmodulin bound will better define the location of these components in the foot-like appendages.
The N-and C-terminal functional and association domains of the ␣ and ␤ isozymes exhibit Ͼ90% and 76% amino acid sequence identity, respectively (24,29). Consequently, we predict that the ␤ isozyme has a dodecameric core that is very similar to the gear-shaped core of tCaM KII␣ (Fig. 2D) and that its N-terminal functional domain is similar to the foot-like appendage of CaM KII␣ (Fig. 2, B and C). In support of this notion, the molecular masses of the ␣ and ␤ isozymes are similar (␣, 600 kDa and ␤, 663 kDa) (29), and their average images of the central hexagonshaped ring of the 6-fold views are indistinguishable (data not shown). The unique domain in the ␤ subunit (residues 314 -387) is positioned between the association and functional domains, which would appear to place it in the neck region of CaM KII␣ between its core and foot-like appendages. Such an arrangement may result in a different orientation for this functional component in the ␤ complex that may be significant in its targeting to distinct subcellular sites in neurons (5)(6)(7)(8).
Structure-Function Relationships-The location of the functional components (the catalytic, calmodulin-binding domains and the Thr-286) on extended tethers at each end of the core complex is an unusual macromolecular architecture. In contrast, most enzyme structures are organized so that their functional components are an integral part of the molecule. Yet this organization appears well suited for the function of the complex. The clustering of six functional domains appears to optimize the FIG. 5. Stereo views of the superimposed CaM KII␣ (wire frame) and tCaM KII␣ core (red) structures (A) and a comparison of the morphology of the foot-like extension of CaM KII␣ with the filtered x-ray structure of CaM KI (B). The semitransparent tCaM KII␣ fits nicely in the core of CaM KII␣ and shows that the core comprises its association domain (residues 315-478) and that the foot comprises the functional domains (residues 1-314) (Fig. 3). The bilobate CaM KI ribbon representation (49) is enclosed in its structure filtered to 25 Å. The similarity between the morphology of the two structures indicates that the crevice between the two lobes in the x-ray structure is near the middle of the foot and that the lobe with abundant helices resides toward the distal end of the foot. Based on the sequence homology between CaM KI and CaM KII␣, the ATP-and calmodulin-binding sites are near the middle of the foot. The foot-like extension of CaM KII␣ is oriented so that the bottom of the foot is toward the viewer and its neck is in the opposite direction. autophosphorylation reactions of the Thr-286 that require the juxtaposition of two of the foot processes so that the catalytic domain of one may phosphorylate the Thr-286 on the other. Even though the significant distance between the foot-like extensions does not permit the close encounter required for the autophosphorylation reaction, the binding of calmodulin to these components may optimize their orientation for the autophosphorylation. In this regard, Ca 2ϩ /calmodulin, which is required to bind to proximate subunits for autophosphorylation (46), is known to undergo a significant conformational change upon interaction with the binding domain of CaM KII (47). This interaction could also affect the conformation of the catalytic domains so that pairs of foot processes are in the proper orientation. Because there are three pairs of juxtapositional catalytic domains, we propose that the initial autophosphorylation event involves nearest neighbors followed by the autophosphorylation of the remaining three domains. The structure shows that the two sets of catalytic centers 200 Å removed presumably function independently of each other.
A considerable body of evidence implicates CaM KII␣ as an essential component of the long lasting synaptic enhancement denoted long term potentiation (reviewed in Ref. 13). It was proposed that the long term potentiation involves the phosphorylation of the ␣-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor by CaM KII␣ (48). The arrangement of six catalytic domains clustered in close proximity seems well suited for the kinase to phosphorylate the predicted five subunits of the AMPA receptor simultaneously or in rapid succession. The extended arrangement of the catalytic domains may serve to promote access to the sites of phosphorylation on the receptor. Because the N-methyl-D-aspartate and AMPA receptors are juxtapositioned on the dendritic spine (13), the arrangement of the two catalytic centers, one at each end of the CaM KII␣ structure, may be optimally arranged to phosphorylate the receptors in response to Ca 2ϩ influx. Furthermore, the interdigitation of the gear-shaped cores would permit a close packing arrangement of the CaM KII␣ complexes with the receptors that may be important in the postsynaptic signaling system.
As indicated previously, CaM KII is involved in numerous biological processes, and the present example of its putative role in the neuronal dendritic spine is just one example of how its unusual architecture appears to be well suited to meet its multifunctional requirements in biology.