Molecular Dynamics Characterization of the C2 Domain of Protein Kinase Cβ*

Protein kinase C (PKC) isozymes comprise a family of related enzymes that play a central role in many intracellular eukaryotic signaling events. Isozyme specificity is mediated by association of each PKC isozyme with specific anchoring proteins, termed RACKs. The C2 domain of βPKC contains at least part of the RACK-binding sites. Because the C2 domain contains also a RACK-like sequence (termed pseudo-RACK), it was proposed that this pseudo-RACK site mediates intramolecular interaction with one of the RACK-binding sites in the C2 domain itself, stabilizing the inactive conformation of βPKC. βPKC depends on calcium for its activation, and the C2 domain contains the calcium-binding sites. The x-ray structure of the C2 domain of βPKC shows that three Ca2+ions can be coordinated by two opposing loops at one end of the domain. Starting from this x-ray structure, we have performed molecular dynamics (MD) calculations on the C2 domain of βPKC bound to three Ca2+ ions, to two Ca2+ ions, and in the Ca2+-free state, in order to analyze the effect of calcium on the RACK-binding sites and the pseudo-RACK sites, as well as on the loops that constitute the binding site for the Ca2+ ions. The results show that calcium stabilizes the β-sandwich structure of the C2 domain and thus affects two of the three RACK-binding sites within the C2 domain. Also, the interactions between the third RACK-binding site and the pseudo-RACK site are not notably modified by the removal of Ca2+ ions. On that basis, we predict that the pseudo-RACK site within the C2 domain masks a RACK-binding site in another domain of βPKC, possibly the V5 domain. Finally, the MD modeling shows that two Ca2+ ions are able to interact with two molecules of O-phospho-l-serine. These data suggest that Ca2+ ions may be directly involved in PKC binding to phosphatidylserine, an acidic lipid located exclusively on the cytoplasmic face of membranes, that is required for PKC activation.

Protein kinase C (PKC) 1 is a family of protein kinases that undergo translocation from one intracellular compartment to another when activated by neurotransmitters, hormones, and growth factors. Most members of this family are activated by phosphatidylserine (PS), diacylglycerol (DG), and, to different extents, by other lipid second messengers and Ca 2ϩ ions (1,2). There are at least three subfamilies of PKCs, classified according to their homology and sensitivity to activators (2)(3)(4). ␤PKC belongs to the so-called classic PKC subfamily, or cPKC (␣, ␤I, ␤II, and ␥ kinases). The members of this class contain four conserved domains (C1, C2, C3, and C4) inter-spaced with isozyme-unique (variable or V) domains. The ␤I and ␤II PKC isozymes are splice products of the same gene and therefore differ only in their C-terminal V5 domain (5).
The C2 domain is found also in proteins other than PKC (4). Because many of these proteins bind lipids and particularly PS in a calcium-dependent manner, it was obvious to suggest that the calcium-and PS-binding sites reside within this domain (6). Sequence alignment studies (4) revealed that the C2 domains exhibit two types of topologies (type I and II), but all have a ␤-sandwich fold composed of four ␤-strands in each face of the structure (7). Based on this homology, the C2 domain was suggested to contain the calcium switch required for localization to the membrane (8). However, immunofluorescence studies did not agree with simple localization of the activated PKC isozymes to the plasma membrane. We found that activated PKC isozymes are each localized to unique intracellular sites and, therefore, suggested that this unique localization is mediated by their selective interactions with specific anchoring proteins, termed Receptors for Activated C-Kinase (RACKs, see Fig. 1) (9 -14). We subsequently demonstrated that the unique cellular functions of PKC isozymes are indeed dependent on the binding of each isozyme to its corresponding RACK, bringing the active PKC within a close proximity to particular subsets of substrates and away from others (3).
Structures of the C2 domain obtained by x-ray diffraction and NMR spectroscopy (18,19) demonstrate that these RACKbinding sequences in ␤PKC-C2 are located on three exposed ␤-strands in the domain (Fig. 2B). Peptides corresponding to these three ␤-strands specifically inhibit activation-induced translocation of the C2-containing cPKC isozymes and their functions in cells (16,20). As predicted, a peptide derived from a non-conserved region of the C2 domain (␤C2-3, amino acids 201-207, IPDPKSE; yellow in Fig. 2), which is not adjacent to the RACK-binding strands, had no effect on PKC binding to RACK, or ␤PKC translocation and function in cells (16). It therefore appears that the C2 domain of PKC has a dual role. Upon activation, the domain binds PS in a calcium-dependent manner, which results in membrane binding of the enzyme. In addition, this domain participates in specific protein-protein interactions with the corresponding RACK, bringing the activated isozyme to a close proximity with a subset of substrates and away from others, thus mediating functional specificity of this family of enzymes.
Using the same logic, we suggested that translocation activators should be agonists of PKC function, independent of the amount of second messengers that normally activate PKC (21). We predicted that such peptide agonists would bind the unstable transition state between inactive and activated PKC, causing exposure of the catalytic site and the RACK-binding site and thus enabling the anchoring of the enzyme to RACKs (Fig. 1). Indeed, a ␤PKC-derived peptide, termed pseudo-RACK1 peptide or ␤RACK because of its homology to RACK1 (amino acids 241-246 within the C2 domain, SVEIWD, green in Fig. 2), binds to ␤PKC, activates it in the absence of PS and calcium in vitro, and acts as a selective agonist of ␤PKC function in vivo (22). We proposed that the ␤RACK site in ␤PKC is an autoregulatory site (17,21). When ␤PKC is in an inactive conformation, the ␤RACK site interacts with the RACK1-binding site; activation of PKC exposes the RACK1-binding site, enabling the association of the enzyme with its anchoring RACK (22). A model for this interaction is shown in Fig. 1, and the relative position of the RACK1binding sites within C2 and the ␤RACK site are indicated in the primary structure and in the secondary structure ( Fig. 2A) as well as in the tertiary structure of the domain (Fig. 2B).
Because ␤PKC binds to RACK1 upon activation with calcium and PS, we studied the conformational changes in the C2 RACK-binding sites in the absence and presence of these factors using computer simulation techniques of molecular dynamics (MD). MD can provide an insight into the structure and dynamics of proteins. Although such simulations are limited by the number of conformational spaces that can be sampled, these techniques are quite effective in monitoring even subtle structural changes and variations in residueresidue interactions, also when the comparison of very similar systems is concerned (23)(24)(25)(26)(27); in particular, the effect of calcium ions on protein structure was successfully evaluated on peroxidases (28). Here, using MD simulations of the C2 domain of ␤PKC, we examined the effects of calcium binding on the RACK1-binding sites and ␤RACK site within the C2 domain. Activated ␤PKC has been reported to bind two (18) or three (19) Ca 2ϩ ions in the C2 domain. Therefore, MD calculations were performed on the C2 domain of ␤PKC with three Ca 2ϩ ions (hereafter referred to as ␤PKC-3Ca), with two Ca 2ϩ ions (␤PKC-2Ca hereafter), and with no Ca 2ϩ ions (␤PKC hereafter), using the available x-ray structure as the starting model (19). Because a direct correlation between calcium and PS binding has also been demonstrated, a simple model was built to examine the possibility that Ca 2ϩ ions bound to the domain are directly involved in PS binding. Three molecules of O-phospho-L-serine, which mimic the headgroup of PS, were positioned near the Ca 2ϩ ions in the MD structure of ␤PKC-3Ca. An MD simulation was also performed on this complex (␤PKC-3Ca⅐PS hereafter). Our MD studies suggest that, although conformational changes occur within the C2 domain due to calcium and PS binding, they are unlikely to mediate the disruption of interaction of FIG. 1. Model of PKC activation. Inactive PKC is depicted as a folded rod with the pseudo-substrate autoinhibitory sequence associated with the substrate site in the catalytic domain, as well as with the pseudo-RACK sequence associated with the RACK-binding site. In the presence of PKC activators (PS/DG/calcium), the rod unfolds and the RACKbinding and substrate-binding sites become exposed, resulting in binding of PKC to its RACK and to its substrate. The pseudo-RACK peptide is thought to bind the unstable transition state between the inactive and active forms, shifting the equilibrium between the two conformations toward the active (open) one. the ␤RACK site with a RACK-binding site within the C2 domain.

MATERIALS AND METHODS
All calculations were carried out using AMBER 5.0 (29), a program commonly used for molecular dynamics simulations of proteins (27, 30 -34). The standard AMBER 1994 force field (35) was used. The x-ray structure of the C2 domain of rat ␤PKC complexed with three Ca 2ϩ ions (2.7-Å resolution, PDB entry 1A25) was used to model the starting structures (19). Although the crystal structure contains two molecules of ␤PKC-C2 related by a dyad axis in the asymmetric unit (19), the modeling was conducted on a single molecule. This domain contains 132 residues (157-288 of the complete sequence), corresponding to 1088 heavy atoms. The water molecules found in the crystal structure were excluded, following the standard procedure for protein simulations in explicit water described in the AMBER manual (29). Hydrogen atoms, which are missing in the 1A25 PDB file, were added through the EDIT module of AMBER, resulting in 2173 atoms in the native structure. The Ca 2ϩ ions were added to the standard AMBER residue data base by the PREP module of AMBER: They were treated as divalent cations, with a van der Waals radius of 1.60 Å, and ⑀ (i.e. 6 -12 potential well depth) of 0.1 kcal mol Ϫ1 , based on previous MD calculations dealing with the effect of calcium ions on the structure of peroxidases (28). The starting positions of the Ca 2ϩ ions were set according to their coordinates in the x-ray structure and were labeled II, III, and IV, following Sutton and Sprang notation (19). For ␤PKC-2Ca, calcium II and III were chosen, on the basis of the model of Ca 2ϩ binding by C2 domains depicted by Shao et al. in Ref. 18: In that work, two calcium ions were found to be bound to the C2 domain through NMR spectroscopy. No direct bond was set between the calcium ions and any protein groups, and no distance constraints were introduced to prevent a bias in the modeling, i.e. the calcium ions are free to leave the protein. A shell of TIP3P (36) water molecules extending 10 Å in every direction from the protein surface was created using the SOL option of the EDIT module of AMBER, resulting in the introduction of about 2100 water molecules for all the model systems. Proper counterions were generated by the CION program of AMBER and positioned near free charged surface residues of the protein to achieve an overall charge of zero on each system, ensuring that electrostatic interactions were not broken. A 10-Å cut-off for the evaluation of the non-bonded interactions was used, resulting in the evaluation of 2.5-2.6 ϫ 10 6 pair interactions. Because proteins are systems where long-range electrostatics are expected to play an important role in determining molecular conformational energies and structures, the choice of the cut-off for non-bonded interactions is significantly important for the quality of the simulation. The value of 10 Å adopted here was shown (28,32,33,(37)(38)(39) to be a good compromise between the requirement for accurate treatment of long-range electrostatics and the requirement for a reasonable calculation time, of which the evaluation of non-bonded interactions is by far the determinant part.
The solvent molecules were equilibrated initially by energy minimization and subsequently by performing 15 ps of MD. After energy minimization of the whole system (protein ϩ water ϩ counterions), MD trajectories were calculated. Temperature was initially increased from 0 to 300 K (performing three MD runs of 3 ps each at 100, 200, and 300 K, respectively) and then maintained constant for the whole simulation time, coupling the protein to a thermal bath (40). The SHAKE algorithm (41)(42)(43) was applied on all bonds during all the MD runs. However, a time-step of 1.5 fs could be used only in the calculation of the trajectory for ␤PKC, whereas a time-step of 1.0 fs was used both for ␤PKC-3Ca and ␤PKC-2Ca, due to instability of the trajectory in the initial steps of the simulations. The pair list was updated every 20 steps, and coordinates and energy values were collected every 100 steps for further analysis. The simulations were performed for 1060 ps, and the final 1000 ps were taken for the analysis. This time frame was chosen because ␤PKC system reached stabilization, i.e. the structure reached an average constant r.m.s.d. 2 with respect to the starting structure, after the initial 60 ps (Fig. 3). ␤PKC-3Ca and ␤PKC-2Ca equilibrated faster, in about 10 ps (Fig. 3). The average structures were calculated by averaging the coordinates at the various steps of the trajectories. The average structures were then subjected to energy minimization and subsequently analyzed by the program PROCHECK (44) to confirm the stereochemical quality of the model structures.
In the ␤PKC-3Ca⅐PS model, three molecules of O-phospho-L-serine were built by the PREP module of AMBER and initially positioned in proximity of the Ca 2ϩ ions of ␤PKC-3Ca average structure, to evaluate 2 r.m.s.d. (root mean square deviation) is defined as (⌺ i ⌬r i 2 /i)1⁄ 2, where ⌬r i is the displacement of an atom in a structure with respect to a reference structure, and the sum is performed over the atoms of the backbone, or over the heavy atoms, i.e. all the atoms excluding hydrogen. the possibility of a direct calcium⅐PS interaction (45). Nevertheless, the initial calcium⅐PS distance was set larger than coordination distance, to prevent a bias in the modeling of such interaction. The internal coordinates and the parameters for O-phospho-L-serine were obtained from data for serine and PO 2 groups already present in AMBER libraries. The procedure for MD modeling of this complex was performed as described above, consisting of solvation, energy minimization, heating, and dynamics. The simulation time was 600 ps, with 1.0 fs as the time-step.

RESULTS
␤PKC-3Ca-The MD average structure, which is shown in The coordination pattern of Ca 2ϩ ions in the MD structure of ␤PKC-3Ca is shown in Fig. 7 and is in good agreement with the x-ray structure (19), validating our modeling procedure. That coordination is maintained during the whole ␤PKC-3Ca simulation. Because neither direct bonds nor distance constraints were set between Ca 2ϩ ions and any protein groups, we infer that these binding sites for calcium are highly stable. The coordination sphere is completed with water molecules. Site IV, which is hexacoordinate in the crystal structure (19), appears heptacoordinate in the present model, with the carbonyl oxygen of Asn 253 as the seventh ligand.
␤PKC-3Ca⅐PS-Phosphatidylserine (PS) is an acidic lipid located exclusively on the cytoplasmic face of biological membranes and is required for PKC activation (1,2). The exact mode of interaction of PS with the enzyme is unknown; however, it has been proposed that Ca 2ϩ ions bound to the protein may be involved in PS binding by providing a positively charged site for the negatively charged headgroup of PS (45). To mimic PS, we have chosen a simpler molecule bearing the same headgroup as PS, i.e. O-phospho-L-serine, which was also used in the determination of the crystal structure (19). As described under "Materials and Methods," in this simulation three molecules of O-phospho-L-serine (PS1, PS2, and PS3) were initially positioned close to the Ca 2ϩ ions in the average structure of ␤PKC-3Ca. After 50 ps, the structure became stable for the remainder of 550 ps of simulation. The snapshots along this simulation (see Fig. 8) show that two molecules of O-phospho-L-serine (PS1 and PS2) are able to substitute the water molecules in the coordination sphere of calcium III and IV (see Fig. 7). In particular, PS2 binds calcium IV in the equatorial plane through an oxygen atom of its phosphate group and PS1 binds to both calcium III and calcium IV in an axial position, through the oxygen atoms of its carboxylate group. Conversely, the third molecule of O-phospho-L-serine (PS3), which binds calcium II for the initial 450 ps of the simulation, eventually moves away, suggesting that this interaction is not stable. Furthermore, the three PS-like molecules establish instantaneous, probably nonspecific, interactions with some of the residues belonging to loop 6 -7, i.e. Leu 249 , Thr 250 , Ser 251 , and Arg 252 . Though none of these interactions are stable along the entire trajectory, altogether they apprecia-bly reduce the mobility of loop 6 -7: the average value of the fluctuations for residues 247-253 is 0.64 Å, compared with 1.00 Å for the same segment in the ␤PKC-3Ca system, therefore, in the absence of PS. Although this simulation does not completely mimic the PKC interaction with the membrane bilayer, it further supports the previous findings that calcium may be directly involved in binding of PS (45,46).
␤PKC-2Ca-The average MD structure of ␤PKC with two calcium ions is shown in Fig. 4B. Removal of calcium IV from ␤PKC-3Ca increases the average r.m.s.d. relative to the crystal structure, with 0.98 Å of r.m.s.d. for the backbone and 1.75 Å  for the heavy atoms, respectively. The values per residue are shown in Fig. 5B. Comparison of Figs. 4A and 4B shows that the effect of removing this Ca 2ϩ ion is mainly observed on loop 6 -7, which is involved in calcium coordination, and on loop 3-4, probably due to the shortening of ␤-strand 4 (from 209 -211 to 210 -211). Additional changes in secondary structures were noted: ␤-strands 5 and 8 are shortened (from 222-230 to 225-230 and from 271-276 to 273-276, respectively), and the first helix (amino acids 233-238) is missing after removal of calcium IV. In contrast, the second helix (␣1) becomes longer (from 265-268 in ␤PKC-3Ca to 263-268 in ␤PKC-2Ca; see Fig.  6, top). The average fluctuation is 0.51 Ϯ 0.09 Å for the backbone and 0.58 Ϯ 0.27 Å for the heavy atoms. Fig. 6 reports the average heavy atom fluctuation per residue. That figure and analysis of Fig. 4 suggest that the slight increase in the mobility of ␤PKC-2Ca as compared with ␤PKC-3Ca is due to a larger flexibility of loops 6 -7 and 3-4. The removal of calcium IV results in an increased flexibility of loop 6 -7, because this calcium is coordinated exclusively by residues in loop 6 -7 (see Fig. 7). The removal of calcium also affects loop 3-4 indirectly, because it results in a shortening of ␤-strand 4.
␤PKC-The MD average structure of the calcium-free form is reported in Fig. 4C. This figure shows that, also in the absence of Ca 2ϩ ions, the global fold of the domain is maintained. However, conformations of certain regions are notably different from the calcium-bound structures.
The average r.m.s.d. of ␤PKC as compared with the crystal structure is higher than in the calcium-bound states; it is 1.65 Å for the backbone and 2.44 Å for the heavy atoms. The largest r.m.s.d. values are found for loops 3-4 and 6 -7 (see Fig. 5C), supporting the observation that these loops are the regions most affected by the progressive removal of calcium. The most striking effect of calcium removal is the disappearance of the short ␤-strand 4 ( Figs. 4 and 6, top). Other secondary structure elements that differ from ␤PKC-2Ca (see Fig. 6, top) include the interruption of ␤-strand 3 in the middle (residues 198 -199), shortening of ␤-strand 7 (from 254 -262 to 259 -262) and lengthening of ␤-strand 8 to the length found in the ␤PKC-3Ca state (271-276). An increase of mobility for loop 3-4 is also induced, as shown in Fig. 6. However, in the absence of calcium, the domain remains quite rigid: The average fluctuation is 0.57 Ϯ 0.08 Å for the backbone and 0.63 Ϯ 0.32 Å for the heavy atoms (these values are only slightly larger than the values found for ␤PKC-2Ca and ␤PKC-3Ca).

DISCUSSION
The present MD simulations allowed the analysis of the effects of Ca 2ϩ ions on the structure of the C2 domain of ␤PKC. We focused mainly on the regions of this domain that have a major role in previously reported biological functions (16 -18, 20, 22). These regions are indicated in Fig. 2 and Table I. The simulations were performed on the C2 domain of ␤PKC bound to three Ca 2ϩ ions (with and without the PS-like headgroups), to two Ca 2ϩ ions, or in the Ca 2ϩ -free state. Using this analysis, we evaluated (i) the effect of Ca 2ϩ on pseudo-RACK1 and RACK1-binding sites, (ii) the effect of Ca 2ϩ on the regions of the domain involved in Ca 2ϩ coordination, and (iii) the possible role of calcium in PS binding by the C2 domain.
Effect of Calcium on Pseudo-RACK1 and ␤C2-2 Sites-As described in the introduction, the pseudo-RACK1 sequence was predicted to be an autoregulatory site for ␤PKC (21). We suggested that, when ␤PKC is in an inactive conformation, pseudo-RACK1 site interacts with the RACK1-binding site within the enzyme (Fig. 1). In the active form, this intramolecular interaction is interrupted, rendering the RACK1-binding site available for the interaction with the RACK ((22) and Fig. 1). According to the structure of the C2 domain (see Fig. 2B), it is evident that, among the three identified RACK1-binding sites (16), only the ␤C2-2 (red) could interact with the pseudo-RACK1 site (green). The pseudo-RACK1 site is part of strand 6, and the spatial organization of the ␤-sheet to which it belongs places it next and anti-parallel to residues 194 -199 of strand 3. In this position, three pairs of hydrogen bonds are present between the amide proton and the carbonyl oxygen of residues Lys 199 -Ser 241 , Lys 197 -Glu 243 , and Tyr 195 -Trp 245 (Fig. 9).
Regardless of the presence or absence of calcium, the distances between the ␤-strands 194 -199 and 241-246, the two central strands of a four-stranded ␤-sheet, remain constant in the three modeled structures. The only exception to this general behavior is the distance between Lys 197 and Glu 243 . In ␤PKC-3Ca and in ␤PKC-2Ca this distance is constant, whereas in the absence of calcium (␤PKC) this distance fluctuates. These fluctuations, which are in the range of 1-1.5 Å, could have some functional relevance, because Glu 243 constitutes the crucial difference between the pseudo-RACK1 sequence in ␤PKC (SVEIWD) and the corresponding site in RACK1 (SIKIWD). The presence or absence of Ca 2ϩ ions does not significantly affect the hydrogen bonds between the ␤C2-2 and the pseudo-RACK1 segments: All the hydrogen bonds are maintained during the MD simulations on the various trajectories, thus stabilizing the interaction between these strands in the ␤-sandwich.
Interesting information is obtained by monitoring the behavior of the side chains (Fig. 10). The fluctuations of several distances during the simulations suggest that, in the presence of three Ca 2ϩ ions, there is a stable conformation in which no interaction is present between the side chains of the ␤-strands 3 and 6. In particular, the amine group of Lys 199 , which in ␤PKC-2Ca and in ␤PKC interacts with the carboxylic group of Glu 243 and, weakly, with the hydroxyl group of Ser 241 , does not interact with these groups in ␤PKC-3Ca. Furthermore, an Hbond is formed between Glu 243 and the amide group of Gln 280 , which belongs to helix 3; the distance between Glu 243 -Gln 280 is quite constant in the presence of three Ca 2ϩ ions, whereas it undergoes large fluctuations in ␤PKC-2Ca and in ␤PKC. These findings may indicate that the third Ca 2ϩ ion stabilizes, at least partially, helix 3 in a position close to pseudo-RACK1, whereas two Ca 2ϩ ions are not sufficient to do that. Fig. 10 shows a comparison between the average structures in which the changes involving residues Lys 199 , Glu 243 , and Gln 280 , induced by calcium binding, are shown.
In conclusion, the interaction between pseudo-RACK1 and ␤C2-2 sites is a very stable strand⅐strand interaction in the ␤-sandwich, which is not markedly altered by the removal of Ca 2ϩ ions. If the activation of PKC requires the dissociation of the intramolecular interaction between these two sites, it seems highly unlikely that this dissociation involves the disruption of the ␤-sandwich structure, and calcium binding is unlikely to result in such a dramatic conformational change. Instead, the intramolecular interaction between the pseudo-RACK1 site and a RACK-binding site may involve side-chain to side-chain interactions; removal of one Ca 2ϩ ion induces an interaction between Glu 243 (part of pseudo-RACK1) and Lys 199 (immediately after ␤C2-2). Therefore, it is possible that calcium binding to the third site results in disruption of intramolecular interactions between the RACK-binding site and the pseudo-RACK1 site. In addition, the surface exposed to the solvent is unaltered (Table II), suggesting that calcium binding does not change the surface characteristics of this site, and the potential interaction with RACK1. Together, these data suggest that the interaction between ␤C2-2 and pseudo-RACK1 do not constitute a major intramolecular interaction site that is disrupted in PKC activation by calcium. Instead, we favor the possibility that the intramolecular interaction involves the pseudo-RACK1 site with a RACK-binding site outside the C2 domain. This possibility is further supported by the finding of a RACK1binding site within the V5 region of ␤PKC (47). Examination of this possibility awaits the availability of structural information on the intact enzyme. Effect of Calcium on the Other RACK1-binding Sites-The ␤C2-1 segment, which contains a part of the RACK1-binding site (16) is a very basic region, on which the removal of Ca 2ϩ ions has the largest effect. As noted above, the short ␤-strand 4, which is formed by residues 209 -211 in ␤PKC-3Ca, is shortened (210 -211) in ␤PKC-2Ca and absent in ␤PKC (Fig. 6, top).
This loss in the secondary structure is associated to an appreciable increase in the fluctuations around this region (Fig. 6). Thus, the stabilization of a certain conformation of the ␤C2-1 segment is clearly a calcium-induced effect and may be related to the activation process. The presence or the absence of calcium does not affect the extent of the solvent-exposed surface (Table II) but has significant effect on its order as a consequence of reduced mobility. This increased rigidity may be presumably necessary for the recognition by the RACK.
Similar to ␤C2-1, the removal of Ca 2ϩ ions determines a decrease in the secondary structure elements of the ␤C2-4 segment (Fig. 6, top), although the surface exposed to the solvent is not significantly modified (Table II). Removal of just a single Ca 2ϩ ion produces the shortening of ␤-strand 5, where the first three residues (222-224) are in a random coil conformation both in the presence of two Ca 2ϩ ions (␤PKC-2Ca) and in the calcium-free form (␤PKC). Nevertheless, the fluctuations in these regions are limited in all three cases (Fig. 5). From these structural changes, we can infer that the ␤C2-4 segment requires the binding of three Ca 2ϩ ions to stabilize the proper conformation for binding to RACK.
Effect of Calcium on Loops 2-3 and 6 -7-Loops 2-3 and 6 -7 are involved in calcium binding in the C2 domain of ␤PKC (19,48). As previously predicted (48), the main effect of removal of Ca 2ϩ ions is to increase the distance between these loops. The removal of one Ca 2ϩ ion does not significantly modify this distance, whereas in the calcium-free state (␤PKC) the electrostatic repulsion among the acidic residues, deputed to bind calcium, makes loop 6 -7 move away from loop 2-3. Comparison of ␤PKC-2Ca and ␤PKC structures (Fig. 4, B and C) shows that, whereas loop 2-3 does not change significantly its position, the position of loop 6 -7 is notably different. This is con- FIG. 9. Interaction between ␤C2-2 and ␤RACK sites in ␤PKC-C2. Residues, which are involved in strand⅐strand interaction, are labeled, and backbone hydrogen (dark gray) and oxygen (light gray) atoms, which form hydrogen bonds, are evidenced as spheres.   with one of its carboxylic oxygens. These differences between ␤PKC-3Ca and ␤PKC structures are shown in Fig. 11. On the basis of NMR studies (18), it was found that ␤PKC-C2 binds two calcium ions (corresponding to sites II and III in Fig.  7), whereas a third ion may bind with a lower affinity. However, the crystal structure of the same domain (19) shows three bound calcium ions per domain. In that crystal structure, one donor atom to the third calcium ion (site IV in Fig. 7) is provided by a residue from the second C2 molecule in the asymmetric unit of the crystal; in solution, this donor atom is replaced by a well-ordered, highly rigid water molecule (the axial water in Fig. 7). Furthermore, Asn 253 remains at a coordination distance for the entire trajectory and completes heptacoordination of calcium IV, together with another water molecule in the equatorial plane (see Fig. 7). Therefore, binding site IV is already well-organized to bind a calcium ion in the crystal structure, which we used as the starting conformation in ␤PKC-3Ca simulation, and it remains very stable as long as the coordination sphere is complete, as it happens in our simulation: The energetics of this system is such that repulsion of calcium IV does not occur, at least on the MD trajectory timescale, i.e. about 1 ns.
How Is Calcium Involved in PS Binding?-It is well known that calcium increases the affinity of cPKC for negatively charged lipids (49,50). Newton and collaborators originally suggested that the positively charged surface of sheet B (␤strands 3, 4, 6, and 7), determined by a cluster of lysine residues at positions 197, 199, 209, 211, and 213, may provide the binding site for PS. The decreased distance between loops 2-3 and 6 -7 due to calcium binding could orient this basic face to interact with the lipids' headgroups (51). This hypothesis is in agreement with the x-ray structure by Sutton and Sprang (19) that was determined using crystals grown in the presence of O-phospho-L-serine as a headgroup analog of PS. Weak electron density was found at the surface of sheet B, such that the phosphate group was in contact with lysines 197, 199, and 211, even though the absence of strong density for the putative seryl group and specific protein contacts with it suggested that the interaction might be nonspecific (19).
Our results show that the accessibility of this basic cluster is not reduced in the absence of calcium as compared with the calcium-bound states (see Table II). Therefore, it does not seem likely that the calcium-induced increased affinity of PKC for negatively charged lipids is due to a larger exposure of the basic cluster on the surface of sheet B. Indeed, subsequent experimental data from Newton's group show that substitution of four lysine residues in the cluster with neutral residues does not inhibit association of ␤PKC with PS-enriched vesicles (52). Instead, our data suggest that this sequence provides part of the docking site for RACK1 (16), and the stabilization of the ␤-sandwich structure induced by calcium binding may be required for this protein⅐protein interaction (52).
As an alternative explanation, it has also been suggested that the Ca 2ϩ ions bound to PKC maintain some of their coordination open to interact with the negatively charged headgroup of phospholipids (45). We have examined this possibility by simulating the interaction of three O-phospho-L-serine molecules with the Ca 2ϩ ions in the structure of ␤PKC-3Ca. This MD simulation shows that only two molecules of O-phospho-Lserine are able to coordinate the three Ca 2ϩ ions through the oxygen atoms of the phosphate or of the carboxylate group, by substituting the water molecules in the coordination sphere of the calcium ions ( Figs. 7 and 8). These data support the possibility that Ca 2ϩ ions are indeed directly involved in PS binding by acting as a bridge.
The Ca 2ϩ bridge model is supported by an x-ray structure of the ␣PKC-C2⅐(Ca 2ϩ ) 2 ⅐PS complex by Verdaguer et al. (53), showing that one molecule of 1,2-dicaproyl-sn-phosphatidyl-Lserine is specifically coordinated to a Ca 2ϩ ion and other residues in the Ca 2ϩ -binding loops. In that work, a membranebinding mechanism of the ␣PKC-C2 domain is suggested in which two calcium ions play different roles in membrane binding. Recent studies of the C2 domains of ␣PKC (54) and cytosolic phospholipase A 2 (55), both of which bind two Ca 2ϩ ions, also showed that two Ca 2ϩ ions play distinct roles, with one primarily involved in inducing conformational changes and the other in Ca 2ϩ bridging. On this basis, we can infer similar differential roles for Ca 2ϩ ions in the membrane targeting of ␤PKC-C2, because ␤PKC-3Ca⅐PS simulation data show that only calcium III and IV are coordinated in a stable way by O-phospho-L-serine, whereas calcium II is not. On the other hand, comparison of data from ␤PKC-3Ca, ␤PKC-2Ca, and ␤PKC simulations (see above) show that all three Ca 2ϩ ions are involved in inducing conformational changes in the domain, especially in loop 6 -7. Therefore, we suggest that calcium III and IV play a dual role in the membrane targeting of ␤PKC-C2, providing a bridge between the C2 domain and phospholipids as well as inducing conformational changes. Conversely, calcium II is not involved in bridging and its role appears limited to the stabilization of the domain structure.
In addition, in the course of ␤PKC-3Ca⅐PS simulation, Ophospho-L-serine molecules appear to interact also with protein residues of loop 6 -7, reducing the mobility of that loop. This suggests that phospholipid binding may determine a further increase in the order of this region of the ␤PKC-C2, thus establishing a positive cooperation with Ca 2ϩ binding to achieve a certain conformation of the domain. This possibility is supported by a recent mutation analysis of the C2 domain of ␣PKC (54). In particular, in agreement with our findings, interactions along loop 6 -7 indicate that at least two residues (Arg 249 and Arg 252 ) participate in electrostatic interaction with anionic lipids and two others (Trp 245 and Trp 247 ) participate in penetration into the membrane. Also in the above-mentioned x-ray structure of ␣PKC complex with 1,2-dicaproyl-sn-phosphatidyl-L-serine (53), this PS-mimicking molecule is found to bind to the C2 domain of ␣PKC both via the calcium coordination and by direct interaction with Trp 247 and Arg 249 . On the other hand, in the same work (53) the authors emphasize that phospholipid analogs, such us O-phospho-L-serine, used to analyze the binding of phospholipids to ␤PKC (19), lack the possibility of some of the interactions seen in the structure of ␣PKC-C2⅐(Ca 2ϩ ) 2 ⅐PS complex. Therefore, the unstable interactions between ␤PKC-C2 and O-phospho-Lserine we identified in the simulation may in fact mimic the direct association of the enzyme with the lipids, and the instability could be due to the inadequacy of O-phospho-L-serine in mimicking PS. The docked model of the ␣PKC-C2⅐(Ca 2ϩ ) 2 ⅐PS complex onto a model membrane proposed by Verdaguer et al. (53) suggested to us that the molecule of PS3 in ␤PKC-3Ca⅐PS simulation (see Fig. 8) could be actually replaced by a longer hydrophobic tail of PS1. This model is, of course, mainly speculative.
Finally, Nalefski and collaborators (56) compared the equilibrium and kinetic parameters of C2 domains binding to calcium and lipids. This study demonstrates that there are at least two steps in the docking of the C2 domain to membranes: A rapid calcium binding followed by a slower membrane binding. These authors further demonstrated that, although the C2 domains of various proteins share sequence homology and similar architecture, they exhibit unique coordination of calcium, resulting in different kinetics of membrane docking (56). Our previous work on the role of RACKs in docking of activated PKC adds a third step in this process of docking, providing further mechanism to ensure specific subcellular localization of the activated enzyme and possibly increased stability for this docking.
Final Considerations-The present MD simulations indicated that the progressive reduction of the number of Ca 2ϩ ions bound to the C2 domain of ␤PKC causes a decrease in the number and the length of secondary structure elements, as well as an increase in the average fluctuations over the time of simulation. On this basis, we infer that calcium binding determines a stabilization of the ␤-sandwich structure of the C2 domain. In particular, we observed that this stabilization affects two regions of the C2 domain, which are involved in binding of the activated form of PKC to its receptor, RACK1. These regions correspond to ␤C2-1 and ␤C2-4. This observation suggests that calcium cooperates in PKC activation by driving the RACK1-binding sites within C2 toward the most favorable conformation for the interaction with RACK1.
In addition to the effect of calcium on the structure of ␤C2, calcium plays an electrostatic role in PKC activation: The binding of two or three Ca 2ϩ ions is expected to produce a drastic change in the electrostatic potential field around their binding sites, neutralizing the negative charge of the acidic cluster of aspartates on C2. This change in electrostatic field has been suggested to represent a molecular switch (50, 57) that contributes to shift the enzyme to its activated form. We propose here two possible but not incompatible mechanisms for the working of this switch. In the former mechanism, Ca 2ϩ ions are directly involved in PKC binding to PS: If this occurs, the same region of the C2 domain would bind both calcium and PS. This region, i.e. loops 2-3 and 6 -7, would be able to interact with PS only after the reversal of its electrostatic properties induced by binding of Ca 2ϩ ions. In the latter mechanism, the changes induced by calcium binding affect the domain⅐domain interactions between the C2 domain and other domains in PKC, which may contribute to maintaining PKC in the inactive state (58).
As mentioned earlier, one possible domain participating in this intramolecular interaction is the V5 domain. It is located in the last 50 amino acids of PKC and constitutes the only difference between ␤I and ␤IIPKC isozymes (3). Our recent study showed that at least part of the RACK1-binding site in ␤IIPKC is located within the V5 domain of the enzyme (47). The ␤IIPKC V5 domain binds directly to RACK1 with an affinity similar to that of the isolated C2 domain enzyme (47). In addition, peptides derived from the ␤IIV5 domain inhibit ␤IIPKC binding to RACK1 in vitro and inhibit ␤IIPKC translocation and function in cells (47). The presence of relatively limited interactions between the pseudo-RACK1 sequence with RACK1-binding sites within the C2 domain in our modeled structures suggests that pseudo-RACK1 site interacts with RACK-binding sites on other PKC domains. Because the V5 domain directly binds to RACK1, it may also interact with the pseudo-RACK1 sequence in the C2 domain, to mask it in the inactive state. Such V5⅐C2 interactions have been previously suggested (59,60). The first reports indicating interaction between the V5 and C2 domains were provided by the work of Newton et al., showing that calcium binding to the C2 domain of ␤PKC isozymes is affected by their V5 domains (reviewed in Ref. 59). In addition, the phosphorylation of Ser 660 in the V5 region increases the affinity of the enzyme to both phospholipids and calcium, presumably due to allosteric effects of the V5 domain on the C2 domain (60). The possibility of inducible intra-domains interactions in PKC, such as these suggested here between the C2 and V5, could be directly addressed when the complete structure of PKC becomes available.
Finally, it is still not known what are the steps that lead inactive cytosolic PKC to anchor to a particular site within the cells. There could be at least three steps, including calcium binding, lipid binding, and RACK binding. The final step leads to maximal stabilization of the activated isozymes, which occurs during protein⅐protein binding of the PKC via the C2 domain and the V5 domain to the pre-anchored RACK1. Future studies including co-crystallographic kinetic analysis with purified components will help elucidate the question.