Structural Determinants for Ca2+ and Phosphatidylinositol 4,5-Bisphosphate Binding by the C2A Domain of Rabphilin-3A*

Rabphilin-3A is a neuronal C2 domain tandem containing protein involved in vesicle trafficking. Both its C2 domains (C2A and C2B) are able to bind phosphatidylinositol 4,5-bisphosphate, a key player in the neurotransmitter release process. The rabphilin-3A C2A domain has previously been shown to bind inositol-1,4,5-trisphosphate (IP3; phosphatidylinositol 4,5-bisphosphate headgroup) in a Ca2+-dependent manner with a relatively high affinity (50 μm) in the presence of saturating concentrations of Ca2+. Moreover, IP3 and Ca2+ binding to the C2A domain mutually enhance each other. Here we present the Ca2+-bound solution structure of the C2A domain. Structural comparison with the previously published Ca2+-free crystal structure revealed that Ca2+ binding induces a conformational change of Ca2+ binding loop 3 (CBL3). Our IP3 binding studies as well as our IP3-C2A docking model show the active involvement of CBL3 in IP3 binding, suggesting that the conformational change on CBL3 upon Ca2+ binding enables the interaction with IP3 and vice versa, in line with a target-activated messenger affinity mechanism. Our data provide detailed structural insight into the functional properties of the rabphilin-3A C2A domain and reveal for the first time the structural determinants of a target-activated messenger affinity mechanism.

Rabphilin-3A is a neuronal C2 domain tandem containing protein involved in vesicle trafficking. Both its C2 domains (C2A and C2B) are able to bind phosphatidylinositol 4,5-bisphosphate, a key player in the neurotransmitter release process. The rabphilin-3A C2A domain has previously been shown to bind inositol-1,4,5-trisphosphate (IP3; phosphatidylinositol 4,5bisphosphate headgroup) in a Ca 2؉ -dependent manner with a relatively high affinity (50 M) in the presence of saturating concentrations of Ca 2؉ . Moreover, IP3 and Ca 2؉ binding to the C2A domain mutually enhance each other. Here we present the Ca 2؉ -bound solution structure of the C2A domain. Structural comparison with the previously published Ca 2؉ -free crystal structure revealed that Ca 2؉ binding induces a conformational change of Ca 2؉ binding loop 3 (CBL3). Our IP3 binding studies as well as our IP3-C2A docking model show the active involvement of CBL3 in IP3 binding, suggesting that the conformational change on CBL3 upon Ca 2؉ binding enables the interaction with IP3 and vice versa, in line with a target-activated messenger affinity mechanism. Our data provide detailed structural insight into the functional properties of the rabphilin-3A C2A domain and reveal for the first time the structural determinants of a target-activated messenger affinity mechanism.
C2 domains are ubiquitous protein modules initially identified in protein kinase C (1). They all consist of ϳ130 residues (2) and share a common fold composed of two four-stranded ␤-sheets arranged in a compact ␤-sandwich and surrounded by variable loops and helices. C2 domain-containing proteins such as phospholipase C, cytosolic phospholipase A2, or protein kinase C are generally involved in signal transduction or membrane trafficking. The main function of the C2 domains in these proteins appears to be the targeting to biological membranes via their Ca 2ϩ -dependent phospholipid binding properties (3).
Primary lipid targets are phosphatidylcholine (for cytosolic phospholipase A2 (4)) and phosphatidylserine (for phospholipase C␦ (5), synaptotagmins (6), and protein kinase C␣ (7)), the most abundant phospholipids in the inner membrane leaflet. Although not known to be classical phosphoinositide binding modules, some C2 domains have been shown to interact rather specifically with phosphoinositides (phospholipase C, protein kinase C␣), especially phosphatidylinositol 4,5bisphosphate (PIP2) 3 (8). PIP2 represents less than 1% of the lipids in the plasma membrane and forms microdomains to regulate several key signaling events (9,10). The interaction between the C2 domain of protein kinase C␣ and PIP2 is essential for the proper subcellular localization of protein kinase C␣ (11,12). Several C2 domains of proteins involved in synaptic vesicle fusion have been described to bind PIP2, e.g. C2 domains of some synaptotagmins (6), CAPS-1 (13), and rabphilin-3A (14). The binding of the C2B domain of synaptotagmin 1 to PIP2 prelocalizes the protein to PIP2-enriched membranes (15). Recently a role for the CAPS-1 PIP2 interaction has been shown in the prefusion step of dense core vesicle fusion (13).
PIP2 is a key player in the different stages of the neurotransmitter release process. It is required for clathrin-mediated endocytosis (16) or for specific stages of exocytosis and synaptic vesicle trafficking (10). PIP2 recruits proteins to specific locations in the plasma membrane to modulate spatially localized events involved in membrane trafficking (17).
Rabphilin-3A was originally identified as a Rab3 effector in neuronal cells (18). It was found that the C-terminal C2 domain tandem (C2A and C2B) of this protein mediates interactions with ␤-adducin (19), calcium/calmodulin-dependent serine protein kinase (20), and annexin A4 (21). For the C2B domain a specific interaction with the tSNARE SNAP25 (25-kDa synaptosomal-associated protein) was identified (22). Recent results indicate that rabphilin-3A increases the number of releasable vesicles at the plasma membrane (22,23) and enables the repriming of these vesicles after extensive evoked neurotransmitter release (23). The C-terminal C2 domains of rabphilin-3A are involved in the Ca 2ϩ -dependent membrane targeting of rabphilin-3A (24), their primary target being the membrane phosphatidylserine. In addition, they have been described to bind PIP2 in vitro (14), but so far no in vivo study has addressed the role of this interaction in rabphilin-3A function.
We have recently shown that rabphilin-3A C2 domains present well defined inositol 1,4,5-trisphosphate (IP3, i.e. PIP2 headgroup) binding sites in solution (25). However, the C2 tandem exhibits asymmetric behavior toward Ca 2ϩ and PIP2 headgroup binding. The C2B domain binds Ca 2ϩ with an affinity in the M range and binds IP3 in a Ca 2ϩ -independent manner with an affinity in the millimolar range (0.43 mM). Oppositely, the intrinsic Ca 2ϩ binding affinity of the C2A domain is low (1.1 mM), and it binds IP3 in a Ca 2ϩ -dependent manner with a relatively high affinity (50 M) in presence of saturating concentration of Ca 2ϩ . IP3 and Ca 2ϩ binding to the C2A domain mutually enhance each other, in line with a target-activated messenger affinity mechanism where both the messenger and the target have to be simultaneously present to allow the protein to interact with them at physiological concentrations (26). The asymmetry of the rabphilin-3A C2 domains toward PIP2 headgroup binding may have functional implications; the C2A domain might be involved in the targeting of synaptic vesiclebound rabphilin-3A to PIP2-enriched subdomains upon Ca 2ϩ signaling. Subsequently, the C2B domain, once bound to the plasma membrane, might interact with the phosphoinositide during the docking step, potentially modulating its binding to SNAP25.
The functional plasticity found across the C2 domains is essential to convey specific functions to their harboring proteins. A more detailed structural characterization of specific C2 domain features should provide deeper insight into their functional properties. Within this line, the C2A domain crystal structure has been solved only in its Ca 2ϩ -free state (27). Thus, the Ca 2ϩ binding properties are not clarified yet. This domain binds Ca 2ϩ cooperatively with a Hill coefficient of 2.24, suggesting that it binds two to three Ca 2ϩ ions (25). But the exact binding mode remains unknown. Moreover, the Ca 2ϩ binding region (CBR) of this domain exhibits several unusual features found only in a few other C2 domains (Fig. 1). In the Ca 2ϩ binding loop 3 (CBL3), glutamic acid residues are found in position 482 (instead of a conserved Asp) and position 475 (instead of a conserved aromatic residue). The role and function of this non-canonical Ca 2ϩ binding loop have been so far only poorly investigated. Finally, the structural features mediating the interrelation between Ca 2ϩ and PIP2 headgroup binding remain mostly unknown. To address these structure-function aspects, we solved the solution structure of the rabphilin-3A C2A domain in its Ca 2ϩ -bound form and investigated the structural determinants of its Ca 2ϩ and IP3 binding properties.
Structure Calculations-Backbone and dihedral restraints were derived from the program TALOS (34) using as input the 1 H, 13 C, and 15 N backbone chemical shifts. Only TALOS predictions with 10 matches agreeing were used as and dihedral restraints. To each of them was attributed an average value with a minimal S.D. range of 10°. Dihedral restraints, 13 C, 1 H/ 15 N HSQC-NOESY spectra and 1 H, 13 C, 15 N chemical shifts were used as input for the ATNOS/CANDID package (35,36) to generate a first ensemble of structures. This early ensemble was used to manually finalize the assignment of the 13 C, 1 H/ 15 N HSQC-NOESY spectra and led to a final set of about 1700 distance restraints. Dihedral angles and distance restraints were then used in the MD protocol of CYANA (37) to generate an ensemble of 100 structures of the rabphilin-3A C2A domain. Hydrogen-bond restraints were also added during this step, based on weak observed temperature coefficients (superior to Ϫ5⅐10 Ϫ3 ppm⅐K Ϫ1 ) (38) correlating with the secondary structure prediction.
The 50 structures with the lowest target function were selected for an RDC refinement protocol performed with Xplor-NIH (39) and the protein-2.0 force field. Ca 2ϩ ions were added subsequently based on Ca 2ϩ titration experiments and comparison with the canonical Ca 2ϩ binding site of C2 domains. For each Ca 2ϩ , six distance restraints were defined between the protein and the ion (Ca 2ϩ -oxygen distances were restrained to 1.8 -2.8 Å). The compatibility of the Ca 2ϩ binding restraints with the force field parameters and the experimental restraints (dihedral angles, distance restraints, and RDCs) was carefully checked.
The 20 structures with the lowest energy were selected as the final ensemble of structures describing the rabphilin-3A C2A domain in solution. Experimental restraints and structural statistics are summarized in Table 1. The lowest energy structure from the final ensemble was considered as the most representative of the ensemble and was used for preparing figures and structure comparisons.
Fitting Procedure-To evaluate titration experiments of the labeled C2A domain with Ca 2ϩ and IP3, the chemical shift deviations of significantly affected residues were plotted against the ligand concentration. A simultaneous fit was performed using Igor Pro (Wavemetrics Inc.) with the following equations. Equation 1 is a single binding site model (40), where [Pt], [Lt], K D and B max are the total protein concentration, ligand concentration, dissociation constant, and the chemical shift deviation at saturation, respectively.
For the cooperative binding model, the total ligand concentration was used in the Hill equation to calculate the free ligand concentration, where [Pt], [Lt], K Dapp , n, and B max are the total protein concentration, ligand concentration, global dissociation constant, the Hill coefficient, and the chemical shift deviation at saturation, respectively.
Docking Models for the IP3-C2A Domain Complex-Models of the C2A-PIP2 complex were calculated using the docking program HADDOCK (high ambiguity-driven protein docking) (41) in combination with CNS (42) as described in our previous work (25). The five structures from the final ensemble with the lowest energy were used as starting coordinates for the C2A domain. Only solvent-accessible residues undergoing significant chemical shift perturbation (⌬␦ larger than 0.1 ppm) upon IP3 titration (performed at pH 5) were defined as active residues. More details about the docking protocol including a list of active residues and a full description of the docking consecutive stages are summarized in Table 2.
The 50 structures obtained after water refinement were analyzed and ranked according to their HADDOCK score ( Table  2). The three best models were selected as the final representative ensemble of the complex structure.

Structural Statistics and Overall
Structure of the Ca 2ϩ -bound C2A Domain of Rabphilin-3A-The solution structure of the Ca 2ϩ -bound C2A domain of rabphilin-3A was solved with the aim of investigating its IP3 binding properties including the way they are modulated by Ca 2ϩ . Because amide resonances of residues from CBL3 were broadened below detection level at pH 7, all spectra were acquired at pH 5. The N-terminal portion of the C2A domain fragment (residues 371-383) was not included in the structure calculation as it has been observed to be highly dynamic. Indeed, only few sequential NOE effects were observed for this segment (as also revealed by the very narrow line-widths of 1 H, 15 N HSQC cross-peaks for these residues).
The 20 final structures of the Ca 2ϩ -bound C2A domain of rabphilin-3A at pH 5 exhibit an atomic root mean square distribution about the mean coordinate positions of 0.38 Ϯ 0.11 Å for the backbone atoms and 1.11 Ϯ 0.14 Å for all heavy atoms, excluding the residues from CBL1 and CBL3, which are slightly less well defined ( Fig. 2A). The structure ensemble fits the experimental restraints with no interproton distance and torsion angle violations greater than 0.2 Å and 10°, respectively. All of the backbone torsion angles of the non-glycine residues fall in the allowed regions of the Ramachandran plot ( Table 1).
The overall structure is composed by two four-stranded ␤-sheets arranged in a compact ␤-sandwich (corresponding to the type I topology described for the C2 domains) and sur-rounded by two helices, one two-turn ␣-helix between strands ␤ 6 and ␤ 7 and a one-turn 3 10 -helix between strands ␤ 7 and ␤ 8 (Fig. 2B). As observed in many other C2 domains and also in the Ca 2ϩ -free crystal structure of the C2A domain (27), a set of acidic residues is located at one extremity of the protein characterized by three loops (CBL1-3 in Fig. 2A). These residues define the CBR.
Ca 2ϩ Binding Mode-To determine the Ca 2ϩ binding mode of the C2A domain, we first performed a 1 H, 15 N HSQC-based Ca 2ϩ titration at pH 5. As expected, the same residues (located in the CBR) observed previously at pH 7 (25) were affected upon Ca 2ϩ binding (Fig. 3A), indicating that the Ca 2ϩ binding mode is topologically not modified at pH 5 as compared with pH 7. 1 H, 15 N HSQC cross-peaks affected upon Ca 2ϩ addition adopted a linear trajectory throughout the titration, allowing the fitting of their chemical shift perturbation against the Ca 2ϩ concentration. This also clearly indicates that the C2 domain switches from the unbound state to the fully Ca 2ϩ -loaded state without an intermediate state, which means that regardless of the number of Ca 2ϩ ions binding to the C2A domain, they bind concomitantly, i.e. with very similar affinities. Fitting the chemical shift perturbation of residues affected by Ca 2ϩ binding against the Ca 2ϩ concentration with Equation 2 to take in account cooperative binding (as suggested by the sigmoidal shape of the curve in Fig. 3B) gave a global dissociation constant of 3.76 Ϯ 0.05 mM with a Hill coefficient of 2.27 Ϯ 0.05 (Fig. 3B). The lower affinity measured at pH 5 as compared with pH 7 (1.10 Ϯ 0.01 mM) is expected considering the increased protonation state of the acidic residues in the CBR at lower pH. Nevertheless, the pH 5 data are in agreement with those measured at pH 7, i.e. the intrinsic Ca 2ϩ binding affinity is in the mM range. Noteworthy, the Hill coefficient of 2.27 Ϯ 0.05 measured at pH 5 is identical to the one measured at pH 7 (2.24 Ϯ 0.03). This value indicates that three Ca 2ϩ ions should bind in a cooperative way to the C2A domain of rabphilin-3A.
Based on the Ca 2ϩ titration and the geometry of the canonical Ca 2ϩ binding mode of C2 domains, Ca 2ϩ ions were added to the C2A domain structure during the RDC refinement. Only two Ca 2ϩ ions were positioned successfully into the canonical sites Ca1 and Ca2 (Fig. 3C). Attempting to add a third Ca 2ϩ to site Ca3 led to severe RDC and distance restraints violations, in line with mutagenesis results for the C2B domain of synaptotagmin I where the mutation of glycine 368 into a serine residue that is conserved in other C2 domains was a prerequisite to form a third low affinity Ca 2ϩ binding site (Fig. 3C) (43). Thus, a third Ca 2ϩ binding site, indicated by the Hill coefficient of 2.27, may exist in the CBR of the C2A domain. But our NMR data do not allow at present to locate this site.
The Ca 2ϩ binding region of the rabphilin-3A C2A domain is shown in Fig. 3D. Both Ca 2ϩ ions are hexacoordinated, involv-  ing four aspartate side chains (Asp-413, -419, -474 and -476), two carbonyl groups (Met-412 and Glu-475), and one glutamate side chain (Glu-482). The involvement of a glutamate side chain has not been observed so far in any structure of a C2 domain.
Structural Comparisons-To compare the Ca 2ϩ -free crystal structure with the Ca 2ϩ -free conformation in solution, we measured 1 D1 H ,15 N RDCs on a Ca 2ϩ -free sample (in the presence of 5 mM EGTA) aligned in Pf1 bacteriophage at pH 6. Under these conditions most of the CBL resonances are broadened below detection, whereas they are observable in the Ca 2ϩbound state. Nevertheless, RDC values observed in the core domain were used to determine the alignment tensor using singular value decomposition (44,45) as implemented in the software PALES (46). The RDCs fit to the Ca 2ϩ -free crystal structure with a correlation coefficient of 0.92 and a Q factor of 0.196 (supplemental Fig. 1). These values indicate that in solution, the Ca 2ϩ -free form of the C2A domain adopts a conformation very similar to the one observed in the crystal structure.
The Ca 2ϩ -free crystal structure and the Ca 2ϩ -bound solution structure superpose with an r.m.s.d. value of 0.82 Å for residues 384 -509, excluding CBL1 and CBL3 (Fig. 4A). Both structures are very similar, indicating that Ca 2ϩ binding does not modify the overall structure of the protein. Nevertheless, significant changes are observed within CBL1 and CBL3 (Fig.  4B). Both loops adopt very different conformations and positions upon Ca 2ϩ binding. Such dramatic changes are in agreement with the finding that in the Ca 2ϩ -free crystal structure the conformations and positions of the acidic side chains of the prospective Ca 2ϩ binding residues within these loops are not compatible with Ca 2ϩ binding.
In the Ca 2ϩ -bound structure, CBL1 adopts an open conformation by digressing from CBL3. More specifically, the conformation of Asp-413 changes in order to chelate Ca 2ϩ in position Ca1. Residues Asp-419 and -476 also chelate this ion, but these two residues already adopt a rather favorable conformation in the Ca 2ϩ -free crystal structure. As mentioned above, the Ca 2ϩinduced conformational change of residue Asp-413 seems also to go along with a large rearrangement of most of the CBL1 residues (Fig. 4B). In addition, CBL3 bends toward the tip of the domain, leaving the basic concave side more exposed to solvent. Moreover, Ca 2ϩ binding induces important conformational changes for residues Asp-474 and Glu-482 (Fig. 4B). In particular, two hydrogen bonds observed in the Ca 2ϩ -free crystal structure are not present anymore in the Ca 2ϩ -bound form, one between the side chain of residue Glu-482 and the amide group of residue Asp-413, the other one between the carboxylic group of Glu-475 (from the DED motif located in CBL3) and the guanidinium group of Arg-437 (from strand ␤ 3 ). This last observation is also confirmed by the fact that ␤ and ␥ protons of Glu-475 exhibit chemical shift degeneracy, revealing a high flexibility of the Glu-475 side chain. Noteworthy, conformations of residues throughout CBL3 also change significantly, including the basic residues His-480 and Asn-481 involved in PIP2 headgroup binding (see below) (Fig. 4C).
IP3 Binding to the C2A Domain-As shown above, Ca 2ϩ binding to the C2 domain of rabphilin-3A induces large conformational changes in CBL1 and CBL3. These changes might play an important role in the Ca 2ϩ -dependent IP3 binding to the C2A domain. Therefore, we probed the IP3 binding properties of this C2 domain by performing a 1 H, 15 N HSQC-based titration at pH 5 and in presence of 20 mM Ca 2ϩ . Large chemical shift perturbations were observed for a limited set of residues located on the concave side of the core domain, namely on the strands ␤ 3 and ␤ 4 , and on CBL3, forming a polybasic stretch (Fig. 5A). The saturation was not reached at 3 mM IP3. Nevertheless, a reasonable simultaneous fit with Equation 1 using chemical shift perturbations from 16 HN cross-peaks could be performed and gave a K D value of 1.30 Ϯ 0.03 mM (Fig. 5B). This much lower affinity compared with the value measured at pH 7 (55 M) (25) can be explained by the increased protonation state of the IP3 phosphate groups, loosening the interaction with the polybasic stretch of the C2A domain. Nevertheless, the chemical shift perturbation pattern perfectly overlaps with the one observed at pH 7 (data not shown), indicating that the IP3 binding mode at pH 5 does not differ from pH 7. However, most resonances of residues located in CBL3 are broadened out at pH 7, which makes the accurate evaluation of their involvement in IP3 binding difficult. At pH 5, these resonances are clearly observable, allowing the investigation of their involvement in IP3 binding. Indeed, as shown in Fig. 5C, most of CBL3 constitutes the third region of the IP3 binding site in addition to the strands ␤ 3 and ␤ 4 . Moreover, the large amplitude of chemical shift deviations of crosspeaks of residues in CBL3 (Fig. 5A), compared with those of both ␤ strands, suggests the active involvement of this Ca 2ϩ binding loop in IP3 binding. Resonances of the basic residues His-480 and Asn-481 are the most perturbed resonances of CBL3 upon IP3 titration. To probe the role of these basic residues in IP3 binding, two single point mutants have been designed: H480A and N481A. These two mutants were 15 N-labeled, and 1 H, 15 N HSQC-based IP3 titrations were performed at pH 7 in the presence of 20 mM Ca 2ϩ to determine affinity values at more physiological conditions. K D values were obtained from a simultaneous fitting procedure. The H480A and N481A mutants showed significantly higher K D values of 360 Ϯ 10 and 930 Ϯ 30 M, respectively (supplemental Fig. 2) as compared with the wild type protein (55 M). Meanwhile the intrinsic Ca 2ϩ binding affinities of these mutants were only slightly affected. The K D was 1.46 mM for H480A and 1.1 mM for N481A compared with a K D of 1.1 mM for the wild type C2A domain (data not shown). The strong decreases in IP3 affinity clearly indicate the active involvement of His-480 and Asn-481 in IP3 binding. Thus, residues from CBL3 define in addition to the polybasic stretch the PIP2 headgroup binding site of the C2A domain of rabphilin-3A.
IP3-C2A Domain Docking Models-The 1 H, 15 N HSQCbased IP3 titration performed at pH 5 and the data available for the H480A and N481A mutants provide a very well defined view of the IP3 binding mode to the C2A domain of rabphilin-3A. To derive a model of the IP3-C2A complex, we used these experimental data in a docking approach. The docking procedure was driven with HADDOCK 2.0, using the five lowest energy structures of the Ca 2ϩ -bound rabphilin-3A C2A domain as starting coordinates. PIP2 (1,2-diacetoyl-sn-3-[phosphoinositol-3,4,5-trisphosphate]) was docked to mimic a polar head from a lipid embedded in the plasma membrane. ambiguous interaction restraints were defined between the protein and the IP3 moiety of PIP2 to match the experimental conditions, where IP3 was used in the titrations. The docking protocol used led to an ensemble of 50 structures after water refinement. The 50 structures were ranked according to their HADDOCK score (Table 2). A single cluster of solutions was found within the final ensemble, and the three best models were selected as the final representative ensemble of the complex structure. Statistical data for the final ensemble are summarized in Table 2. Ligand position and orientation are well conserved (Fig. 6A), in line with the low r.m.s.d. value measured at the protein/ligand interface of 1.32 Ϯ 0.59 Å.
As shown in Fig. 6B, the PIP2 molecule binds to the concave side of the domain and interacts with basic residues from CBL3 and strands ␤ 3 and ␤ 4 . The residue His-425 interacts with phosphate 4, residues Lys-435 and Arg-437 interact with phosphate 3, and residues Lys-423, His-480, and Asn-481 interact with phosphate 1 of PIP2. In particular, the His-480 -Asn-481 pair acts as a clamp fixing phosphate 1. Because of these interactions, the PIP2 molecule lies flat on the concave surface of the ␤-sheet, with the diacetoylglycerol chain pointing toward the CBR (Fig. 6B).
Comparison of the C2A domain conformation in its IP3-free and IP3-bound forms suggests that PIP2 binding does not require major protein backbone conformational changes (backbone r.m.s.d. of 0.67 Å between the two forms). Even the CBL3, which was defined as a fully flexible segment during the calculations, exhibits only slightly different conformations, indicating that the Ca 2ϩ -bound form of the C2A domain adopts a conformation adjusted for the interaction with IP3. Overall, the Ca 2ϩ /IP3-bound C2A domain structure is much closer to its Ca 2ϩ -bound structure than to its Ca 2ϩ -free structure, revealing structural features underlying Ca 2ϩ dependence of IP3 binding.
Ca 2ϩ /IP3 Binding Cooperativity-Ca 2ϩ binding to the C2A domain triggers IP3 binding. Moreover, Ca 2ϩ loading of the C2A domain is greatly increased in the presence of IP3 (25). This phenomenon is known as target-activated messenger affinity mechanism, recently defined by Corbin et al. (26). It is reasonable to assume that IP3 binding will promote a similar conformational change on the CBR as the one induced by Ca 2ϩ and subsequently enhance Ca 2ϩ binding to the C2A domain.
The very low affinity of the Ca 2ϩ -free form for IP3 (0.8 mM) hinders NMR structural studies with conventional methods such as the measurement of RDCs or intermolecular NOEs. As an alternative approach to address this question we designed two mutants of the C2A domain with the aim to mimic, directly or indirectly, IP3 saturation.
The R437A mutant was designed based on our recent observation that Glu-475 from the DED motif on CBL3 interacts with Arg-437 from strand ␤ 4 via a hydrogen-bond network between the carboxylic group of Glu-475 and the guanidinium group of Arg-437 in the Ca 2ϩ -free crystal structure (27). In the Ca 2ϩ -bound solution structure, this interaction does not exist anymore (Fig. 4B). Moreover, the ␤ and ␥ protons of Glu-475 exhibit chemical shift degeneracy, revealing a high flexibility of the Glu-475 side chain. This implies that Ca 2ϩ binding breaks the interaction between Glu-475 and Arg-437, releasing the Arg-437 side chain, thus allowing a stronger interaction of Arg-437 with IP3, as shown by our docking model.
The N481D mutant was designed to mimic the presence of a phosphate group on the concave side of the core domain (namely the IP3 binding site). Our working hypothesis was that the Asp-481 side chain should interact with the basic residues of the concave side, shifting CBL3 into a conformation similar to the one observed in the IP3-C2A docking model.
Both mutants, R437A and N481D, were 15 N-labeled, and 1 H, 15 N HSQC-based Ca 2ϩ titrations were performed at pH 7. K D values were obtained from a simultaneous fitting procedure using Equation 2 (see "Experimental Procedures"). Both mutants showed significantly lower K D values of 477 Ϯ 7 and 305 Ϯ 4 M, respectively (supplemental Fig. 3), as compared with the wild type C2A domain (1.1 mM). Moreover, both mutants exhibited lower cooperativity than the wild type domain with Hill coefficients of 1.62 Ϯ 0.01 and 1.63 Ϯ 0.01 for Ϫ1753 Ϯ 10 a TA, torsion angle space; CA, cartesian space. b Defined as the weighted sum of different terms (1ϫE VDW ϩ 0.2E elec ϩ 0.1E AIR ϩ 1 ϫ E desolv ).
R437A and the N481D, respectively. Thus, both mutations facilitate Ca 2ϩ binding. The affinity of these mutants for IP3 was also measured in presence of saturating concentration of Ca 2ϩ . For both mutants the saturation was not reached at 3 mM IP3. The R437A mutant exhibits a much higher K D value of 673 Ϯ 36 M as compared with the wild type protein (55 M) (supplemental Fig. 4). For the N481D mutant, the IP3 binding to the concave side almost vanished; only very weak chemical shift perturbations were observed at 3 mM IP3. The Arg-437 and Asn-481 mutant data clearly demonstrate that not only residues from the basic patch but also residues on CBL3 constitute the structural link between Ca 2ϩ and PIP2 binding to the C2A domain of rabphilin-3A. This points out the functional importance of the conformational change of CBL3 for PIP2 binding as predicted from our docking model.

DISCUSSION
The aim of this work was to gain further insight of the interplay between the C2A domain of rabphilin-3A, Ca 2ϩ , and PIP2 binding at the atomic level. To achieve this goal, the NMR structure of the Ca 2ϩ -bound C2A domain was required because only the Ca 2ϩ -free crystal structure of this domain had been described so far (27). Because of the low Ca 2ϩ binding affinity of the isolated C2A domain in solution (320 M in presence and 1.1 mM in absence of IP3 at pH 7 (25)), its Ca 2ϩ -bound NMR structure had to be determined at an apparently unphysiologically high Ca 2ϩ concentration (20 mM). This low intrinsic Ca 2ϩ binding affinity is well known for C2 domains. It turns into a high apparent Ca 2ϩ binding affinity in the presence of the target phospholipids, which complete the Ca 2ϩ coordination spheres. The gain between the low intrinsic Ca 2ϩ binding affinity in the absence of phospholipids and the high apparent Ca 2ϩ binding affinity of C2 domains in their presence can reach up to 3 orders of magnitude (47). In the case of rabphilin-3A, the C2B domain should actually dock to the membrane first (25) due to its much higher intrinsic Ca 2ϩ binding affinity (28) compared with the C2A domain, resulting in a high local concentration of phosphatidylserine, the primary lipid target of the C2 domains of rabphilin-3A, and thus, in a dramatic decrease of the K d of the C2A domain for Ca 2ϩ binding. Synergistically, an increase of the Ca 2ϩ binding affinity upon Ca 2ϩ -dependent binding of the C2A domain to PIP2 may then further strengthen the interaction between the C2A domain and the membrane at physiological Ca 2ϩ concentrations, in line with our Ca 2ϩ titration data in the presence of IP3 (25).
In this context the question arises if IP3 is a good mimic at all to study the interaction of PIP2 with the Ca 2ϩ -loaded C2A domain. As already mentioned above, phosphatidylserine was identified as the key phospholipid enabling high apparent Ca 2ϩ binding affinity for the C2 domains of rabphilin-3A via the Ca 2ϩ bridging mechanism, whereas PIP2 was shown to further increase the membrane binding affinity by specific Ca 2ϩ -dependent interaction with both C2 domains (14). Here we have confirmed that in solution IP3 interacts indeed in a specific Ca 2ϩ -dependent manner with the C2A domain. We were able to delineate the molecular determinants of this interaction that increases the intrinsic Ca 2ϩ binding affinity of this domain but differs from the Ca 2ϩ bridging mechanism. This is in agree-ment with the higher membrane binding affinity of the C2A domain in the presence of PIP2 observed in liposome binding experiments (14). Moreover, the Ca 2ϩ binding affinity resulting from IP3 binding to the C2A domain is comparable with the Ca 2ϩ binding affinities observed for the C2 domains of synaptotagmin I in solution (48). Noteworthy, the Ca 2ϩ binding affinity determined here is close to the concentrations of Ca 2ϩ microdomains required for the vesicle fusion process (49). For several other C2 domains it was also shown that the formation of localized pools of PIP2 plays an important role for enhancing Ca 2ϩ binding affinity (26,50). This may also be the case for the C2 domains of rabphilin-3A.
It is widely assumed that Ca 2ϩ binding to C2 domains does not induce major conformational changes and that the interaction with acidic phospholipids takes place by the change of the electrostatic potential of the CBR upon Ca 2ϩ binding (47). Here we have shown that the C2A domain of rabphilin-3A does not follow this rule. Indeed, structural comparison of the Ca 2ϩbound form with the Ca 2ϩ -free form revealed that Ca 2ϩ binding induces large conformational changes within CBL1 and CBL3 and leads to the rupture of two hydrogen bonds (between Glu-482 and Asp-413 and between Glu-475 and Arg-437 (Fig.  4B)). These conformational changes are probably associated with a high activation energy, which explains the low affinity of the C2A domain for Ca 2ϩ and the strong, structurally determined cooperativity between the two Ca 2ϩ binding sites. Very similar properties were also observed for EF-hand Ca 2ϩ binding modules found e.g. in calmodulin or calbindin where the transition from the Ca 2ϩ -free to the Ca 2ϩ -bound form involves major conformational rearrangements (51).
Next, comparison of the Ca 2ϩ -bound structure with the Ca 2ϩ -free form (27) reveals that the Ca 2ϩ -induced conformational change of the CBR appears to trigger the interaction with IP3 by shifting the CBL3 into a favorable conformation. Indeed, the NMR structure of the Ca 2ϩ -bound C2A domain shows a conformational change in the CBR (Fig. 4), resulting in a larger exposure of the concave side of the C2 domain, which harbors the basic cluster essential for PIP2 binding to C2 domains (52). Moreover, our mutagenesis studies and our docking model show that residues His-480 and Asn-481 from CBL3 also play a central role in IP3 binding by clamping the phosphate 1 of IP3, whereas the other phosphates interact with basic residues from the strands ␤ 3 and ␤ 4 of the core domain (Fig. 6). In particular, Asn-481 is clearly essential for IP3 binding as its mutation into alanine led to the most pronounced decrease in affinity (930 Ϯ 30 M) as compared with H480A (360 Ϯ 10 M) and R437A (673 Ϯ 36 M). Notably, Asn-481 is well conserved in PIP2 binding C2 domains (Fig. 1).
Previously, only the basic patch located on the concave side of the domain has been described as IP3 recognition site (8).
Here, we have shown that, in addition, residues from CBL3 are essential for IP3 recognition. Moreover, as CBL3 is involved in Ca 2ϩ binding, these residues very likely couple IP3 and Ca 2ϩ binding. Indeed, Ca 2ϩ binding to the C2A domain facilitates IP3 binding via a conformational change of the CBR, particularly within CBL3. On the contrary, in the Ca 2ϩ -free form CBL3 has to undergo a large conformational change to promote IP3 binding. This energetically unfavorable rearrangement is in good agreement with the very low IP3 affinity observed for the Ca 2ϩ -free form of the C2A domain (25). It is also consistent with the fact that Ca 2ϩ loading of the C2A domain is greatly increased by the presence of IP3 in a target-activated messenger affinity mechanism. Indeed, if the binding of IP3 provokes a conformational change of CBL3, the energetic cost of activation for Ca 2ϩ binding should be reduced, which is supported by our mutagenesis studies showing that directly (N481D) or indirectly (R437A) induced conformational changes of CBL3 facilitate Ca 2ϩ binding.
The docking model presented in this work is based on the Ca 2ϩ -bound C2A solution structure. It differs from our previous, Ca 2ϩ -free crystal structure-based model by including the direct interaction of His-480 and Asn-481 with the PIP2 headgroup. This docking model takes into account the structural changes in CBL3 upon Ca 2ϩ binding and the direct involvement of CBL3 in PIP2 binding as shown by our extensive chemical shift mapping of IP3 binding at pH 5.0 as well as by the IP3 binding of several mutants at pH 7.0. The resulting positioning of PIP2 parallel to the main axis of the C2A domain implies that the C2A domain must be sitting rather flat on the membrane surface to keep the phosphate 1 group of PIP2 as well as the Ca 2ϩ ions in the phosphate layer of the membrane (53). The perpendicular orientation of the inositol ring to the main axis of the C2A domain in our previous model (25) was also in line with a flat positioning of the C2A domain on the membrane. But in the light of the data for the Ca 2ϩ -bound C2A domain presented here this former docking model lacked the interaction with CBL3 residues as an essential feature for a detailed structural and functional description of the Ca 2ϩ -dependent C2A domain-PIP2 interaction.
We were able to explain in this work the remarkable interconnection of the Ca 2ϩ and PIP2 binding of the C2A domain of rabphilin-3A. This was possible because of the availability of the structures in Ca 2ϩ -free and Ca 2ϩ -bound form combined with the full description of the IP3 binding site on the C2A domain. Taken together, our results show how the interdependency of Ca 2ϩ and IP3 binding is mediated by CBL3. As a functional consequence, this structural feature implies that both the messenger (Ca 2ϩ ) and the target (PIP2) have to be simultaneously present to trigger an efficient targeting of the C2A domain to the plasma membrane. It would be highly interesting to find out if the structural determinants that we found for the interdependency of PIP2 and Ca 2ϩ binding in this C2A domain are unique or if this architecture is used more generally in C2 domains with similar functional features.