Structural Basis for Ca2+-mediated Interaction of the Perforin C2 Domain with Lipid Membranes*

Background: Perforin is a critical component of immune homeostasis, responsible for clearing virally infected cells. Results: The molecular details of calcium binding by the perforin C2 domain are revealed. Conclusion: Calcium-mediated structural rearrangement activates perforin for membrane binding. Significance: The C2 domain regulates membrane binding by calcium-dependent events, which have now been defined for a mammalian perforin C2. Natural killer cells and cytotoxic T-lymphocytes deploy perforin and granzymes to kill infected host cells. Perforin, secreted by immune cells, binds target membranes to form pores that deliver pro-apoptotic granzymes into the target cell. A crucial first step in this process is interaction of its C2 domain with target cell membranes, which is a calcium-dependent event. Some aspects of this process are understood, but many molecular details remain unclear. To address this, we investigated the mechanism of Ca2+ and lipid binding to the C2 domain by NMR spectroscopy and x-ray crystallography. Calcium titrations, together with dodecylphosphocholine micelle experiments, confirmed that multiple Ca2+ ions bind within the calcium-binding regions, activating perforin with respect to membrane binding. We have also determined the affinities of several of these binding sites and have shown that this interaction causes a significant structural rearrangement in CBR1. Thus, it is proposed that Ca2+ binding at the weakest affinity site triggers changes in the C2 domain that facilitate its interaction with lipid membranes.

trations of Ca 2ϩ . Perforin is thus activated to bind membranes only in the extracellular environment, where the Ca 2ϩ concentration is estimated to be between 1 and 3 mM (8 -10). It is suggested that the requirement for extracellular concentrations of Ca 2ϩ provides a critical control point for perforin function and helps regulate unwanted perforin activity within the cell (11).
The perforin C2 domain comprises eight ␤-strands in a ␤-sandwich fold (7). Three Ca 2ϩ -binding regions (CBRs) contain conserved Asp residues and mediate the interaction with both Ca 2ϩ and membranes. Different types of C2 domain bind different numbers of Ca 2ϩ ions. For example, the synaptotagmin I C 2 B domain binds two Ca 2ϩ ions (12,13). In contrast, the synaptotagmin I C 2 A domain coordinates three Ca 2ϩ ions (14,15). However, some C2 domains, for example the PKC Apl II (16), can interact with membranes without first binding Ca 2ϩ .
In the original crystal structure of perforin (7), two Ca 2ϩ ions were observed in the CBRs, which were scavenged from the environment and therefore presumably interact with the strongest binding sites. One Ca 2ϩ ion (site II) was coordinated within CBR1 and CBR2 by conserved Asp residues (Asp-435 and Asp-483). The second Ca 2ϩ ion was coordinated by a nonconserved Asp residue (Asp-490) and found outside the CBR3 in an unusual binding position that appears to be unique to the perforin C2 domain (Fig. 1). Currently, it is unknown how many Ca 2ϩ ions are coordinated by perforin in the fully liganded state. However, the crystal structure of a related (39% identity) C2 domain-only protein (SmC2P1 from Scophthalmus maximus) (Fig. 1B) revealed three Ca 2ϩ ions coordinated within the CBRs (17).
In the CBRs of the perforin C2 domain, two pairs of conserved and solvent-exposed aromatic residues in CBR1 (Trp-427 and Tyr-430) and CBR3 (Tyr-486 and Trp-488) have been shown to be critical for membrane binding, as substitution of all four residues with alanine resulted in complete loss of function in effector target cell-based and red blood cell lysis-based assays (17). It is likely that these four aromatic residues are thus essential for tight interaction of the perforin C2 domain with lipid membranes. By analogy with SmC2P1, it is suggested that Ca 2ϩ binding in the CBRs induced a conformational change in CBR1 and subsequently that the relative positions of the four aromatic residues changed to orientations facilitating interaction with lipid membranes (17). However, there is no direct structural evidence of Ca 2ϩ -dependent rearrangements of the CBR1 in the perforin C2 domain, and the numbers of bound Ca 2ϩ ions and their binding sites remain unclear.
To address these questions, we have investigated the interactions of perforin C2 domain with Ca 2ϩ ions by both NMR spectroscopy and x-ray crystallography, using highly soluble and stable variants of mouse perforin C2 domain in which Trp-427, Tyr-430, Tyr-486, and Trp-488 were mutated to alanine. We propose a detailed Ca 2ϩ -binding mechanism of the perforin C2 domain and a role for bound Ca 2ϩ in its interaction with lipid membranes.

Experimental Procedures
Protein Expression and Purification-The C2 quad mutant (W427A/Y430A/Y486A/W488A) was cloned into the pha-gemid vector pComb3X for expression of the protein in the periplasmic space of Escherichia coli Top10FЈ (Life Technologies, Inc.) with C-terminal HA and His 6 tags (Fig. 1B). For crystallography, residues 410 -535 of murine perforin (C2 quad (410 -535)) were cloned into pComb3X by introduction of 5Ј and 3Ј asymmetric SfiI sites using standard molecular biology techniques. For NMR, residues 410 -526 (C2 quad(410 -526)) were cloned into pComb3X and, to avoid an unpaired cysteine in the construct, Cys-524 was mutated to serine by QuikChange mutagenesis according to the manufacturer's protocol (Stratagene). The clones for expression of the D429N, D435N, D483N, D490N, and D491N variants of C2 quad(410 -526) were constructed using pComb3X as the template to introduce the mutation by overlap PCR with flanking 5Ј EcoRI and 3Ј NcoI sites. The amino acid sequences of the constructs used in this study are shown in Fig. 1B. For expression of all constructs, cells were grown overnight in 3 ml of SB (Super Broth) media contained 0.5% glucose (starting culture) at 37°C and subsequently inoculated into 100 ml of SB media (subculture). The subculture was grown at 37°C until an A 600 of 0.5-0.6 and then transferred into 4 liters of SB media (main culture). The main culture was continuously grown at 37°C to an A 600 of 0.6. The temperature was reduced to 23°C, and protein expression was induced by addition of 10 M isopropyl ␤-D-1-thiogalactopyranoside (IPTG) followed by incubation for ϳ20 h. Uniformly 15 N-or 13 C/ 15 N-labeled proteins were produced by using a high cell density method (18) with modifications. Cells were grown using a similar procedure to that for expression of unlabeled protein, except that all SB media contained 0.5% glucose to prevent leaky protein expression, and the main culture was grown overnight at 37°C without IPTG induction. After harvesting cells from 4 liters of main culture, cells were resuspended in 2 liters of minimal media containing 15 NH 4 Cl as the sole nitrogen source or 15 NH 4 Cl and [ 13 C]glucose as the sole nitrogen and carbon sources for the expression of 15 N-or 13 C/ 15 N-labeled proteins, respectively. Cells were incubated for 1 h at 23°C to adapt to minimal media, and protein expression was induced by adding 10 M IPTG. Cells were harvested ϳ20 h after IPTG induction.
Purification of both the C2 quad(410 -535) and C2 quad (410 -526) mutants was performed by periplasmic protein extraction with osmotic shock (19). The cell pellets were resuspended in binding/resuspension buffer (50 mM Tris/Cl (pH 8.0), 300 mM NaCl, 20 mM imidazole, and 0.02% NaN 3 (typical volume is 5 ml per 1 g of wet cells)). An osmotic shock buffer (50 mM Tris/Cl (pH 8.0), 150 mM NaCl, 1 M sucrose, and 2 mM EDTA) was added into the cell mixture with a 1:1 volume ratio and mixed well. After incubation at room temperature for 30 min, the osmotically shocked cells were collected by centrifugation at 10,000 rpm for 20 min. The pellets were resuspended in release buffer (5 mM MgSO 4 and protease inhibitor (Roche Applied Science), typical volume 1 ml per 1 g of osmotic shocked cells) and incubated on ice for 30 min. The supernatant containing periplasmic proteins was collected by centrifugation at 15,000 rpm for 20 min and added to an equal volume of binding/resuspension buffer, and then ϳ2% (v/v) Ni 2ϩ -charged chelating Sepharose (GE Healthcare) was added. After 1.5 h of incubation at 4°C, the mixture was transferred into a column, and the supernatant was isolated from the resin by gravity flow. The resin was washed with 10 column volumes of binding/ resuspension buffer followed by the same volume of wash buffer (50 mM Tris/Cl (pH 8.0), 300 mM NaCl, 40 mM imidazole, and 0.02% NaN 3 ). The purified protein was eluted in 20 ml of elution buffer (50 mM Tris/Cl (pH 8.0), 300 mM NaCl, 300 mM imidazole, and 0.02% NaN 3 ). The eluted fraction was applied onto a Superdex 75 10/300 GL or 16/60 size exclusion column (GE Healthcare) equilibrated in either crystallography buffer (25 mM HEPES (pH 7.4), 150 mM NaCl with 0.02% NaN 3 ) or NMR buffer (20 mM Tris/Cl (pH 8.0), 300 mM NaCl, and 2 mM EDTA). The peak fractions were analyzed by SDS-PAGE and significant protein-containing fractions pooled.
Crystallographic Analysis-Crystallization was carried out by the hanging drop method with a 1:1 mixture of protein and mother liquor at 20°C with the protein concentrated to 9.75 mg/ml. Apo-C2 quad(410 -535) crystals were obtained in 0.1 M MES (pH 6.0), 0.2 M NaCl, and 20% (w/v) polyethylene glycol 2000 monomethyl ether. Crystals of holo-C2 quad(410 -535) were obtained in 0.1 M HEPES sodium (pH 7.5), 0.2 M calcium chloride dihydrate, and 14% v/v polyethylene glycol 400. The crystals were flash-cooled in liquid nitrogen using 25% (v/v) glycerol as the cryoprotectant. Data sets were collected at the Australian Synchrotron MX2 beamline at 100K (20). The data were merged and processed using XDS (21), POINTLESS, and SCALA (22). Five percent of the data set was flagged as a validation set for calculation of the R free . Molecular replacement was carried out using the C2 domain of 3NSJ as a search probe (7). One molecule was found per asymmetric unit cell, and an initial model was generated using PHENIX. Model building was performed using COOT (23), and refinement was performed using PHENIX (24). Crystallographic and structural analysis was performed using the CCP4 suite (25), unless otherwise specified. The figures were generated using MacPyMOL (26), and the structural validation was performed using MolProbity (27). All atomic coordinates and structural factors were deposited in the PDB under codes 4Y1S (apo-C2 quad(410 -535)) and 4Y1T (holo-C2 quad(410 -535)).
NMR Spectroscopy-All NMR measurements were conducted in 20 mM HEPES buffer (pH 7.0) and 150 mM NaCl with 10% 2 H 2 O at 25°C on a Bruker Avance 600 MHz NMR spectrometer equipped with a cryoprobe. Backbone resonance assignments of C2 quad(410 -526) were obtained from analysis of three-dimensional HNCA, HN(CO)CA, HNCACB, and CBCA(CO)NH spectra of a 0.5 mM uniformly 13 where n is the total number of titration points, and i is ith titration point at each [Ca 2ϩ ] from 0 to 30 mM. ␦ i is the chemical shift of either 1 H or 15

Results
Expression and Purification of Perforin C2 Domain Mutants-Expression of the perforin C2 domain in isolation at levels required for crystallography and NMR spectroscopy has not been successful to date, despite extensive attempts over many years (17). We reasoned that the four aromatic residues in the CBRs (Trp-427, Tyr-430, Tyr-486, and Trp-488) (7) represent the major barrier to expression, possibly through driving association of the C2 domain with E. coli membranes during protein expression (Fig. 1A). By mutating these four hydrophobic aromatic residues to alanine (Fig. 1B), using the crystal structure of perforin to accurately define the domain boundaries, and exploiting periplasmic expression, two different C2 mutants W427A/Y430A/Y486A/W488A were expressed successfully (C2 quad(410 -535) and C2 quad(410 -526)). Twodimensional 1 H-15 N SOFAST-HMQC spectra of the both constructs showed well resolved resonances, confirming that both were properly folded (data not shown). C2 quad(410 -535) was used in crystallography experiments and C2 quad(410 -526) for NMR spectroscopy, as it yielded better quality spectra than C2 quad(410 -535).
Crystal Structure of C2 Quad(410 -535)-The structure of apo-perforin remains unknown as the perforin structure contains the aforementioned two Ca 2ϩ ions (7). Thus, we determined the crystal structure of C2 quad(410 -535) in both the partially and fully calcium-liganded states ( Fig. 2 and Table 1). In the 1.6 Å apo-C2 quad(410 -535) structure (Fig. 2, A and B), a single Ca 2ϩ ion was coordinated in the noncanonical position at residue Asp-490, which mirrors one of the two atoms bound in the full-length perforin structure. This Ca 2ϩ ion was modeled as a 0.5 fractional occupancy, which indicates that, despite the presence of EDTA, some Ca 2ϩ was scavenged during the purification process. The 2.6 Å holo-C2 quad(410 -535) structure revealed five Ca 2ϩ ions, four of which occupy the groove in the jaws of the C2 domain in addition to the noncanonical Ca 2ϩ ion coordinated by Asp-490 (Fig. 2, A and C). In comparing the apo-and holo-C2 quad(410 -535) structures, the most striking feature is the absence of the 427-431 portion of CBR1 loop (Fig.  2, A and B), consistent with predictions that this region in perforin is mobile in the absence of calcium (17).
The most significant structural rearrangement driven by Ca 2ϩ binding involves CBR1. This region contains the critical lipid-binding residues Trp-427 and Tyr-430 (both mutated to alanine in this protein) that make key contacts with Ca 2ϩ ions in sites I-III. Here, Ca 2ϩ binding is achieved through Thr-432 and Asp-429. Thr-432, which is the only Ca 2ϩ -binding residue visible in electron density in the apo-structure, moves 6.8 Å (Fig. 2B). Asp-429 moves 11.4 Å, from its position in the fulllength perforin structure to engage Ca 2ϩ (Fig. 2C), thus driving this significant structural rearrangement in CBR1. Elsewhere, conformational changes in the side chains of Asp-485 and Asp-491 in CBR3 position these residues so that they interact with sites I, II, and IV. Finally, a modest rearrangement of Asn-454 on CBR2 completes the coordination of site III (Fig. 2, D and E).
CBR3 undergoes only modest conformational changes upon calcium binding; indeed, the backbone atoms of the previously identified lipid-binding residues 486 and 488 (both alanine in this protein) remain essentially unchanged between the apoand holo-forms. Thus, the key event driven by Ca 2ϩ binding must be the conformational change in CBR1 and the repositioning of Trp-427 and Tyr-430 into an orientation suitable for interacting with membranes. These data are consistent with previous studies showing that the W427A and Y430A mutations caused the most substantial functional defects with respect to perforin membrane binding (17). In comparison with the SmC2P1 structure (PDB codes 3W56 and 3W57), similar positioning of Ca 2ϩ ions at sites I-III was observed. This bonding pattern differs from that of MUNC13-C 2 B, which is the most structurally similar C2 domain characterized previously (7,17,31). Unlike full-length perforin and the structures of apoand holo-C2 described here, SmC2P1 does not coordinate Ca 2ϩ at the noncanonical position.
In the murine perforin structure, the CBR1 loop is stabilized via crystal contacts; our new data confirm the suggestions that CBR1 is largely mobile in the absence of Ca 2ϩ . Given that the site II Ca 2ϩ is occupied in murine perforin (7), we reason this site is probably the strongest affinity Ca 2ϩ -binding site inside the jaws of the CBRs. However, Ca 2ϩ binding to site II alone is clearly insufficient to drive conformational change in CBR1. Disulfide bonds between Cys-496 and Cys-509, and between Cys-524 and Cys-535 are represented by green and cyan sticks, respectively. The residues (410, 411, 461, 471, and 472), whose resonances were undetected in the current NMR study, are shown in yellow. B, sequence alignment of mouse perforin (Prf) C2 and the C2 quad mutant constructs used in the analyses carried out in this study. Amino acid sequences of the mouse perforin C2 domain (C2 WT) and the C2 quad mutants used for the current crystal (C2 quad(410 -535)) and NMR studies (C2 quad(410 -526)) together with SmC2P1 are aligned. The N-terminal signal sequence and the C-terminal additional sequences, including the hemagglutinin (HA) and/or His 6 tag, are shown in gray. The four alanine residues substituted from the WT aromatic residues are illustrated in red. Five conserved Asp residues are highlighted in yellow. The positions of Asp-Asn mutations (Asp-429, Asp-435, Asp-483, Asp-490, and Asp-491) are colored in blue. Sequence numbers of the mouse perforin C2 domain and SmC2P1 are shown above and below the primary sequences, respectively. The positions of the CBRs are boxed. The signal sequence was cleaved during the export process at the position indicated by the arrow, resulting in an alanine overhang at the N terminus. Pairs of Cys residues that form disulfide bonds are highlighted in the same colors as in A. In the C2 quad(410 -526), Cys-524 was mutated to serine (cyan) to remove a free thiol in the construct.
Furthermore, because site IV makes no contact with CBR1, we reasoned that binding of Ca 2ϩ to site I or III must represent the key event that brings Trp-427 and Tyr-430 into a position suitable for interacting with membranes. To further study these events, and to understand the order of Ca 2ϩ -binding events, we conducted NMR studies.
Analysis of Ca 2ϩ -binding Mode of the C2 Quad Mutant-We undertook Ca 2ϩ titration experiments in solution, monitored by NMR. The 1 H-15 N SOFAST-HMQC spectrum of the C2 quad(410 -535) used for crystallography showed a number of random coil chemical shifts in the central region of the spectrum, mostly arising from additional non-native residues at the FIGURE 2. Crystal structure of apo-and holo-C2 quad(410 -535) superimposed with the C2 domain from full-length perforin. A, C2 domain of 3NSJ (magenta) superimposed with both the apo-C2 (cyan) and holo-C2 (orange) quad(410 -535). CBRs 1 and 3 are identified, and the corresponding colored spheres represent the Ca 2ϩ ions for each structure. The view is rotated by 90°to view the C2 domain from the membrane perspective. CBR1 in the apo-C2 quad(410 -535) structure is disordered, and the loop was not built into the density (cyan dashed line). B, superimposed structures of the apo-C2 (cyan) and holo-C2 (orange) quad(410 -535) to illustrate the re-organization that occurs in the residues involved in Ca 2ϩ coordination. The residues from Ala-427 to Ala-431 (CBR1, cyan dashed line) were not visible in the electron density map of the apo-structure and were not modeled into the final structure. Significant movement is observed in the CBR1, where residue Thr-432 moves through 6.8 Å, after which the CBR1 loop becomes well ordered and visible in the holo-C2 structure. Upon movement of the loop, Asp-429 is re-positioned to engage Ca 2ϩ in positions I-III. C, superimposition of the C2 domain from murine perforin and the holo-C2 quad(410 -535) further demonstrates the significant movement of residue Asp-429 over 11 Å. Throughout, key residues are represented in stick form labeled by residue number; arrows indicate movement, and disulfide bonds are represented as yellow sticks. Ca 2ϩ is numbered as described previously (17), and the additional observed Ca 2ϩ is numbered sequentially (position IV). D, key residues of the holo-C2 quad(410 -535) structure making contact with Ca 2ϩ . E, schematic representation of Ca 2ϩ -binding interactions in the CBRs. The coordinates of the side chain carboxyl groups of conserved Asp residues and Asp-490 with Ca 2ϩ ion at each site are indicated as dashed lines. The Ca 2ϩ ions at each site are shown as orange spheres with corresponding site numbers. The weakest affinity site (III) is labeled in red.
C terminus introduced during cloning (Fig. 1B). Therefore, we created a new construct in which the C-terminal residues were removed after residue 526 (C2 quad(410 -526)) and only a hexa-His tag was attached at the C terminus (Fig. 1B). The 1 H-15 N SOFAST-HMQC spectrum of this new construct was less cluttered in the central region (supplemental Fig. S1A).
Complete backbone resonance assignments of C2 quad (410 -526) (supplemental Fig. S1B) were obtained by conventional triple-resonance experiments, except for proline residues, two residues at the N terminus (410 and 411) and residues 461, 471, and 472. Residues 461, 471, and 472 are located in loop regions in the crystal structure (Fig. 1A) and are subject to exchange broadening among more than two conformers and/or exchange with water molecules, resulting in undetectably broad resonances under our solution conditions.
Prior to Ca 2ϩ titrations, protein samples were treated with EDTA to ensure that residual Ca 2ϩ ions were completely removed, and then the EDTA was removed. The 13 spectra with different Ca 2ϩ concentrations are superimposed in Fig. 3. The chemical shifts of many resonances in the CBRs changed smoothly with increasing Ca 2ϩ , indicating that the exchange between free and bound Ca 2ϩ is fast on the chemical shift time scale (Fig. 3A). However, several of these chemical shift changes were nonlinear, as exemplified by the resonances of Gly-428 or Asp-483. This reflects the fact that different Ca 2ϩ -bound states have different backbone chemical shifts, and several states contribute simultaneously to the observed chemical shifts at any given [Ca 2ϩ ]. Complex titration profiles were also observed for the resonances of Ala-484 and Asp-490 (Fig. 3, B and C).
The calcium titrations were conducted at a protein concentration of 0.15 mM and a [Ca 2ϩ ] range of 0 -30 mM. The initial chemical shift perturbations (Fig. 3, B and C) were largely complete by 0.3 mM Ca 2ϩ , implying K d values Ͻ0.05 mM. The next transition is evident at around 3 mM, reflecting K d values Ͻ0.4 mM for those binding events, whereas the chemical shift perturbations at the highest [Ca 2ϩ ] were still incomplete at 30 mM, implying a K d value of 5-10 mM for the final Ca 2ϩ -binding event. At the typical physiological Ca 2ϩ concentration of the extracellular space, ϳ1-3 mM (8 -10), the weakest Ca 2ϩ -binding site would therefore not be fully occupied.
As expected, the number of Ca 2ϩ -binding sites is five based on the crystal structure of holo-C2 quad(410 -535), and the K d values at several of those sites are expected to be similar to one another. Therefore, the changes in backbone amide chemical shifts are separately summarized up to 3 mM and from 3 to  a Data combine the clashscore, rotamer, and Ramachandran evaluations into a single score, normalized to be on the same scale as the x-ray resolution.
As noted above, most of the perturbed resonances were detected at all [Ca 2ϩ ], but some disappeared with increasing [Ca 2ϩ ] (e.g. Thr-431 in Fig. 3A). This is a consequence of intermediate exchange among two or more Ca 2ϩ -free and multiple Ca 2ϩ -bound states for these resonances. Several of those resonances reappeared at high [Ca 2ϩ ] (e.g. Asn-455 and Asp-485 in Fig. 3A), indicating that they are in a single major conformation at the highest concentrations examined. By contrast, the resonances of Ala-430, Thr-431, Thr-432, and Ala-486, shown in yellow in Fig. 4B, disappeared with increasing [Ca 2ϩ ]. These residues presumably sample more than one conformation even at the highest [Ca 2ϩ ], which is close to saturation of all Ca 2ϩ binding, and these conformers are exchanging with one another on an intermediate time scale that gives rise to peak broadening. These residues are involved in the quad mutation sites, which contain the hydrophobic aromatic residues Tyr-430 and Trp-486 in the wild-type perforin C2 domain.
Detailed Ca 2ϩ -binding Mechanism of the C2 Quad Mutant-To analyze details of the Ca 2ϩ -binding mechanism of C2 quad(410 -526), we mutated the conserved Ca 2ϩ -binding Asp residues and nonconserved Asp-490 to Asn (D429N, D435N,  D483N, D491N, and D490N) and repeated the Ca 2ϩ titration experiments. Two distinct areas of Ca 2ϩ titration spectra for each mutant are shown in Fig. 5 (extended spectra are shown in supplemental Figs. S2-S5). The Ca 2ϩ titration spectra of the mutants showed significant differences from those of C2 quad(410 -526), confirming that these Asp residues bind Ca 2ϩ ions, in agreement with the crystal structure of the holo-C2 quad(410 -535).
In the D491N mutant (Fig. 5), significant differences in the perturbation curves were observed for Gly-428, Asn-455, and Ala-484 in comparison with the Ca 2ϩ titration spectra of C2 quad(410 -526) (Fig. 3). Because the crystal structure of holo-C2 quad(410 -535) shows that the carboxyl group of Asp-491 coordinates two Ca 2ϩ ions at sites I and IV (Fig. 2), mutation of Asp-491 to Asn substantially reduced the Ca 2ϩ binding affinity at these sites. A turning point in the titration curves was still observed at 3 mM [Ca 2ϩ ], as seen for Gly-428 and Asp-483, indicating that the weakest affinity site still exists in the D491N mutant. Therefore, the weakest affinity site is either site II or III. Because site II was occupied by Ca 2ϩ scavenged from the buffer in the crystal structure of full-length perforin (7), we argue that site III must be the weakest affinity site.
Site III was confirmed as the weakest affinity site by examination of the D483N mutant. Ca 2ϩ titration spectra of the D483N mutant showed that the resonances of Gly-428 and Asn-483 continued to shift beyond 3 mM [Ca 2ϩ ], suggesting that the weakest affinity site is maintained in this mutant (Fig.  5). The crystal structure of the holo-C2 quad(410 -535) showed that the carboxyl group of Asp-483 interacts with two Ca 2ϩ ions at sites I and II but not site III (Fig. 2), so the observation of chemical shift perturbations at [Ca 2ϩ ] of Ͼ3 mM in this mutant supports the conclusion that site III is the weakest affinity site. This was also supported by Ca 2ϩ titration experiments of the D429N and D435N mutants. The carboxyl groups of both Asp-429 and Asp-435 coordinate a Ca 2ϩ ion at site III (Fig. 2). In the Ca 2ϩ titration spectra of the D429N mutant, similar chemical shift perturbation profiles to those for the D491N mutant were observed for resonances of Gly-428, Asn-455, and Asp-483 at low [Ca 2ϩ ], but beyond 3 mM [Ca 2ϩ ], in contrast, the chemical shift perturbations were very small, indicating loss of the weakest affinity site (Fig. 5). Similarly, the chemical shift changes of the D435N mutant at high [Ca 2ϩ ] (Ͼ3 mM) were as small as those of the D429N mutant (Fig. 5), confirming the absence of the weakest affinity site.
Asp-429 coordinates three Ca 2ϩ ions at sites I-III (Fig. 2). However, the D429N mutant may still be able to coordinate a single Ca 2ϩ ion at site II, as its chemical shift perturbation patterns were similar to those of the D491N mutant at low [Ca 2ϩ ]. As mentioned previously, the D491N mutation does not affect Ca 2ϩ binding at site II. In the D435N mutant, significantly different chemical shift perturbation patterns were observed for residues Gly-428, Ala-484, Trp-453, Asn-455, and Asp-483, showing that the Asp-435 mutation affects a different Ca 2ϩbinding site from those of Asp-491 and Asp-429 (Fig. 5). This is in good agreement with the crystal structure of holo-C2 quad(410 -535), in which the side chain carboxyl of Asp-435 coordinates Ca 2ϩ at sites II and III but not sites I and IV (Fig. 2). The D435N mutant is still capable of coordinating Ca 2ϩ ions at sites I and IV (and the noncanonical position), and the observed titration curves reflect Ca 2ϩ binding events only at these sites. In the crystal structure of holo-C2 quad(410 -535) (Fig. 2), all of the backbone amides in CBR2 are Ͼ10 Å away from sites I and IV, implying that the resonances of backbone amides in the CBR2 should be less affected by the Ca 2ϩ -binding events at these sites. Indeed, the resonances of Trp-453, Asn-454, and Asn-455 in the CBR2 were almost unchanged over the entire Ca 2ϩ titration in the D435N mutant.
Asp-490 coordinates a single Ca 2ϩ ion at a noncanonical position (site V) outside CBR3 (Fig. 2). Ca 2ϩ titrations of the D490N mutant showed similar chemical shift perturbation patterns to those of the C2 quad(410 -526) (supplemental Fig. S6), except that several resonances appeared to be affected by local conformational differences associated with the point mutation. At 30 mM [Ca 2ϩ ], the spectrum of the D490N mutant was similar to that of C2 quad(410 -526) (supplemental Fig. S7), sug-gesting that Ca 2ϩ bound at the noncanonical position had no significant effect on the saturated Ca 2ϩ -binding state of the C2 quad mutant.

Interaction of C2 Quad Mutant with Membranes and Role of Ca 2ϩ
Ions-Because of the replacement of four hydrophobic aromatic amino acid residues with alanine, the C2 quad mutant is considered to have lost its membrane binding capacity (17). However, NMR is capable of detecting weak interactions that cannot be readily monitored by other methods. To analyze the interaction of the C2 quad mutant with membranes, dodecylphosphocholine (DPC) titration experiments were performed. DPC titration spectra of C2 quad(410 -526) at three different  (Fig. 6E). The perturbed residues were localized to a surface containing the CBRs. In addition, the side chain amide resonance of Asn-454 (Asn-454⌬) was also strongly affected by DPC titration, disappearing beyond 3 mM [DPC] (Fig. 6A). The side chain of this residue is exposed and oriented downward in the right view in Fig. 6E. These results strongly suggest a specific interaction with DPC micelles. Significantly changed 1 H chemical shifts (⌬␦ 1 H) of Asp-429, Asp-485, Asp-489, and Asp-490 (Fig. 6D) were fitted to Equation 3 (see under "Experimental Procedures") ( Fig. 7A), and the dissociation constant (K d ) of C2 quad(410 -526) against DPC micelles was determined as ϳ1 mM.
To investigate the effect of Ca 2ϩ ions on the interaction with membranes, DPC titration experiments were performed at different [Ca 2ϩ ]. In the absence of Ca 2ϩ (Fig. 6B), resonances indicative of unfolded or disordered protein appeared in the central region of the spectrum at 3 mM [DPC], and these resonances intensified at 10 mM [DPC]. These data imply that the interaction with DPC micelles partially unfolds the structure of the C2 quad(410 -526) in the absence of Ca 2ϩ .
To assess how many Ca 2ϩ ions are required for interaction with membranes, we undertook a DPC titration in the presence of 2 mM [Ca 2ϩ ], in which the weakest affinity site should be largely unoccupied (Fig. 6C). No CSPs were observed except for the C-terminal residues. In addition, the C2 quad(410 -52) was partially unfolded at 10 mM [DPC], and adding increasing [DPC] resulted in loss of the original well resolved resonances.
Importance of the Weakest Ca 2ϩ -binding Site for Interaction with Lipid Membranes-The titration of C2 quad(410 -526) with DPC demonstrated that occupancy of Ca 2ϩ at the weakest affinity site is critical for interactions with membranes (Fig. 6). The importance of the weakest affinity site was confirmed by DPC titrations with the D491N, D429N, D435N, D483N, and D490N mutants in the presence of 30 mM [Ca 2ϩ ], as shown in Fig. 7, B-F (extended spectra are shown in supplemental Figs. S11-S15 resonance of Asn-454⌬ was clearly perturbed (Fig. 7B). The CSP of Asn-454⌬ was also observed in C2 quad(410 -526) at 30 mM [Ca 2ϩ ]. In addition, no unfolded resonances were observed up to 50 mM [DPC]. Therefore, the D491N mutant can still bind to DPC micelles in the proper orientation. The much smaller chemical shift change indicates that the affinity of DPC micelles became much weaker because of the loss of Ca 2ϩ -binding sites I and IV.
By contrast, resonances indicative of unfolding caused by nonspecific interaction with DPC micelles were observed in the D429N and D435N mutants in the presence of 30 mM [Ca 2ϩ ] with increasing [DPC] (Fig. 7, C and D). Both mutants lose the weakest affinity site, resulting in partial Ca 2ϩ -bound states even at 30 mM [Ca 2ϩ ]. Indeed, in the presence of 50 mM [DPC], the spectra of both mutants are very similar to the C2 quad(410 -526) at 2 mM [Ca 2ϩ ], in which the weakest affinity site should be  unoccupied (Fig. 6C). These data support the critical role of Ca 2ϩ binding at the weakest affinity site in enabling the C2 quad mutant to interact with lipid membranes. Similar destabilization was observed as the D483N mutant interacted with DPC micelles, which maintains the weakest affinity site (Fig. 7E). However, this mutant lost Ca 2ϩ -binding sites II and possibly I, indicating that sequential Ca 2ϩ occupancy at Ca 2ϩ -binding sites II and III is important for proper interaction of the C2 quad mutant with DPC micelles. The DPC titration of the D490N mutant showed similar chemical shift perturbation patterns to that of the C2 quad(410 -526) (Fig. 7F), indicating the Ca 2ϩ binding to Asp-490 is not important for interaction with membranes.

Discussion
Our studies demonstrate that the C2 quad mutant binds multiple Ca 2ϩ ions in the CBRs, the positions of which were identified in the crystal structure. Substitution of the four aromatic residues Trp-427, Tyr-430, Tyr-486, and Trp-488 with alanine resulted in a loss of perforin activity (17), although these four residues are not involved in Ca 2ϩ binding among known C2 structures (14 -16, 32-37). Indeed, the mutations of all four residues to alanine did not influence Ca 2ϩ binding as confirmed by thermodynamic stability (17). The C2 quad mutant is therefore a suitable model for analyzing Ca 2ϩ binding to the perforin C2 domain.
The crystal structures of the apo-and holo-C2 quad(410 -535) (Fig. 2, C and D) demonstrate good agreement with previous results identified through mutational analyses and inferred from the C2 domain-only protein, SmC2P1 from S. maximus (17,38). These structures have identified the exact positions of the coordinated Ca 2ϩ ions and demonstrated that the C2 domain is capable of binding five Ca 2ϩ ions. These structures also provided further evidence for the noncanonical Ca 2ϩ ion coordinated at Asp-490, which was observed previously in the full-length perforin crystal structure but is known to be nonessential for perforin function (38). CBR1 is highly mobile in the absence of Ca 2ϩ , placing it in a conformation that is suboptimal for binding membranes. Upon Ca 2ϩ binding, CBR1 undergoes a very large movement, and the "jaws" of the C2 domain coordinate four Ca 2ϩ ions within the hydrophobic groove, predominantly mediated by Asp residues. Furthermore, Ca 2ϩ binding is a critical regulatory step in the hydrophobically dependent membrane binding of perforin, bringing the CBR1 and CBR3 into close proximity forming a hydrophobic cleft/groove. Although the structures of the C2 domains do not reveal any obvious rearrangements that could possibly trigger conformational change within other domains of perforin, in particular the MACPF domain, our holo-C2 quad(410 -535) structure is the first evidence of the locations of Ca 2ϩ in the C2 domain of perforin and the first indication of an additional canonical Ca 2ϩ coordination position, site IV.
As demonstrated by our NMR titrations with Ca 2ϩ , the affinity of one Ca 2ϩ -binding site is significantly weaker compared with other sites. From mutation analyses of the four conserved Asp residues Asp-429, Asp-435, Asp-483, and Asp-491 in the CBRs, we defined the weakest affinity site to be site III. The crystal structure of the holo-C2 quad(410 -535) showed that the side chain carboxyl group of Asp-429 in CBR1 coordinates three Ca 2ϩ ions at sites I-III, whereas this residue, together with the flanking two residues, is disordered in the apo-form (Fig. 2). The crystal structure of the perforin C2 domain revealed that the CBR1 domain swings out from the Ca 2ϩ -binding pockets, and the carboxyl side chain of Asp-429 is directed toward solvent (7). Comparing the holo-C2 quad(410 -535) and the perforin C2 domain, significant structural differences are seen only in the relative orientation of CBR1 (Fig. 2). A previous study (17) suggested that bound Ca 2ϩ ions induced conformational rearrangement of CBR1, which then facilitated re-positioning of the four key hydrophobic residues to interact with lipid membranes.
The crystal structure of the holo-C2 quad(410 -535) showed that Asp-435 binds two Ca 2ϩ ions at sites II and III, and Asp-491 coordinates Ca 2ϩ ions at site I (and site IV), which is consistent with the crystal structure of SmC2P1 (17). In contrast, Asp-429 interacts with three Ca 2ϩ ions at sites I-III, which differs from SmC2P1, where this residue coordinates Ca 2ϩ ions only at sites I and II. Based on Ca 2ϩ titration results by NMR, Asp-429 will be able to coordinate two Ca 2ϩ ions at sites I and II at low [Ca 2ϩ ]. At higher than extracellular [Ca 2ϩ ], in contrast, Asp-429 is capable of coordinating one more Ca 2ϩ ion at site III; subsequently, a conformational rearrangement of CBR1 is induced. The coordination of Ca 2ϩ ion at site III by the carboxyl group of Asp-429 is also seen in other C2 domains, for example synaptotagmin I (12,14) and the PKC-␤ (35), both of which bind three Ca 2ϩ ions in the CBRs.
The importance of the conformational rearrangement of CBR1 for the C2 domain interaction with membranes was demonstrated by DPC titration experiments. It is clear that bound Ca 2ϩ stabilizes the structure of the C2 quad mutant in its interaction with DPC micelles. These data are in good agreement with previous evidence indicating that perforin is stabilized in the presence of Ca 2ϩ , as determined by a thermal stability assay (17). Importantly, stabilization of the C2 quad mutant by Ca 2ϩ is highly dependent on the [Ca 2ϩ ], and complete stabilization requires full occupancy of Ca 2ϩ ion at the weakest affinity site (site III). If the weakest affinity site is unoccupied, DPC micelles interact with the C2 quad mutant nonspecifically, leading to partial unfolding. The Ca 2ϩ titration spectra showed that Ca 2ϩ binding at the weakest affinity site starts around 3 mM [Ca 2ϩ ], which is well matched with physiological extracellular [Ca 2ϩ ]. However, the Ca 2ϩ -free state at the weakest affinity site is still dominant around these [Ca 2ϩ ]. Because the exchange between the free and the Ca 2ϩ -bound forms is fast in the NMR time scale, the NMR spectra are observed as a population average of the free and the Ca 2ϩ -bound forms. Therefore, observation of the fully Ca 2ϩ -bound state requires excess [Ca 2ϩ ] (ϳ30 mM), at which the Ca 2ϩ -bound state is dominant at the weakest affinity site III. A previous study (38) demonstrated that the Ca 2ϩ concentration required for full activation of perforin in a sheep red blood cell lysis assay is ϳ250 M, which is less than the physiological extracellular [Ca 2ϩ ]. In the case of native perforin, the interaction with membrane should be tighter than the C2 quad mutant. Once Ca 2ϩ binds to the weakest affinity site, the native C2 domain can interact with the membrane, which may occur when a small proportion of the protein is Ca 2ϩ -bound, rather than requiring the dominant population to be in a fully Ca 2ϩbound state. This allows for oligomerization of perforin and activity at less than physiological extracellular [Ca 2ϩ ].
The importance of occupancy of Ca 2ϩ ion at site III was confirmed by the results of DPC titration of the D491N mutant, in which the C2 quad(410 -526) was still stable at 30 mM [Ca 2ϩ ] even when the Ca 2ϩ -binding site I was not occupied. By contrast, the site III-deficient mutants, D429N and D435N, were unstable in the presence of DPC micelles even at 30 mM [Ca 2ϩ ]. In addition, the NMR results for the D483N mutant clearly indicated that the proper conformational rearrangement of CBR1 is required for Ca 2ϩ binding at site III as well as site II. Our results are consistent with previous studies that Asp-429, Asp-435, and Asp-483, but not Asp-491, are critical residues for plasma membrane binding and cell lysis (38).
The conformational rearrangement of CBR1 enables the C2 quad(410 -526) to interact with DPC micelles thorough the CBRs. Based on our observations, we propose the following mechanism of perforin C2 domain interaction with lipid membranes. At less than extracellular [Ca 2ϩ ], although the C2 domain coordinates two Ca 2ϩ ions in the CBRs, the CBR1 does not change conformation and is in the "inactive form." At higher than extracellular [Ca 2ϩ ], three Ca 2ϩ ions are bound within the CBRs. The Ca 2ϩ binding at site III, which is the weakest affinity site, induces a conformational rearrangement of CBR1 that leads to the "active form," facilitating interaction with membranes (Fig. 8). The affinity of the C2 quad(410 -526) for DPC micelles is very low (ϳ1 mM), which is insufficient for functional activities of perforin. Indeed, no perforin activity was observed in the full-length perforin quad mutant (17) because of the lack of four hydrophobic aromatic residues that are crucial for proper interaction with membranes. On the basis of the Ca 2ϩ titration results, the regions, including the quad mutations, will not be fixed as a single conformer even at close to saturating concentrations of Ca 2ϩ , which may reduce the ability of the C2 quad(410 -526) to interact with the membrane. As indicated above, however, the mutation of four key hydrophobic aromatic residues to alanine does not affect Ca 2ϩ binding to the perforin C2 domain. Therefore, the Ca 2ϩ -dependent conformational switch from the inactive form to the active form should also occur in the native perforin C2 domain. The conformational plasticity in the regions including the quad mutations may be necessary for the initial interaction with membranes. Subsequently, in the case of native perforin C2 domain, the four key hydrophobic aromatic residues would be rearranged into the orientation required for tight binding to lipid membranes. Importantly, higher extracellular [Ca 2ϩ ] promotes membrane binding of the C2 domain on granule exocytosis, whereas low [Ca 2ϩ ] prevents premature activation of perforin (3, 38 -40). Our crystal and NMR data are in excellent agreement with these biological results.
In conclusion, we have clarified the mechanism of the perforin C2 domain interaction with membranes and the role of Ca 2ϩ in that process. Our results represent the first observation of structural details regarding the interaction of the perforin C2 domain with lipid membranes and will facilitate further understanding of perforin function.