A novel dimeric active site and regulation mechanism revealed by the crystal structure of iPLA2β

Calcium-independent phospholipase A2β (iPLA2β) regulates several physiological processes including inflammation, calcium homeostasis and apoptosis. It is linked genetically to neurodegenerative disorders including Parkinson’s disease. Despite its known enzymatic activity, the mechanisms underlying pathologic phenotypes remain unknown. Here, we present the first crystal structure of iPLA2β that significantly revises existing mechanistic models. The catalytic domains form a tight dimer. The ankyrin repeat domains wrap around the catalytic domains in an outwardly flared orientation, poised to interact with membrane proteins. The closely integrated active sites are positioned for cooperative activation and internal transacylation. A single calmodulin binds and allosterically inhibits both catalytic domains. These unique structural features identify the molecular interactions that can regulate iPLA2β activity and its cellular localization, which can be targeted to identify novel inhibitors for therapeutic purposes. The structure provides a well-defined framework to investigate the role of neurodegenerative mutations and the function of iPLA2β in the brain.

ARs can be stacked side-by-side forming elongated linear structures. Five conserved amino acids form a hydrophobic core holding the helical repeats together. The remaining amino acids are variable, but the 3D structure of the AR is highly conserved.
The cellular localization of iPLA 2 β is tissue-specific and dynamic 13,20,51 . Different variants of iPLA 2 β are associated with the plasma membrane, mitochondria, endoplasmic reticulum and the nuclear envelope. iPLA 2 β lacks trans-membrane domains, but is enriched in putative protein-interaction motifs. Those include several proline-rich loops and the extended ANK domain with 7 or 8 ARs capable of interacting with multiple cognate receptor proteins 49,50,52 . However, relatively little is known about iPLA 2 β protein interaction mechanism.
It binds calmodulin kinase (CaMKIIβ) in pancreatic islet β-cells 53 and the ER chaperone protein calnexin (Cnx) 54 . The functional significance and mechanisms of both interactions remains unknown. Pull-down of proteins isolated from β-cells under mild detergent treatment revealed a number of other proteins from different cellular compartments, including transmembrane proteins 54 . iPLA 2 β was also found in the Arf1 interactome, which regulates cell morphology 55 .
Understanding the mechanisms of the diverse iPLA 2 functions requires knowledge of its spatial and temporal localization, which are most likely guided by poorly understood proteinprotein interactions. Overall, structural studies are currently limited to identification of the putative calmodulin binding sites 56 , molecular modeling, and mapping of the membraneinteraction loop using hydrogen/deuterium exchange mass spectrometry [57][58][59] .
Here, we present the first crystal structure of a mammalian iPLA 2 , which revises previous structural models and reveals several unexpected features critical for regulation of its catalytic activity and localization in cells. The protein forms a stable dimer mediated by CAT domains, with the active sites poised to interact cooperatively, facilitating transacylation and, potentially, other acyl transfer reactions. The structure suggests an allosteric mechanism of the inhibition, where a single CaM molecule interacts with two CAT domains altering the conformation of dimerization interface and active sites. Surprisingly, ANK domains in the crystal structure are oriented toward the membrane interaction interface and are ideally positioned to interact with membrane proteins. Structural data suggest a novel ATP-binding site in the AR and the role of ATP in regulating protein activity.

Results
Structure of iPLA 2 β. The structure of the short variant of iPLA 2  (SH-iPLA 2 , 752 amino acids) was solved by a combination of selenomethionine single-wavelength anomalous diffraction (SAD) with molecular replacement (MR) using two different protein models. Those include the patatin 48 , with 32% sequence identity to the CAT domain, and four ARs of an ankyrin-R protein 60 (Fig. S1). Five additional ARs and several loop regions in CAT were modeled into the electron density map. The sequence assignment was guided by position of 51 selenium peaks and the structure was refined using 3.95 Å resolution data (Table S1,  The core secondary structure elements of the CAT domain are similar to that of patatin with root-mean-square deviation (rmsd) of 3.1 Å for 186 C atoms (Fig. S3a). Consequently, the fold of the CAT domain also resembles that of cPLA 2  catalytic domain 61 , but to a significantly lesser extent. The active site is localized inside the globular domain as in the patatin structure. However, in iPLA 2 , the catalytic residues are more solvent-accessible than in patatin (Fig. S3b). In the latter, the active site is connected to the surface through two narrow channels (Fig. S3c), insufficient for phospholipid binding without significant conformational changes. By contrast, in iPLA 2 , the active site cavity is wide open and can accommodate phospholipids.
The periphery and loop regions differ significantly from those in the patatin structure, with two unique extended proline-rich loops in iPLA 2 β. A long C-terminal -helix (7 in patatin 62 ) is kinked in the iPLA 2  structure and participates in dimerization (described below).

Conformation of the ANK domain.
The electron density map reveals nine ARs in the structure of SH-iPLA 2 , instead of the previously predicted eight. AR 1 is formed by residues 120-147 with a less conserved AR signature sequence motif (Fig. S1). The outer helix of AR 1 is poorly ordered and was omitted from the current model. The C-terminal AR 9 is formed by residues 376-402. Gln396, which is substituted by the 54-residue proline-rich insert in the long variant (L-iPLA 2 ), locates in the short loop connecting two helixes of AR 9 (grey arrow in Fig.   1b). The orientation of the entire ANK domain is completely unexpected (Figs. 1b, 2b). It is attached to the CAT domain at the side opposite to the membrane-binding surface and was expected to form an extended structure oriented away from the membrane to participate in oligomerization 63 . In the crystal structure, it wraps around the CAT domain towards the predicted membrane interfacial surface. This is achieved by the extended conformation of an eighteen amino acid-long connecting loop, illustrated in Figure S6d. Part of the linker is unresolved due to poor electron density, however, the assignment of the ANK and CAT domains to the same molecule is unambiguous in the crystal packing. The outer helices of AR 7 and AR 8 form an extensive hydrophobic interface with CAT. AR 9 partially contributes to this interface as well.
ANK interaction with ATP. iPLA 2 β is the only known phospholipase that interacts with ATP 12 .
The glycine-rich motif was initially proposed as an ATP-binding site 64 . However, this motif is highly conserved through patatin-like phospholipases, where it forms part of the active site. It is also a common element of α/β hydrolases, where it functions as an oxyanion hole coordinating charge distribution during catalysis 65 . To identify the location of ATP binding in iPLA 2 , we soaked protein crystals with 2'MeSe-ATP and collected 4.6 Å anomalous data. A single anomalous peak was consistently found near Trp293 of AR 6 ( Fig. S6a). An electron density near this residue was also found in the 2Fo-Fc map calculated from the SeMet crystal ( Fig.   S6b), confirming a potential interaction with ATP at this location. The density was less pronounced in the native crystal, and since the low resolution of the Se-Met and Se-ATP data did not permit unambiguous modeling of ATP, we did not include it in the final refinement. Importantly, AR 6 (residues 282-308) adopts an unusual conformation. One of its helices is two amino acids shorter than a conventional AR helix. There is a kink of the entire ANK domain at this position, as compared to ankyrin-R. Potential binding of ATP at this location, where an elongated ANK domain structure is disrupted by the short -helix of AR 6 , suggests the importance of ATP to the regulation of ANK domain conformation and thermodynamic stability.
To our knowledge, the interaction of ATP with ARs has been reported only once in the literature. TRPV1 binds ATP within the positively charged inner concave surface of three ARs 66 . In iPLA 2 β, AR 6 and AR 7 also possess several basic residues in a corresponding surface area. However, the position of the anomalous peak next to Trp293 suggests a potential stacking interaction. Interestingly, it was shown that both purine nucleotides, ATP and GTP, have similar effects on iPLA 2 β activity 15,67 . To probe the dimerization interface, we substituted Trp695 with Glu (W695E). Trp695 forms extensive hydrophobic interactions with the opposite monomer, including a stacking interaction with its counterpart (Fig. S6c). The mutant is a monomer in solution and is inactive (Fig. 4A,   S4g). The W695A mutant exists in equilibrium with both monomeric and dimeric peaks and is catalytically active (data not shown).

CAT-mediated dimerization
The monomers are related by a two-fold axis rotational symmetry. Two active centers and the predicted membrane-binding loops 57 are oriented in the same direction (Fig. 3d).
Importantly, the active sites are in the immediate vicinity of the dimerization interface and in close spatial proximity to each other (Fig. 3). The catalytic Asp598 is at the beginning of a helical loop (599-603) and two leucines of this loop form contacts with the long -helix (604-624) of the opposite monomer. This arrangement suggests a strong allosteric association between the two active sites and dependence of the catalytic activity on the dimer conformation.
Calmodulin binding mechanism. Calmodulin inhibits iPLA 2 β enzymatic activity in the presence of calcium. It was proposed to tightly interact with iPLA 2 β even at low calcium concentrations 68 and to be displaced by active mechanisms, such as covalent modification of the active site by acyl-CoA 69 or by interaction with a calcium influx factor released from the ER during calcium depletion 70 . Two putative CaM-binding peptides containing the canonical IQ and 1-9-14 motifs were previously isolated by tryptic footprinting and affinity chromatography using CaM-agarose 56 . We measured the K i of iPLA 2  inhibition by CaM using a fluorogenic activity assay with Pyrene-PC fluorescent phospholipid liposomes (Fig. S5a-e). The results revealed a tight calcium-dependent interaction with CaM with a K i of 23 ± 1.5 nM (Fig. 4b) and a Hill coefficient n = 2.2 ± 0.2, indicating potential cooperativity. Next, we measured the direct interaction of CaM with iPLA 2  using fluorescent polarization with fluorescein (FAM)-labeled CaM (Fig. 4c). The dissociation constant of the interaction of CaM with iPLA 2 β (K d ) of 112 ± 5 nM was higher than the K i measured with unmodified CaM; however, it corresponds to the K i of FAM-CaM (Fig. S5f). No cooperativity was observed in the direct binding experiment.
Remarkably, the interaction of FAM-CaM with the monomeric W695E mutant was at least an order of magnitude weaker, with a K d >1400 nM (Fig. 3c), suggesting that iPLA 2  dimerization is crucial for CaM binding. The interaction of CaM with synthetic isolated FAMlabeled peptides corresponding to 1-9-14 and IQ motifs was even weaker. Affinity towards the 1-9-14 motif (K d = 2500 ± 400 nM) was comparable to that of the monomeric W695E mutant.
Finally, an excess of CaM enabled W695E dimerization in sedimentation velocity AUC experiments (Fig. 3a). These data strongly support the model where a single CaM molecule interacts with an iPLA 2 β dimer and explain potential cooperativity in the inhibition assay. in the IQ motif did not affect iPLA 2  inhibition by CaM (data not shown). Together, results from solution studies and the conformation of potential CaM-binding sites in the iPLA 2 β dimer suggest that one CaM molecule interacts with two monomers of the iPLA 2 β dimer, most likely through the 1-9-14 motifs.
Discussion. The first crystal structure of iPLA 2 β has revealed several unexpected features underlying its enzymatic activity, mechanisms of regulation and structural domains potentially involved in tissue-specific localization. Previous computer modeling studies used the patatin structure and proposed an interfacial activation mechanism whereby interaction with membrane leads to opening of a closed active site 37 . In the iPLA 2  crystal structure, the active site adopts an open conformation in the absence of membrane interaction (Fig. S3b). Both active sites of the dimer are wide open and provide sufficient space for phospholipids to access the catalytic centers. This is in contrast to patatin, where only two narrow channels connect the catalytic dyad with the solvent exposed surface, and conformational changes are required for substrate to access the active site (Fig. S3c). An open conformation of the active site explains the ability of iPLA 2 β to efficiently hydrolyze monomeric substrates 16  Close proximity of active sites provides a plausible explanation of the previously reported activation mechanism through autoacylation of Cys651. The reaction occurs in the presence of oleoyl-CoA and the modified enzyme is active even in the presence of CaM/Ca 2+ 69 . Cys651 is located at the entrance to the active site at the base of the membrane-binding loop as well as at the dimerization interface (Fig. 3d). Covalent attachment of a long fatty acid chain at this position should increase protein affinity to membrane and can alter the conformation of a CaMbound dimer. The close proximity of two active sites provides an explanation for this autoacylation phenomenon important for the mechanism of enzyme activation in the heart during ischemia.
An intimate allosteric connection of active sites and the dimerization interface provides a plausible mechanism for inhibition by CaM. Indeed, solution studies and location of the putative CaM-binding site strongly suggest that a single CaM binds two molecules of the dimer. We hypothesize that such interactions will lead to conformational changes in the dimerization interface and alter conformation of both active sites.
A hypothetical model of two potential states of iPLA 2  with CaM-bound inactive and CaMfree active dimers is illustrated in Figure 5. In both states, the enzyme is a dimer. The conformation of the dimerization interface differs in the two states depending on interaction of CaM with the 1-9-14 motif. Allosterically, CaM binding stabilizes a closed conformation of the active sites, which remain open in the absence of CaM. The positions of ATP and of acyl modification are shown in the active form. However, the exact mechanism of activation through autoacylation and the effect of ATP-binding on protein activity remain to be further investigated. ANK domains are likely to move out of the conformation observed in crystal structure upon approaching the membrane. In crystals the dimer is shaped as an arch standing on legs formed by the ANK domains (Fig. 2b), with the CAT domains at the top and with their active sites facing downward. The inner radius of the arch is ~80-100 Å. Therefore, in this conformation the ANK domains can prevent the membrane surfaces of larger radii from accessing the catalytic domains. However, the non-specific hydrophobic interactions permit rotational flexibility of interacting domains. Therefore, ANK domains can rotate out of this inhibitory position, while maintaining hydrophobic contacts with the CAT domain. In fact, the relative orientation of the ANK and CAT domains is slightly different between the two monomers of the same crystal. Upon superposition of CAT domains of two monomers, the resulted orientation of ANK domains differs with N-terminal ends shifted by ~12 Å (Fig. S6e).
Similar variation of the ANK domain orientation is a major source of non-isomorphism between different crystals, as observed between native and SeMet crystal forms (data not shown).
The model also illustrates a hypothetical interaction of ANK domains with cytosolic CAMKII and with cytosolic C-terminus of transmembrane calnexin, discussed below.

Methods
Protein purification. The pFastBac vector containing iPLA 2 β cloned from CHO cells with a Cterminal 6XHisTag was used for protein expression as previously described 16 . The CHO iPLA 2 β protein was expressed in Sf9 cells using the Bac-to-Bac system (Invitrogen). Bacmid Growth of suitable protein crystals (diffracting to better than 4Å resolution) was enabled by the counter-diffusion method in capillaries, originally using the Granada Crystallization Box (Hampton Research) 85 , and, later, using a modified capillary method. This method relies on precipitant diffusing through an agarose plug and mixing with the protein solution pre-filled into a ~7 cm long 1 mm diameter capillary. Importantly, counter-diffusion crystals grew over 1-2 weeks and retained diffraction for up to 2 months. Crystals were harvested from drops or capillaries into a cryoprotectant solution containing 68% mother liquor, 10% PEG3350, 10% ethylene glycol, 10% glycerol, 2% ethanol and cryo-cooled in liquid nitrogen. SeMet-labeled  Table 1. To identify parts of crystals suitable for data collection, more than 400 samples were tested with the raster method using a small (5-20 m) beam. The best data sets were collected from elongated crystals using the helical method in order to spread the absorbed dose over larger volume of the crystal and thus reduce the radiation damage to the samples 86 . Data were processed and scaled with HKL2000 87 . It was important to use the "Autocorrection" option during scaling. While it reduced the data set completeness and yielded strongly anisotropic data at resolution higher than 4.4 Å (in the highest resolution range (3.95-4.09 Å), 66% of the data had intensity less than 1σ), it also resulted in data with a lower Wilson B factor and significantly more detailed electron density maps.
Data from the SeMet protein crystal were collected at the selenium absorption peak and inflection wavelengths using helical mode and inverse beam geometry with a 30° wedges.
Analysis of MAD data at two wavelengths with the Phenix suite 88 did not yield a solution. SAD data using peak wavelength produced a solution with 7 selenium peaks. A MR replacement solution was obtained using two different protein models, a patatin 62 and four ARs of an ankyrin-R protein 60 . The structure of patatin was manually trimmed to retain only structural core elements overlapping with CAT domain residues accordingly to the sequence alignment.
The Sculptor program within Phenix was used to prepare four ARs from PDB 1N11. The MR solution contained two copies of each domain. Combination of SAD with MR solution resulted in 51 selenium peaks and a high-quality electron density map (Fig. S2a)  Fluorescent iPLA 2 β Activity Assay. The continuous activity assay was adapted from a protocol used for sPLA 2 89 . 1-hexadecanoyl-2-(1-pyrenedecanoyl)-sn-glycero-3-phosphocholine (Pyrene-PC, ThermoFisher #H361) (Fig. S5a,b) was dissolved as a 1 mM stock in DMSO. The solution was injected into a glass vial containing assay buffer (25 mM HEPES 7.5, 150 mM NaCl, 10% glycerol) over 1 min with shaking to create the substrate mixture. This method resulted in liposomes averaging 100 nm in diameter as determined by dynamic light scattering.
100 L of substrate mixture was added to a black 96-well microplate with a non-binding surface (Corning #3650). 0.2% fatty acid free BSA in the buffer acted as an acceptor for the hydrolyzed 1-pyrenedecanoic acid. Proteins were dialyzed against the assay buffer. iPLA 2 β was incubated with different concentrations of CaM with 1 mM CaCl for 15 min. The baseline fluorescence of the substrate was recorded for 3 min at 340 nm excitation / 400 nm emission using the monochromator of a Biotek Synergy 4 plate reader. 10l of the protein mixture was added to initiate reaction. After a 5 sec mixing step, the fluorescence was read every 30 sec for 1 h or until the signal reached a plateau (Fig. S5c). The linear slope of the first 5 min of the reaction was used as the initial velocity (Fig. S5d,e). The calmodulin inhibition data were fit to the Hill equation using Origin 8.6 software. The velocity in fluorescence units/time was quantified in moles using a standard curve of the 1-pyrenedecanoic acid product. Fluorescence Anisotropy Binding Assays. As calmodulin has no native cysteine residues, a mutant was engineered at Thr34, as described previously 90

Figure 3. Extensive interactions of CAT domains and integrated active sites. a)
Interaction of the CAT domain of molecule B (CAT-B), shown as cyan surface, with the CAT domain of molecule A (CAT-A), shown as yellow cartoon with highlighted catalytic dyad residues (magenta sticks) and the oxyanion hole (green). b) The proximity of the active site to the dimerization interface is illustrated with surface representation of CAT-B (light cyan) and structural elements of the CAT-A active site shown as yellow cartoon, along with the Ser-Asp dyad of CAT-A (magenta stick representation), the oxyanion hole formed by poly-Gly loop (green), and the -helix (red) which contains the catalytic Asp. The structured fragment of 1-9-14 motif is shown in blue. c) The view from the membrane binding surface of the active sites of a dimer with secondary-structure elements and the individual residues color-coded as in b) for molecule A and by light cyan for molecule B. A transparent surface of the dimer is shown in grey. C671 residues of the dimer are represented by yellow and light cyan spheres. These cysteines were previously reported to be acylated in the presence of acyl-CoA and are located on the membrane side of the protein surface. d) Side view of the same structural elements in orientation orthogonal to that in c), illustrating the distance of catalytic dyad residues from the membrane interacting surface and the location of Cys671 at this surface as well as near the dimerization interface.  (83 kDa), and of W695E in the presence of CaM/Ca 2+ in red (138 kDa). Lower MW can be due to unresolved contribution of minor fraction of monomeric species in a latter case. b) Inhibition of iPLA 2  enzymatic activity by CaM in the presence (black) and absence (blue) of Ca 2+ . c) Interaction of FAM-CaM/Ca 2+ with iPLA 2 β as measured by fluorescence anisotropy upon titration by wild-type iPLA 2 β (black) and the W695E mutant (red). Error bars represent average ± s.e.m of triplicate experiments, which were performed at least twice independently. d) and e), two orthogonal views of the iPLA 2 β dimer with monomers color coded in cyan and yellow. CaM in the conformation as reported in the 3SJQ PDB structure is placed next to 1-9-14 motifs (highlighted in blue) to illustrate the possibility of CaM interaction with two 1-9-14 motifs of a dimer. Figure 5. The proposed mechanism of iPLA 2 β regulation and macromolecular interactions. a) Schematic representation of iPLA 2 β dimer in a hypothetical inhibited state bound to CaM. CAT domains are shown in blue and yellow, ANK domains in navy and orange, and a single CaM molecule is represented by two connected circles in pink. Active site cavities are represented by narrow channels (grey lines) leading from the solvent-exposed surface to the Ser/Asp catalytic dyad depicted by magenta. b) An active conformation of the dimer. CaM dissociation leads to the opening of the active sites. ANK domains are available for interactions with protein partners as illustrated for CAMKII (light cyan transparent sphere), known to interact with ANK domain, and with transmembrane Cnx (shown as transmembrane helix with the C-terminal cytosolic peptide in pale yellow), which could recruit iPLA 2  to the membrane. The Cnx-binding site of iPLA 2 β is not known and the hypothetical interaction with ANK domain is based on similar interaction of AnkB and sodium channel peptide. ATP binding (red) in the middle of the ANK domain could trigger additional conformational changes of the AR. Acylation of C671 by oleoyl-coA (green) can facilitate interaction with the membrane and/or opening of active site channels. Other conformational states are feasible as well, such as CaM-bound inhibited protein at the membrane or an open conformation of active sites in CaM-free form in cytosol, corresponding to the crystallized form.   Figure S1. Structure-based alignment of the iPLA 2 β amino acid sequence with patatin (PDB: 4PK9) and the eight ankyrin repeats of ankyrin-R (PDB: 1N11) using TM-align. The ARs are highlighted in light purple, the active site residues starred in magenta, the oxyanion hole in green, and the 1-9-14 CaM-binding motif in blue. Helices and sheets from the structure are depicted with cylinders and straight arrows, respectively, and predicted secondary structure elements in the N-terminus and membrane interacting regions are shown with wavy lines and arrows. Figure S2. iPLA 2 β structure determination by MR/SAD. 2Fo-Fc electron density maps around several regions of the protein, including a) the dimerization interface, b) AR 9 and c) Nterminal -helix contoured at 1.5σ.