Abstract
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.
Introduction
Calcium-independent phospholipase A2β (iPLA2β, also known as PLA2G6A or PNPLA9) hydrolyses membrane phospholipids to produce potent lipid second messengers1,2. Due to its emerging role in neurodegeneration and a strong genetic link to Parkinson’s disease (PD)3-11, the gene coding for iPLA2β was designated as PARK14. Originally isolated from myocardial tissue as an activity stimulated during ischemia12,13, the enzyme displays several specific features including calcium-independent activity, a preference for plasmalogen phospholipids with arachidonate at the sn-2 position, an interaction with ATP14,15 and inhibition by calmodulin (CaM) in the presence of Ca2+ 16. It was also isolated from macrophages, where it was thought to act as a housekeeping enzyme, maintaining the homeostasis of the lipid membrane14. Subsequent studies using the mechanism-based inhibitor bromoenol lactone (BEL) revealed involvement of the enzyme in 1) agonist-induced arachidonic acid release17; 2) insulin secretion18; 3) vascular constriction/relaxation by Ca2+ signaling through store-operated calcium-entry (SOCE) 19,20; 4) cellular proliferation and migration21,22; and 5) autophagy23,24. Alterations in iPLA2β function have demonstrated its role in multiple human pathologies including cardiovascular disease1,25-27, cancer28-31, diabetes32,33, muscular dystrophy34, nonalcoholic steatohepatitis35 and antiviral responses36. Correspondingly, novel inhibitors of iPLA2β have been sought for therapeutic applications37,38. Highly selective mechanism-based fluoroketone inhibitors were designed37,39,40 and successfully applied in mouse models of diabetes41 and multiple sclerosis42. Recently, numerous mutations have been discovered in patients with neurodegenerative disorders such as infantile neuroaxonal dystrophy (INAD)43-45 and PD3-11. The protein was also found in Lewy bodies and its function was connected to idiopathic PD24,46.
More than half of the iPLA2β sequence is comprised of putative protein interaction domains and motifs (Figs. 1a, S1). The sequence can be divided into three parts: the N-terminal domain, the ankyrin repeat (AR) domain (ANK) and the catalytic domain (CAT)47. Hydrolysis by iPLA2β is executed by a Ser-Asp catalytic dyad in close spatial proximity to a glycine-rich motif. The CAT domain is homologous to patatin, a ubiquitous plant lipase48. The AR is a 33-residue motif consisting of a helix-turn-helix structure followed by a hairpin-like loop forming a conserved L-shaped structure. ARs are found in thousands of proteins and have evolved as a highly specific protein recognition structural scaffold49,50. In different proteins, 4 to 24 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 iPLA2β is tissue-specific and dynamic13,20,51. Different variants of iPLA2β are associated with the plasma membrane, mitochondria, endoplasmic reticulum and the nuclear envelope. iPLA2β 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 proteins49,50,52. However, relatively little is known about iPLA2β protein interaction mechanism. It binds calmodulin kinase (CaMKIIβ) in pancreatic islet β-cells53 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 proteins54. iPLA2β was also found in the Arf1 interactome, which regulates cell morphology55. Understanding the mechanisms of the diverse iPLA2β functions requires knowledge of its spatial and temporal localization, which are most likely guided by poorly understood protein-protein interactions. Overall, structural studies are currently limited to identification of the putative calmodulin binding sites56, molecular modeling, and mapping of the membrane-interaction loop using hydrogen/deuterium exchange mass spectrometry57-59.
Here, we present the first crystal structure of a mammalian iPLA2β, 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 iPLA2β
The structure of the short variant of iPLA2β (SH-iPLA2β, 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 patatin48, with 32% sequence identity to the CAT domain, and four ARs of an ankyrin-R protein60 (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, Fig. S2). Residues 1-80, 95-103, 113-117, 129-145, 405-408, 630-631 and 652-670 were omitted from the final model. Regions 81-94, 104-112 and 409-416 were modeled as alanines. The short variant lacks a proline-rich loop in the last AR (Fig. 1) and sequence numbering in the paper corresponds to sequence of the SH-iPLA2β. The structure of the monomer is shown in Figure 1b.
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 Cα atoms (Fig. S3a). Consequently, the fold of the CAT domain also resembles that of cPLA2α catalytic domain61, but to a significantly lesser extent. The active site is localized inside the globular domain as in the patatin structure. However, in iPLA2β, 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 iPLA2β, 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 iPLA2β. A long C-terminal α-helix (α7 in patatin62) is kinked in the iPLA2β structure and participates in dimerization (described below).
Conformation of the ANK domain
The electron density map reveals nine ARs in the structure of SH-iPLA2β, instead of the previously predicted eight. AR1 is formed by residues 120-147 with a less conserved AR signature sequence motif (Fig. S1). The outer helix of AR1 is poorly ordered and was omitted from the current model. The C-terminal AR9 is formed by residues 376-402. Gln396, which is substituted by the 54-residue proline-rich insert in the long variant (L-iPLA2β), locates in the short loop connecting two helixes of AR9 (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 oligomerization63. 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 AR7 and AR8 form an extensive hydrophobic interface with CAT. AR9 partially contributes to this interface as well.
ANK interaction with ATP
iPLA2β is the only known phospholipase that interacts with ATP12. The glycine-rich motif was initially proposed as an ATP-binding site64. 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 catalysis65. To identify the location of ATP binding in iPLA2β, 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 AR6 (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, AR6 (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 α-helix of AR6, 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 ARs66. In iPLA2β, AR6 and AR7 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 iPLA2β activity15,67.
CAT-mediated dimerization of iPLA2β
The crystallographic asymmetric unit is formed by a dimer of iPLA2β. In contrast with the original hypothesis of ANK-mediated oligomerization63, this dimer is formed through CAT domains. The ANK domains are oriented outwards in opposite directions, forming a ∼150 Å-long elongated structure. Analytical ultracentrifugation (AUC) experiments support the existence of an elongated dimeric structure in solution (Fig. 4a). The experimental MW was 152 kDa, corresponding to a dimer with a theoretical MW of 170 kDa. The friction ratio of 1.9 (compared to 1.4 for BSA) corresponds to a significant deviation from a globular shape.The isolated ANK domains did not oligomerize in an AUC experiments (Fig. S4b).
CAT domains interact through an extended, largely hydrophobic dimerization interface with a contact area of ∼2800 Å2 (Fig. 3) formed by three α-helices, several loops, including the long loop 554-570, and a part of the central β-sheet. Such an extensive interaction supports a stable dimer. Correspondingly, iPLA2β dimerizes even at nanomolar concentrations (Fig. S4a). 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).
The monomers are related by a two-fold axis rotational symmetry. Two active centers and the predicted membrane-binding loops57 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 π-helical loop (599-603) and two leucines of this loop form contacts with the long α-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 iPLA2β enzymatic activity in the presence of calcium. It was proposed to tightly interact with iPLA2β even at low calcium concentrations68 and to be displaced by active mechanisms, such as covalent modification of the active site by acyl-CoA69 or by interaction with a calcium influx factor released from the ER during calcium depletion70. 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-agarose56. We measured the Ki of iPLA2β 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 Ki 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 iPLA2β using fluorescent polarization with fluorescein (FAM)-labeled CaM (Fig. 4c). The dissociation constant of the interaction of CaM with iPLA2β (Kd) of 112 ± 5 nM was higher than the Ki measured with unmodified CaM; however, it corresponds to the Ki 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 Kd>1400 nM (Fig. 3c), suggesting that iPLA2β dimerization is crucial for CaM binding. The interaction of CaM with synthetic isolated FAM-labeled peptides corresponding to 1-9-14 and IQ motifs was even weaker. Affinity towards the 1-9-14 motif (Kd = 2500 ± 400 nM) was comparable to that of the monomeric W695E mutant. Binding of the IQ motif was even weaker (Kd=5900 ± 800 nM) (Fig. S5h).
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 iPLA2β dimer and explain potential cooperativity in the inhibition assay. Furthermore, the two 1-9-14 motifs are located on the same side of the dimer and are ∼30 Å apart from each other (Figs. 4d,e). In the structure of the small conductance potassium channel complex with CaM (PDBID: 3SIQ)71, a single CaM molecule in an extended conformation interacts with the channel dimer and the distance between CaM-binding helixes is also 30 Å. In Figures 4d and 4e, CaM from the 3SIQ complex is placed next to an iPLA2β dimer to illustrate comparable distances. At the same time, the conformation of the IQ motif in tertiary structure makes it as unlikely target of CaM binding. This motif overlaps with a β-strand of the conserved structural core of the molecule and is inaccessible for binding without protein unfolding. Moreover, mutation of the most conserved hydrophobic Ile to a charged Asp (I701D) in the IQ motif did not affect iPLA2β inhibition by CaM (data not shown). Together, results from solution studies and the conformation of potential CaM-binding sites in the iPLA2β dimer suggest that one CaM molecule interacts with two monomers of the iPLA2β dimer, most likely through the 1-9-14 motifs.
Discussion
The first crystal structure of iPLA2β 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 site37. In the iPLA2β 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 iPLA2β to efficiently hydrolyze monomeric substrates16 and the lack of a strong interfacial activation such as observed with cPLA2, where membrane binding increases activity by several orders of magnitude72.
Dimeric active site
The dimer is formed by CAT domains tightly interacting through an extensive interface, while ANK domains are oriented outwards from the catalytic core. The existence of the dimer in solution was confirmed by quantitative sedimentation velocity and cross-linking experiments. This configuration was verified by mutagenesis of the observed dimerization interface and a lack of oligomerization of isolated ANK domains. The elongated shape of the dimer contributes to an overestimation of the previously reported oligomeric state in gel filtration analysis due to faster migration of elongated molecules through the size-exclusion matrix. A remote iPLA2β homolog from C. elegans also forms a dimer in solution24.
The catalytic centers are in immediate proximity to the dimerization interface and the activity is likely to depend on the conformation of the dimer. Disruption of the dimer by the W695E mutation yields an inactive enzyme. The active sites are also in close proximity to each other and allosterically connected. Concerted activation of closely integrated active sites should promote rapid responses upon stimulation by ligands, rendering the enzyme an efficient sensor of external perturbations.
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/Ca2+ 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 CaM-bound 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 iPLA2β with CaM-bound inactive and CaM-free 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 CAMKIIβ and with cytosolic C-terminus of transmembrane calnexin, discussed below.
Novel structure suggests a role for ANK domains in membrane localization
The iPLA2β field currently lacks a coherent explanation of how it localizes to the membrane versus cytosol in different cell types and tissues. Elongated ARs form highly specific docking surfaces for different protein partners and 4-8 ARs are capable of binding several proteins73. The finding that the ANK domains are oriented towards the membrane-facing side of iPLA2β suggests that ARs represent the most logical site of putative interaction with membrane proteins. Indeed, specific protein recognition is the only known function of ARs. The canonical ARs tether cytoskeletal proteins to ion channels60.
The interaction of CAMKIIβ with the N-terminal part of iPLA2β has been reported. The function of such interaction remains poorly understood. It can either modulate activity of each protein or mediate recruitment of iPLA2β to membrane since CaMKIIβ form complexes with actin, an L-type Ca2+ channel regulatory protein a1C, and synaptic proteins like NMDA receptors, densin-180 (a trans-synaptic protein), and α-actinin (an actin-binding protein). In case of strong interaction, CaMKIIβ may also play a role of iPLA2β storage pool due to high abundance of protein in neurons (2% of total protein in hippocampus).
The Cnx-binding domain of iPLA2β remains unknown. This interaction is particularly interesting in light of growing number of data implicating iPLA2β function in ER stress response in β-cells and neurons24,74,75. Cnx is the ER chaperone protein. It consists of the lumenal domain, single transmembrane helix and 90 amino acids long C-terminal cytosolic tail, which can interact with iPLA2β. Interestingly, interaction of elongated unstructured peptides was previously reported for the AnkB protein with an autoinhibitory peptide and a peptide of the Nav1.2 voltage-gated sodium channel76. Hypothetically, the ANK domain of iPLA2β could similarly interact with a portion of Cnx C-terminal peptide.
The proline-rich 54-residue insert in the long variant is predicted to form an unstructured loop protruding away from AR9, which can also interact with other proteins. Alternatively, it can disrupt the conformation of AR9 and alter orientation of the ANK domain. The hydrophobic interface between ANK and CAT domains and long flexible linker can allow for significant movement of the ANK domain, similarly to those observed in Rep helicase77 or prothrombin78. Additional studies are required to uncover the mechanism and functionality of alternative conformations of iPLA2β.
Neurodegenerative mutations
are found in all domains and therefore can affect the enzymatic activity, regulation or protein interaction properties of iPLA2β. In 2006, INAD was linked to mutations in iPLA2β gene (PARK14)43, which later was connected to a spectrum of neurodegeneration disorders, correspondingly termed PLAN (recent summary and references in 79). Those include infantile neuroaxonal dystrophy (INAD1/NBIA2A), atypical neuroaxonal dystrophy (NAD) and idiopathic neurodegeneration with brain iron accumulation including Karak syndrome (NBIA2B). A different set of mutations was linked to a rapidly progressive young-adult onset dystonia-Parkinsonism (PARK14)5,7,9,11,80-82. As shown in Figures 1a and 6, mutations are spread throughout all domains. Several tested PARK14 mutants retain full24,83 or partial activity5, while several tested INAD mutations lead to catalytically inactive enzyme83. An interesting example of sensitive allosteric regulation is Arg 687 (corresponding number in L-iPLA2β is 741) located at the dimerization interface, which is mutated to Trp in INAD, leading to an inactive enzyme, and to Gln in PD with the activity retained. While an Arg to Trp mutation can significantly alter conformation of the dimerization interface important for catalytic activity, it is unclear what effect a minor Arg to Gln mutation will have and why it cause a late offset (comparatively to INAD) disease. Surprisingly, the A341T mutation in the ANK domain was found to be inactive83. This residue is at the ANK/CAT interface and can affect the interactions and stability of the protein. It should be noted that there are very few enzymatic and biochemical studies of the protein and mutants, mostly limited to semi-quantitative measurements. The new structure will facilitate in-depth analysis of known mutants and their effect on biochemical properties. This will lead to a better understanding of protein function and the mechanism of activity and regulation in numerous cellular pathways and disease states.
The structure should also facilitate ongoing design of small molecule modulators of iPLA2β for therapeutic purposes. Combined with the analysis of disease-associated mutations, our results clearly demonstrate the importance of N-terminal and ANK domains as well as of peripheral regions of the CAT domain, such as the dimerization interface, for the catalytic activity and its regulation. Together with further knowledge of iPLA2β binding partners, such allosteric regions can be targets for small molecule binding to inhibit either enzymatic functions or signaling in downstream pathways.
Methods
Protein purification
The pFastBac vector containing iPLA2β cloned from CHO cells with a C-terminal 6XHisTag was used for protein expression as previously described16.The CHO iPLA2β protein was expressed in Sf9 cells using the Bac-to-Bac system (Invitrogen). Bacmid DNA was transfected into Sf9 cells with Trans-IT transfection reagent (Mirus Bio). After 4 days, the media was collected as the p0 viral stock. This stock was amplified by adding 1 mL of p0 to 100 ml of 2x106 cells/mL for 96 h, creating the p1 viral stock. 25 mL of the amplified p1 was used to infect 500 ml shaker flasks of 2x106 cells/mL for 60 h. The cell pellet was washed with cold PBS and suspended in purification buffer (25 mM HEPES pH 7.5, 20% glycerol, 0.5 M NaCl, 1mM TCEP) containing 50 µg/ml each of leupeptin and aprotinin. The cell suspension was frozen in liquid nitrogen and lysed by thawing and sonication at 50% power, 50% duty cycle 4 times for 2 min each. The lysate was cleared by ultracentrifugation at 35,000 rpm for 1 h. 0.5 M urea and 1 mM TCEP were added to the supernatant and mixed with 5 ml of TALON cobalt resin (Clontech) to bind for 1 h at 4°C. The resin was centrifuged at 1,000 rpm for 1 min to remove the flow-through fraction in batch mode. The resin containing the bound protein was then applied to an empty column, washed sequentially with purification buffer containing 10 mM imidazole (100 mL), 40 mM imidazole (40 mL), and eluted with 15 mL purification buffer containing 250 mM imidazole. iPLA2β and all mutants were >98% pure as determined by SDS-PAGE and Coomassie staining.
The CaM expression plasmid was a gift from M. Shea (University of Iowa). CaM and its mutants were expressed and purified as previously described84.
Crystallization
iPLA2β was concentrated to 6-8 mg/ml in 10 mM HEPES pH 7.5, 500 mM NaCl, 10% glycerol, 5 mM ATP and 1 mM TCEP. Initial crystallization trials were conducted in sitting-drop plates with a Phoenix robot. iPLA2β forms crystals within 24 h in several conditions, and after extensive optimization, two primary conditions were selected: 0.1 M bis-Tris pH 5.5, 10% PEG3350, 0.2 M Na/K tartrate and 0.1 M bis-Tris pH 5.5, 10% PEG3350, 0.2 M sodium acetate. Crystals in sitting drop conditions displayed poor diffraction (5-7Å), high X-ray sensitivity and quick deterioration of diffraction power after few days, even while continuing to grow in size. Alternatively, a higher concentration of protein solution was obtained in the presence of CaM. An equimolar amount of purified CaM was mixed with iPLA2β, reduced with 5 mM DTT, and dialyzed in 10 mM HEPES pH7.5, 150 mM NaCl, 10% glycerol, 1 mM CaCl2. 2 mM ATP was added and the proteins concentrated to 10-12 mg/ml. However, the crystals obtained from iPLA2β in the presence of CaM were identical to those obtained without CaM and SDS-PAGE analysis demonstrated the absence of CaM in the crystals.
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 iPLA2β was produced in Sf9 cells by using methionine-deficient medium (Expression Systems, Davis, CA) and supplemented with 100mg/L L-selenomethionine 16 h after infection. The I701D mutant, which had 2-3 times greater expression than wild-type, was used for the production of the SeMet protein. Purification and crystallization of the SeMet derivative was the same as for native protein.
Structure Determination
X-ray diffraction data were collected on GM/CA@APS beamlines 23ID-B and 23ID-D at the Advanced Photon Source, Argonne National Laboratory. Data collection and refinement statistics are shown in Supplementary 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 µm) 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 samples86. Data were processed and scaled with HKL200087. 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 suite88 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 patatin62 and four ARs of an ankyrin-R protein60. 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) sufficient for modeling of five additional ARs and several loop regions within the CAT domain. Connectivity of CAT and ANK domains was verified by analysis of all pairs of symmetry-related ANK and CAT domains in the crystal lattice which yielded only one pair with sufficiently short distance. The large number of methionines spread throughout the entire sequence permitted an unambiguous assignment of amino acids. During consecutive steps of structural modeling, combined MR/SAD electron density maps were calculated with one of the domains omitted to avoid model bias. Only one copy of each domain was modeled and the structure was refined using a global NCS function and secondary-structure geometry restriction. After completion of model building, the structure was subsequently refined using 3.95 Å resolution data from the native protein crystal. Simulated annealing composite omit maps were extensively used in model building. Several rounds of Rosetta refinement in Phenix were used for the final model. Phi-psi values of 82% of the residues in the final model are in a favorable region of the Ramachandran plot with 1% in an unfavorable conformation. The latter were in loop regions with poor electron density. Residues 1-80, 95-103, 113-117, 129-145, 405-408, 630-631 and 652-670 were omitted from final model and regions 81-94, 104-112 (numbering in both regions is based on secondary structure prediction) and 409-416 were modeled as alanine residues.
4.6 Å resolution SAD data were collected at selenium peak wavelengths from the protein crystals soaked with 2’MeSe-ATP (Jena Bioscience). Combined MR/SAD analysis revealed a single peak. Several alternative models with different omitted domains or domain fragments were used to avoid model bias. All calculations resulted in an identical position of the selenium peak.
Fluorescent iPLA2β Activity Assay
The continuous activity assay was adapted from a protocol used forsPLA89. 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 µL 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. iPLA2β 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. 10µl 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 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 previously90, to enable coupling of FAM fluorophore in a site-directed manner. This enabled to measure direct binding of FAM-CaM the using fluorescence anisotropy method. The CaM T34C mutant was created by mutagenesis, confirmed by sequencing, and purified with the same procedure as described for the native protein. The labeled protein was separated from excess FAM with phenyl sepharose in the same procedure as for purification. The concentration of labeled protein was measured at 495nm with a molar extinction coefficient of 68,000 M-1 cm-1. For the fluorescence binding assay, proteins were dialyzed to the assay buffer (described in activity assay). CaM-FAM (30 nM final concentration) was incubated with a series of iPLA2β concentrations obtained by 2-fold serial dilution in a 384 well non-binding plate (Corning #3573) in a total volume of 80µL. After 15 min incubation at 25°C, the overall fluorescence intensity and the parallel and perpendicular components were read on a Biotek Synergy 4 with 485 nm excitation and 528 nm emission filters. The fluorescence anisotropy was calculated by the Biotek Gen5 software using the following equation: A=(F∥−F⊥)/ (F∥−2F⊥), where F∥ and F⊥ are the parallel and perpendicular intensities, respectively. Each experiment was conducted in triplicate at least two independent times and values shown are the average ± SEM.
Analytical Ultracentrifugation
Proteins were extensively dialyzed against AUC buffer (25mM HEPES 7.5, 500mM NaCl, 10% glycerol). Sedimentation velocity studies were performed in a Beckman XL-A analytical ultracentrifuge at 20°C and 35,000rpm. The absorbance at 280nm was collected every 4 min for a total of 200 scans. The buffer viscosity and density as calculated by Sednterp (http://www.rasmb.org/sednterp) were 1.04913 ρ and 0.01436 η, respectively. These values were used to fit the data to the Lamm equation in SEDFIT software91 using the continuous c(s) distribution model. Graphs were prepared using GUSSI software (UT Southwestern).
Author Contributions
K.R.M. and S.K designed, performed, and analyzed experiments, including crystallographic data collection and refinement. C.M.J. and R.W.G. provided purified protein for initial crystallization experiments, plasmid for protein production, and provided expertise in studying the effectors of this enzyme. O.K. cloned multiple protein isoforms in E. coli and insect cells, developed protein purification and SeMet labeling protocols and contributed to crystallization and activity measurements. I.M. assisted with cloning of mutants, cell culture, protein purification, activity and AUC assays. R.S. collected data set from native protein crystal and helped with data collection strategies for selenium protein derivatives. K.R.M. and S.K wrote the paper with input from all authors.
Competing Financial Interests
The authors declare no competing financial interests.
Accession codes and data availability
Atomic Coordinates and structure factors for the iPLA2β structure have been deposited in the Protein Data Bank under accession code PDB ID 6AUN. All reagents and relevant data are available from the authors upon request.
Acknowledgements
We thank Aaron Naatz for help in purifying of E. coli expression constructs, Praveen Subramanian for preliminary work on fluorescence assays of PLA2 enzymes and all members of the Korolev lab for helpful discussions. We are grateful to Nicola Pozzi, Enrico Di Cera, David Ford, Jane McHowat and William S. Sly for extremely helpful discussions and to Joel Eissenberg for manuscript preparation. GM/CA-CAT beamlines 23-ID-B and 23-ID-D at the Advanced Photon Source, Argonne National Laboratory are funded in whole or in part from the NCI (ACB-12002) and the NIGMS (AGM-12006). This research was supported by the American Heart Association Grant-in-Aid #0665513Z, Center for Advancement of Science in Space (CASIS) grant CASIS-2016-1, NIH/NINDS grant R21NS094854, to S.K. and NIH/NHLBI grant R01HL118639, to R.W.G.
References
- 1.↵
- 2.↵
- 3.↵
- 4.
- 5.↵
- 6.↵
- 7.↵
- 8.
- 9.↵
- 10.
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.
- 27.↵
- 28.↵
- 29.
- 30.
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵