Introduction

Phospholipase D (PLD) is a ubiquitous class of enzyme found in organisms ranging from viruses and bacteria to yeasts, plants, and animals1. It is capable of hydrolyzing the phosphodiester bonds of glycerophospholipids, resulting in the production of their free head groups and phosphatidic acid (PA)2. PA is regarded not only as an important structural element of membranes but also as a secondary messenger of signal transduction3. It has been shown that the PA generated by PLD is associated with various physiological and pathological processes, such as cellular signaling, metabolism, inflammation, and tumorigenesis1,4,5. Therefore, PLDs in mammalian cells and pathogenic organisms are considered to be valuable therapeutic targets for several human diseases including infectious diseases, neurodegenerative disorders, and cancers5,6,7,8.

To date, many PLDs have been identified in prokaryotic and eukaryotic cells, including in the genus Streptomyces9, Bacillus cereus10, Escherichia coli11, Pseudomonas aeruginosa12, Arabidopsis thaliana13,14, and various mammals15,16. In prokaryotic cells, there is commonly one isoform PLD and amino acid sequences of that PLD exhibiting 68.6–85.57% identity within a given genus such as Streptomyces. In contrast, low identity, about 7.89–20.63%, is shown between different genera17. In mammalian cells, there are six isoforms of PLD, namely canonical PLD1 and PLD2, PLD3, PLD4, PLD5, and PLD68. In plant cells, there is a much larger number of PLD isoforms18,19.

Despite the poor PLD sequence similarity between different organisms, a conserved sequence motif, HxK(xxxx)nD (in which “x” refers to any amino acid residue, known as the HKD motif), has been identified among various prokaryotic and eukaryotic PLD isoforms20. Most of these proteins contain two HKD motifs, which pack together to form the core of the catalytic domain21. The current model of PLD catalysis is a two-step “**-pong” mechanism. Residues of histidine (H), lysine (K), and aspartic acid (D) are directly involved in hydrolysis of the phosphodiester bonds of phospholipids. The histidine residue from one motif acts as a nucleophile that attacks the phosphorus atom of the substrate, forming a phosphoenzyme intermediate; then, a histidine from the other motif functions as a general acid in the cleavage of the phosphodiester bond20,22.

In addition to the two HKD motifs located well in the middle of the sequence, PLDs from eukaryotes typically encode an N-terminal phox homology/pleckstrin homology (PX/PH) or C2 domain23,24,25. Based on their N-terminal domains, PLDs can be grouped into PX/PH-PLDs and C2-PLDs. Although the various PLDs differ significantly in their N-terminal domains, these domains are closely related to the binding of membranes and lipids1,26,27. In addition, it has been reported that the C2 domain is required for the in vitro activity of C2-PLD, and its activity depends on the concentration of Ca2+28, whereas the activity of PX/PH-PLDs seems to be independent of its PX/PH domain29,4f).

To validate the residue interactions observed in the PldA–PA3488 structure, we generated several mutants and measured their binding affinity by surface plasmon resonance (SPR). The results suggested that PldAD643A and PA3488R50A dramatically decrease the PldA–PA3488 interaction, which is consistent with their functions in stabilizing the PldA–PA3488 complex through forming a tight salt bridge (Fig. 4e, Table 1 and Supplementary Fig. 10). In addition, the PldAR969A and PA3488D273A mutants severely disrupted the PldA–PA3488 interaction, indicating that these two residues also function as the major structural determinants in PldA–PA3488 recognition (Fig. 4c, f, Table 1 and Supplementary Fig. 10).

Table 1 Kinetics and affinity constants for wild-type and mutant PldA–PA3488 complex
Full size table

Conformational change of PldA

A comparison of the apo-PldA and PA3488-inhibited structures of PldA enables us to investigate the conformational changes that are associated with complex formation. Close inspection of the structures obtained in this study indicated that the conformational changes mainly occurred in the lid and the PD region (Fig. 5a–e).

In PldAFL, the lid region forms a short 310 helix (residues 120–122), packing with the β2-α2 loop and the α8 helix. In such case, the lid region is positioned distant from the entrance of the active-site pocket, which becomes accessible to the solvent so that the substrate can gain access (Fig. 5a–c). In contrast, in the inhibited PldA, on account of the binding of PA3488, the 310 helix melts, allowing the region to instead form an extended loop, stretching out and following the α3 helix direction and resulting in the formation of a shorter α3 helix than that of apo-PldA (Fig. 5a–c). This is associated with a translation of the lid toward the active site and a rotation of roughly 140° (based on Cα of Ser121), so that the backbone atoms in the lid region are as much as 4.8 Å away from their counterparts in PldAFL (Fig. 5a–c). Conclusively, the rotation transforms the relative “phase” of the lid region in the two structures. On binding PA3488, the whole lid region slightly changes its position and orientation with respect to the active site, sitting on the top of it and almost entirely occluding the entrance of the solvent and substrate (Fig. 5b). Coincidently, activity assays showed that the PldA almost but not completely lost its enzyme activity upon PA3488 binding (Fig. 5g); hereafter, we refer to PA3488-inhibited PldA as “pre-closed” PldA.

The altered position and orientation of the lid region is directly coupled to the PD region. Of note, the α5-α8 loop of the PD1 domain in the inhibited PldA is structurally invisible in the electron microscopy map, implying that these helices are highly dynamic after PA3488 binding, and this accordingly imparts flexibility to the lid region. In PldAFL, the lid region stacks against the α8 helix, creating a stabilized 310 helix. In PA3488-bound PldA, the flexibility of α8 substantially relieves the steric hinderance, making space for the lid region to move toward the active site pocket and block it. The flexibility of α8 also influences the conformation of the following α9 helix, where the Leu426, Thr422, and Ser418 pack extensively with the Met515, Leu511, and Vla507 in the α11 helix of PldAFL. On binding PA3488, the C-terminal side (residues 429-432) of α9 unravels into a loop and moves relative to the HKD1 domain, with concomitant displacement of the associated helices α11 and α12 (Fig. 5d). Meanwhile, the PD2 domain also moves slightly inside (Fig. 5e). In PldAFL, the α15–α19 helices in the PD2 domain point in the opposite direction relative to the HKD2 domain. In PA3488-bound PldA, however, a long β13-β14 loop in PA3488 is incorporated into α15-α19 through five hydrogen bonds (with residues 646 and 655–658) (Fig. 4f). To accommodate the loop of PA3488, a 310 helix (residues 656–658) of PldA unwinds, and its melting on binding PA3488 enables the adjacent helix α16 to move toward α18, tilting away from their positions in PldAFL. The interconnected α17 and following α19 are also coupled to the changes that α16 and α18 undergo (Fig. 5e). Consequently, the movements of the lid region and the PD1 and PD2 domains are correlated.

In PldAtruncate, the only partially visible loop of the lid represents the fully open state of the substrate-binding pocket (Fig. 5a, c), and the entrance of the substrate-binding pocket is shallower and more open than those of PldAFL and inhibited PldA (Fig. 5b). Furthermore, as the first door shield, the PD region also further unbolts, accompanied by the disassembly of the PD1 and PD2 domains. This is consistent with the movement of the indicator α11 shifting 120° toward the outside (Figs. 3a, c and 5a–c).

In summary, these structures provide insights into the mechanisms by which PA3488 inhibits the toxicity of PldA by sequential conformational changes.

Discussion

P. aeruginosa, a causative agent of hospital-acquired infections, secretes a eukaryotic-like protein, phospholipase D (PLD), known as PldA, which is responsible for its virulence2,31,38. PldA has extensive homology with the PLDs of eukaryotes, but not with those of prokaryotes12. Given its comparatively larger molecular weight of 122 kDa and its inherently disordered region, it has been challenging to perform a structural characterization of PldA. Here, we have determined, for the first time, the crystal structure of the full-length eukaryotic-like PldA (PldAFL) in the apo state. The high-resolution structure that we present here allows illustration of the similarities and differences in structure and function between PldAFL and PLDs from different species (bacterial PLD, human PLD, and plant PLDα). Indeed, conserved structural features include the canonical “horse saddle” topology, the two HKD domains, and the active site in which residues involved in catalysis are located. Mutagenesis studies confirmed again the importance of these critical catalytic residues (Fig. 1e). These similarities support the conservation of the mechanisms of substrate binding and catalysis in both bacterial and eukaryotic PLDs.

PldA also exhibits some structural features that are distinct from other PLDs. First, a PD region, which does not exist in other PLDs is situated adjacent to the catalytic domain, possibly imparting unique characteristics to the active site in PldA. This arrangement of helices, combined with the lining of the hydrophobic residues inside the helices, makes it easy to hypothesize that these helices may play an important role in regulating enzymatic activity. Human PLDs possess tandem phox homology (PX) and pleckstrin homology (PH) domains, which are known to mediate interactions with lipid membranes and are believed to regulate PLD localization within the cell. In view of the significant homology between PldA and hPLDs, it is tempting to propose that the PD region may have functional links with the PX–PH domain of hPLDs. Defining the nature of this region involved in binding of the substrate and lipid membrane, as well as characterizing the physiological relevance with the PX–PH domain of hPLDs, may be critical for understanding the mechanism of infection of PldA.

In addition, the core catalytic domain of PldAFL has much more in common with hPLDs than with plant PLDα (Supplementary Fig. 2b). One main feature makes PldA and hPLDs different from their plant cognate: the “open” or “closed” state generated by the position and orientation of the lid region. Our structures of PldAFL and the existing hPLD structures show that they adopt an “open” conformation, with the potential lid region dra** away from the active site and allowing the accessibility of the active site to the solvent and the substrate. In plant PLDα, the reverse applies: in the native PLDα structure, the lid hinders the access of water and the substrate to the active site, while the binding of reaction product PA8 triggers the conformational change of the lid region to an open state, making the active site accessible to both solvent and substrate24. One pioneering crystallographic study on hPLD1 described the possibility of occluding the active site by auto-inhibition23. This study assumed that the two amino acids (Trp381 and Arg917) may inhibit the entrance of the substrate due to their position relative to the active site. By analogy with the either “open” or “closed” structure24 of plant PLDα and our structure of PldA, we suggest that these residues rarely contribute to the state of the catalytic pocket, as they adopt the same conformation in these structures. It is unknown whether PldA and hPLDs readily interconvert between “open” and “closed” conformations, as observed for plant PLDα. However, in this study, we have also determined the crystal structure of PldA using in situ proteolysis (named PldAtruncate) and the cryo-EM structure of PldA in complex with PA3488—an immunity protein that can suppress the virulence of PldA. Compared to the open conformation exhibited by PldAFL (pre-open state), PldAtruncate presents a highly open conformation (open state) and the PA3488-inhibited PldA reveals an almost closed conformation (pre-closed state). The transition between these conformations involves the realignment of the PD domain and lid regions. The concerted movements of these two regions eventually affect their positions and orientations relative to the active site and hence free or obstruct the entrance of the solvent and substrate. These observations indicate that PldA may undergo large conformational changes in response to substrate binding.

Based on the structures obtained in this study, we propose the following mechanism for the invasion of PldA into a target cell. Inside P. aeruginosa cells, the PD2 of PldA interacts with PA3488, and an unknown chaperone protein interacts with PD1 of PldA to stabilize and tightly close the PD region and active pocket together (closed state). Once receiving a physiological signal to secrete PldA to adjacent cells, PA3488 and an unknown protein will be released from PldA to open the PD door shield, and the active pocket gradually opens (pre-closed and pre-open states). Upon anchoring the cell membrane of the target cell, the PD domain and active pocket are fully open to degrade phospholipids efficiently (open state) (Fig. 6). It has been reported that the VgrG proteins are encoded adjacent to PldA and can directly interact with it31. Recently, effectors linked by VgrG have been reported to require effector-specific chaperones for stability and/or to favor their interaction with VgrG39. Therefore, it is possible that there are unknown chaperones that can work with PA3488 to “lock” PldA in the closed state. Indeed, we found the coexistence of VgrG4b (PA3486) and PA3488 can completely inactivate enzymatic activity of PldA (Fig. 5g), as leakage of this enzyme activity would cause cell death by self-toxicity.

Fig. 6: Mechanism of PldA-mediated invasion.
figure 6

Proposed model for the invasion of PldA into target cell. PA3488 and an unknown chaperone (probably VgrG4b) interact with PldA to tightly close the active site (closed state) inside their own cells; once receiving a physiological signal to secrete PldA to adjacent cells, PA3488 and an unknown chaperone is released from PldA to open the PD door shield and the active pocket gradually opens (pre-closed and pre-open states); upon anchoring the cell membrane of the target cell, the PD region and active pocket are fully open to degrade phospholipids efficiently.

Full size image

There is increasing evidence that PldA is associated with promoting chronic infection and targeting eukaryotic cells to produce cell invasion. Given its sequence and particularly its structural similarities, one can easily speculate that PldA is capable of mimicking certain functions of the host PLD; this may partly account for its mechanism of activity as a virulence factor. The crystal structure of PldAFL, PldAtruncate, and the cryo-EM structure of the PldA–PA3488 complex we present here thus provide a structural basis that sheds light on the activation and inhibition mechanisms of PldA. This should aid the future design of PLD-targeted inhibitors and drugs.

Methods

Protein expression and purification

P. aeruginosa PldAFL and PldAtruncate were expressed in E. coli Rosetta cells using a modified pGEX-6t vector with an N-terminal 6×His tag prior to the GST fusion tag that is removable by cleavage with tobacco etch virus (TEV) protease. The cells containing expression plasmid were induced at OD600 = 0.6 for 16 h with 0.1 mM isopropyl-β-D-1-thiogalactopyranoside at 16 °C, and the cells were harvested by centrifugation. The cells were then resuspended in lysis buffer (20 mM Tris pH 7.5, 500 mM NaCl, 5% Glycerol, and 1 mM phenylmethylsulfonyl fluoride) and lysed by sonication. Cell lysate was cleared by ultracentrifugation at 15,000 × g for 40 min, and the supernatant was incubated with nickel-nitrilotriacetic acid resin (Bio-Rad) at 4 °C for 30 min. After washing, the bound proteins were eluted by 300 mM imidazole, followed by hydrolysis with TEV protease to remove the 6× His-GST tag and reloaded with 20 mM imidazole. The protein sample was further purified by gel-filtration chromatography (Superdex 200, GE Healthcare) equilibrated with a buffer (20 mM Tris pH 7.5, 150 mM NaCl). The purified proteins were concentrated to 10 mg/mL and stored at −80 °C. P. aeruginosa PA3488 (20–376, without the N-terminal 19 signal peptide) was expressed as stated above, PA3488 and PldAFL were co-purified using the same protocols for the PldA. The purified PA3488-PldA complex were concentrated to 10 mg/mL and stored at −80 °C. Mutations were produced by PCR-based site-directed mutagenesis, and the designed mutated proteins were purified using the same strategy as described above.

Crystallization and structure determination

Crystallization screening of PldAFL and PldAtruncate was carried out at 20 °C using the sitting-drop vapor-diffusion technique. The best crystals of the native PldAFL were grown under the conditions of 0.2 M potassium chloride, 0.05 M magnesium chloride hexahydrate, 0.05 M Tris hydrochloride pH 7.5, and 10% polyethylene glycol 4000. The crystals of PldAtruncate were obtained by adding chymotrypsin to the crystallization mixture in a ratio of 1:100; in this way the enzyme acts on the PldAFL under crystallization conditions and allows the proteolytic fragment to form crystals in the same drop. The best crystals of the PldAtruncate were grown under the conditions of 0.2 M potassium phosphate dibasic pH 9 and 20% polyethylene glycol 3350. Datasets of PldAFL were collected at 100 K on the BL19U1 beamline of the Shanghai Synchrotron Radiation Facility (SSRF), and datasets of PldAtruncate were collected at 100 K on the BL17U beamline of the SSRF. A 2.0 Å native dataset of the PldAFL and a 3.0  Å native dataset of the PldAtruncate were processed with the XDS software package40.

Attempts have been made to solve the phase problem via soaking the crystals in various buffers containing heavy metals, such as Pt, Hg, and Au, and incorporating selenium in the form of selenomethionine, but these were unsuccessful. Ultimately, despite the low sequence homology of PldA with PLDα (23% identity), the PldAFL structure was determined through molecular replacement using PHASER from the PHENIX software package41. The structure of the A. thaliana PLDα (PDB code: 6KZ9) was used as the search model. The model was further refined using PHENIX and manual rebuilding in COOT42. The PldAtruncate structure was solved by molecular replacement using PHASER with PldAFL as the search model. PHENIX and COOT were used repeatedly for refinement and manual building. All structural figures were prepared using UCSF Chimera43, Chimera X44, GraphPad Prism 5 (https://www.graphpad.com/), and PyMOL (https://www.pymol.org). All the data collection and structure refinement statistics of PldAFL and PldAtruncate are summarized in Supplementary Table 1.

PldA activity assay

The assay was performed using the commercial Amplex Red Phospholipase D Assay Kit (Invitrogen, A12219) to analyze the enzyme activity of PldA in vitro. The assay was carried out at 37 °C for 10 min containing 0.2 µg of PldA, 125 µM di8:0 PC (dissolved in 20 mM Tris-HCl, pH 8.0, 150 mM NaCl) and the supplied regents of the kit. The fluorescence was excited at 530 nm and detected at 590 nm. For the assay of PldA inhibition by PA3488 and VgrG4bCt (aa 693–808), The molar ratio of the samples are: PldA: PA3488 = 1:1, PldA: VgrG4bCt = 1:1, and PldA: PA3488: VgrG4bCt = 1:1:1, repectively. Kinetic study of PldAFL and PldAtruncate activity were performed under standard assay conditions with various substrate concentrations, ranging from 1 to 200 μM di8:0 PC.

Molecular docking

Molecular docking was performed using the AutoDock 4 software package45. AutoDockTools was used to generate the PA8 molecule and produce the area for docking. Ten states for the PA8 docking were generated, and the model with the lowest binding energy was selected for analysis. The images of these structures were generated by PyMOL.

Molecular dynamics simulations

Molecular dynamics simulations of PldAFL were performed using the GROMACS package46 with the CHARMM36 force field. The protein was centered in a cube and dissolved with TIP3P waters. Sodium and chloride ions were added to neutralize the electric charge. The particle mesh Ewald method was used to compute the electrostatic interactions with a real-space cut-off distance of 1 nm, and the van der Waals interactions were set at the same cutoff value. The system was energy minimized and heated to 300 K before the production process. The trajectory production processes proceeded for 70 ns, and the root-mean-square deviations and fluctuations were calculated on the Cα atoms of each residue.

Cryo-EM sample preparation and single-particle dataset acquisition

4 µL of PldA-3488 complex was applied onto a plasma-cleaned R1.2/1.3 300-mesh Au holey-carbon grid (Quantifoil), blotted using a Vitrobot Mark IV (Thermo Fisher Scientific), and plunge frozen in liquid ethane (15 s wait time, 8 °C, 100% humidity, 3–6 s blot time, −1 blot force). The sample quality was screened using a Titan Krios electron microscope equipped with a Falcon III direct electron detector (Thermo Fisher Scientific). Samples with good quality were imaged on the same electron microscope at 300 kV, equipped with a GIF Quantum energy filter (the slit width was set to 15 eV) and a K3 Summit direct electron detector (Gatan) in the super-resolution counting mode. A total of 3732 movie stacks were collected using SerialEM47 at a magnification of 130,000. During data collection, the defocus value was set in the range from 1.5 to 2.0 µm. Each movie stack was dose-fractioned into 32 frames with a total dose of about 50 e− Å−2. More data collection details are given in Supplementary Table 2.

Image processing

All movie stacks were motion-corrected, binned two-fold (0.668-Å pixel size after binning) and dose-weighted using the MotionCor2 program48. Contrast transfer function (CTF) estimation was carried out using the CTFFIND4 program49. The following image processing steps were carried out in RELION 3.150, as shown in fig. S4. A total of 2,099,492 particles with diameters ranging from 10 to 14 nm were auto-picked by a blob picker, and these were randomized split into four subsets. Each subset was subjected to several rounds of 2D classification, and particles with ice contamination or low resolution were removed. A total of 12 good classes with 1000 particles per class were selected to generate an initial model without reference. All selected particles (1,158,720) were combined and subjected to a 100 iterations global angular searching 3D classification with only one class, which to make the initial angle information of each particle more accurate and more reliable. For each of the last three iterations of the 3D classification above, a 25 iterations local angular searching 3D classification was performed with 10 classes. Good classes from each local angular searching 3D classification were selected and merged. The duplicated particles were removed, resulting in 577,905 particles in total. After 3D auto-refinement without any symmetry imposed, these particles yielded a reconstruction at an overall resolution of 3.32 Å. After CTF refinement and post-processing, the map resolution was improved to 3.05 Å. All resolutions were estimated using two independently refined half maps50 with the gold-standard Fourier shell correlation (FSC) = 0.143 criteria51. Directional FSCs were calculated using the 3DFSC server52, and local resolutions were determined with RELION’s own implementation.

Model building, refinement, and validation

The 3.05 Å cryo-EM map was used for model building. The crystal structures of PldA and PA3488 (PDB:5XMG) were fitted into the cryo-EM map using UCSF Chimera43, and these were used as the guide to build the model. The model was checked and corrected manually using COOT42, and it was refined against the map using PHENIX41 with secondary structure restraints applied. For model cross-validation, the coordinates of the final model were randomly shifted by up to 0.5 Å and then refined in PHENIX against one of the unfiltered half-maps. The refined model was then tested against the other unfiltered half-map. The model statistics are shown in Supplementary Table 2.

SPR experiments

The interactions between PldA and PA3488 were explored using a Biacore 8 K (GE Healthcare) instrument against immobilized PldA on the sensors at 298 K. The PldA samples were diluted to 80 nM in 25 mM HEPES, 250 mM NaCl, 0.05% (v/v) Tween 20, pH 7.2, and were immobilized on the flow cells of a Biacore CM chip (GE Healthcare) to 12295, 12500, and 11925 resonance units. Gradient concentrations of PA3488, mutant protein PA3488R50A, PA3488D118A, PA3488D273A, and PA3488R50A/D118A/D273A (from 100 nM to 6.25 nM with two-fold dilution) were then flowed over the chip’s surface. After each cycle, the sensor surface was regenerated with Gly-HCl pH 1.7. The binding kinetics were analyzed using a 1:1 binding model with the BIAevaluation software package (GE Healthcare). In turn, the binding of PldA and its mutants—PldAD643A, PldAR820A, PldAR969A, and PldAD643A+R820A+R969A—with PA3488 were also measured when immobilized by PA3488 on the chip following the same method as above.

μ-XRF data acquisition

For synchrotron micro X-ray fluorescence (μ-XRF) analysis, 100 μL purified PldA protein (15 mg/mL, in buffer containing 20 mM Tris-HCl, pH 8.0 and 150 mM NaCl) was mounted onto Kapton tape. 100 μL 0.1 mM CaCl2 was also analyzed as a reference. The distribution of Ca was measured at the 4W1B beamline in Bei**g Synchrotron Radiation Facility (BSRF, China). The electron energy in the storage ring was 2.5 GeV with a current ranging from 200 to 300 mA. The incident beam was focused with a size of 50 μm × 50 μm. A monochromatic X-ray with a photon energy of 15 keV was used to excite the samples and the count time was 10 s per pixel. PyMca software and origin 8.0 were used to process the data and plot the elemental distribution of PldA protein, respectively.

MALDI-TOF MS sample preparation and data processing

The crystals of PldAtruncate were collected, washed by crystallization buffer and dissolved in 6 μl H2O. The sample was then visualized by 12% SDS-PAGE gel. The protein bands were excised from the SDS-PAGE gel. After dehydration, the gel plugs were incubated in 25 mM NH4HCO3 with 5 mM DTT for 45 min. Then, the samples were alkylated with 40 mM iodoacetamide in 25 mM NH4HCO3 for 45 min at room temperature in the dark and digested overnight with trypsin (40 ng for each band) at 37 °C. The reactions were terminated by adding trifluoroacetic acid to a final concentration of 1%, and desalted using C18 Zip-Tip microcolumns (Millipore, Germany). The samples were then loaded onto the instrument in a crystalline matrix of α-cyano-4-hydroxycinnamic acid (CHCA, 5 mg mL−1).

MALDI-TOF/TOF-MS analysis was performed on an UltrafleXtreme mass spectrometer controlled by FlexControl 3.4 software package (Bruker Daltonics, Bremen, Germany). The instrument was externally calibrated using the Bruker peptides calibration kit. The spectra were acquired in the positive ion reflectron mode over the m/z range from 700 to 3500. For the MALDI-TOF/TOF-MS analysis, precursors were accelerated and selected in a time ion gate, after which fragments arising from metastable decay were further accelerated in the LIFT cell and detected after passing the ion reflector. The MALDI-TOF/TOF MS spectrum was subjected to a database search using the Mascot v2.5 (Matrix Science, London, UK) database search engine (www.matrixscience.com). The search parameters were as follows: enzyme: trypsin, allow up to 2 missed cleavages, fixed modifications: Carbamidomethyl (C) and variable modifications: Oxidation (M). The sequence of PldA was used for search database.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.