Disulfide loop cleavage of Legionella pneumophila PlaA boosts lysophospholipase A activity

L. pneumophila, an important facultative intracellular bacterium, infects the human lung and environmental protozoa. At least fifteen phospholipases A (PLA) are encoded in its genome. Three of which, namely PlaA, PlaC, and PlaD, belong to the GDSL lipase family abundant in bacteria and higher plants. PlaA is a lysophospholipase A (LPLA) that destabilizes the phagosomal membrane in absence of a protective factor. PlaC shows PLA and glycerophospholipid: cholesterol acyltransferase (GCAT) activities which are activated by zinc metalloproteinase ProA via cleavage of a disulphide loop. In this work, we compared GDSL enzyme activities, their secretion, and activation of PlaA. We found that PlaA majorly contributed to LPLA, PlaC to PLA, and both substrate-dependently to GCAT activity. Western blotting revealed that PlaA and PlaC are type II-secreted and both processed by ProA. Interestingly, ProA steeply increased LPLA but diminished GCAT activity of PlaA. Deletion of 20 amino acids within a predicted disulfide loop of PlaA had the same effect. In summary, we propose a model by which ProA processes PlaA via disulfide loop cleavage leading to a steep increase in LPLA activity. Our results help to further characterize the L. pneumophila GDSL hydrolases, particularly PlaA, an enzyme acting in the Legionella-containing phagosome.


Results
PlaA majorly contributes to LPLA activity, PlaC to PLA activity, and both substrate-dependently to GCAT activity of L. pneumophila culture supernatant. To compare the contribution of the GDSL enzymes to different lipolytic activities in L. pneumophila, we analyzed single mutants and a triple mutant for fatty acid release and sterol ester formation after incubation with different lipid substrates ( Fig. 2A). Figure 2B shows amounts of free fatty acids (FFA) liberated by culture supernatants where phosphatidylglycerol (PG) and phosphatidylcholine (PC) hydrolysis reveals PLA, lysophosphatidylglycerol (LPG) and lysophosphatidylcholine (LPC) LPLA and monoacylglycerol (MPG) lipase activities. As expected, wild type bacteria showed highest levels of activities towards all substrates. Reductions in secreted activities were noted for the plaA, the plaC and the plaACD triple mutant but not for the plaD mutant. Specifically, the plaA mutant was strongly reduced in LPG, LPC, and to some extent in MPG hydrolysis confirming major LPLA activity of PlaA. The plaC mutant showed reductions in PG and PC hydrolysis indicating contribution of PlaC to PLA activity. The plaACD mutant revealed lowest activities representing the additive lack of PlaA and PlaC. Nevertheless, hydrolysis of PG, PC and LPG was observed for the plaACD mutant supernatant revealing the presence of further secreted PLA/LPLA(s) in addition to the GDSL enzymes. PlaD under these conditions had no contribution to the secreted lipolytic activities (Fig. 2B). We next asked whether the three enzymes might have a role in cell-associated activities. However, no decreased activities of the mutants were found suggesting that cell-associated activities were independent of PlaA, PlaC or PlaD (Fig. 2C).
Further, we analyzed GCAT activity of L. pneumophila culture supernatants. Here we confirmed the importance of PlaC for transfer of fatty acids from PC to cholesterol, represented by the lack of cholesterolpalmitate (CP) andpropionate (CPP) generation by the plaC and plaACD mutants (Fig. 2D). When, however, LPC was used as a donor of palmitic acid, the wild type did not develop CP at all, indicating that under those conditions palmitic acid transfer is not favored by the GDSL enzymes. Interestingly, only the plaA mutant formed CP, implying that in the absence of PlaA, another enzyme, such as PlaC, may use LPC (which here was not reacted by PlaA) as a donor for palmitic acid (Fig. 2E). Wild type and the single mutants produced CPP, albeit with different efficiencies. Among the single mutants, the plaA mutant displayed the greatest reduction in the generation of CPP confirming a previous observation that PlaA contributes significantly to the CPP production in L. pneumophila 30 . Neither CP nor CPP could be detected with the culture supernatant of the plaACD mutant, displaying exclusive contribution of the three GDSL enzymes to GCAT  . Displayed are cleavage sites in phospholipids and monoglycerides as well as reaction products. R1 and R2 denote long chain fatty acid residues and X the polar headgroup. PLA and LPLA activities of culture supernatans (B) and cell lysates (C) from L. pneumophila Corby wild type and different knock out mutants (plaA − , plaC − , plaD − and plaACD − ) were determined via quantification of FFA after incubation for 16 h with the indicated lipids at 37 °C. Formation of different cholesterol esters by L. pneumophila culture supernatants was studied after incubation for 16 h at 37 °C with PC (D) and LPC (E) followed by lipid extraction and TLC. The results represent the means and standard deviations of duplicate reactions (B,C). All results shown are representative for at least two additional experiments. CP, cholesterol palmitate; CB, cholesterol butyrate; CPP, cholesterol propionate; FFA, free fatty acids; C, cholesterol. activity (Fig. 2D,E). To summarize, PlaA and PlaC activities in L. pneumophila culture supernatant showed clear differences. PlaA contributed majorly to LPLA activity and acyl transfer from lysophospholipids to cholesterol whereas PlaC revealed PLA activity and acyl transfer from diacylphospholipids to cholesterol.
PlaA and PlaC are type II-secreted and zinc metalloproteinase ProA-processed. PlaA and PlaD were recombinantly expressed, purified, and polyclonal antibodies were raised as described earlier for PlaC 30 . Previously, secretion types of PlaA and PlaC have been deduced from the lack of enzymatic activities in the culture supernatants of defined secretion mutants 28,32 . The generation of specific antibodies provided us with a tool to directly detect the proteins in different bacterial fractions, and assess their secretion path and processing. To that end, we prepared culture supernatants and cell lysates from wild type, different secretion mutants and the proA mutant. After Western blot analysis, the pattern observed was quite similar for PlaA and PlaC. Both proteins were found in the supernatant of all strains but not in the lspDE mutant and the negative control (plaACD -), confirming that PlaA and PlaC export is type II secretion system-dependent (Fig. 3). While PlaA in the supernatants was mostly present as a ~26-28 kDa protein, PlaC had a size of about 36 kDa. The latter represents the processed and ProA-activated form of PlaC 30 . Consequently, a larger protein version of about 47 kDa representing unprocessed PlaC could be detected in the supernatant of the proA mutant (Fig. 3, Fig. 1). PlaA was also present in a higher molecular form of ~34 kDa in the proA mutant supernatant suggesting its processing by ProA (Fig. 3, Fig. 1). Interestingly, PlaA and PlaC were not detected in substantial amounts in cell lysates of the different strains unless their secretion was prevented by an lspDE knock out (Fig. 3). On the contrary, PlaD was only detected in the cell lysates, except in the plaACD mutant, but not in culture supernatants (Fig. 3). In summary, we here provide evidence that PlaA and PlaC are Lsp type II-secreted and ProA-processed whereas PlaD remains cell associated. rPlaA shows LPLA, lipase, and GCAT activities which depend on the individual members of the catalytic triade: serine 30, aspartate 279, and histidine 282. For further analysis of PlaA, the protein was recombinantly expressed in E. coli and either the purified protein or cell lysates were used for lipolysis assays. As shown in Fig. 4A, rPlaA possessed lipase and LPLA activities, the latter especially towards LPG and LPC. A site directed mutation, affecting serine 30 in the GDSL motif, as well as the other two potential members of the catalytic triade, aspartate 279 and histidine 282, reduced LPLA or lipase activities to background levels highlighting their importance in catalysis (Fig. 4A,B). GCAT activity of rPlaA was found for the transfer of short chain fatty acids (generation of CPP) and palmitic acid (generation of CP) with MPG and lysophospholipids as donors (Fig. 4C,D). The formation of CP was surprising because PlaA-dependent CP development was not found for L. pneumophila culture supernatants (Fig. 2D,E). As expected the catalytic site mutants did not exhibit GCAT activity (Fig. 4C,D). In conclusion, rPlaA shows LPLA, lipase, and GCAT activities dependent on a serine 30, aspartate 279, histidine 282 catalytic triade. Zinc metalloproteinase ProA boosts LPLA activity of rPlaA but diminishes GCAT activity transferring palmitic acid. PlaA is C-terminally processed by ProA. Next we were interested if the zinc metalloproteinase ProA of L. pneumophila has an impact on PlaA activity as it has been previously demonstrated for PlaC, where especially GCAT activity increases after ProA treatment 30 . To test for a ProA-dependent effect on PlaA activity we initially used culture supernatant of an L. pneumophila plaACD mutant containing ProA but lacking GDSL enzyme-derived activities in the lipid hydrolysis samples. Indeed, a steep increase in LPLA activity with LPG and LPC as substrates was detected when the supernatant was added to the reaction mixture with purified rPlaA. In contrast, no substantial activity was observed with the purified S30N-mutated rPlaA (Fig. 5A). Of note, a slight increase in PG hydrolysis was measured in the case of incubation of rPlaA and rPlaAS30N with the culture supernatant suggesting the presence of a non-GDSL PLA in the supernatant of the plaACD mutant (Fig. 5A).
Western Blot analysis indicated processing of PlaA by ProA (Fig. 3) and therefore, we analyzed whether rProA also acts on rPlaA. Indeed, rPlaA but not rPlaAS30N showed a substantial increase in LPLA activity when incubated with rProA (Fig. 5B). In contrast to the earlier described increase in PlaC activity upon ProA activation, no enhancement of GCAT activity was observed for rPlaA treated with rProA. Instead, rProA addition resulted in loss of palmitic acid transfer from LPC to cholesterol (i.e. no generation of CP) by rPlaA (Fig. 5C). This might explain why no explicit PlaA-derived GCAT activity transferring palmitic acid from lysophospholipids was observed in the L. pneumophila wild type in the experiments presented above (Fig. 2D,E). In contrast, transfer of short chain fatty acids was found independent of ProA (Fig. 5C).
To verify processing of PlaA by ProA, we incubated rPlaA with rProA followed by Western blotting with a PlaA-specific antibody or another antibody directed towards the N-terminal Strep tag. In both cases, rPlaA was found to be processed after 10 minutes incubation, characterized by the appearance of an ~26-28 kDa protein band (Fig. 5D,E), which corresponds to the size of PlaA in the above analyzed L. pneumophila strains, except the proA mutant (Fig. 3). Cleaved rPlaA could still be detected with the anti-Strep antibody suggesting that PlaA is processed on its C-terminal side (Fig. 5D,E). An incubation time extending 1 h resulted in stable detection of the ~26-28 kDa fragment with the PlaA-specific antibody. On the contrary, the Strep-PlaA version was no longer detectable under these conditions (Fig. 5D,E). These data suggest further processing of rPlaA at the N-terminus. However, no second substantial molecular shift of rPlaA could be observed with the PlaA antibody indicating that N-terminal cleavage of rPlaA is likely a minor procession (Fig. 5D,E). After 5 h of incubation of rPlaA without rProA further degradation of the protein from the C-terminal side occurred (Fig. 5D,E). As a conclusion, we showed that PlaA LPLA activity is steeply increased but GCAT activity is diminished by ProA which processes PlaA at its C-terminus.
Deletion of 20 amino acids in a predicted disulphide loop of rPlaA results in boosted LPLA activity but diminished GCAT activity transferring palmitic acid. L. pneumophila PlaC GCAT is activated through truncation of a disulphide loop by ProA 30 . Like PlaC, PlaA contains a predicted disulphide loop, which in this case consists of 28 amino acids (Fig. 1). Indeed, the ~26-28 kDa PlaA fragment observed in wild type L. pneumophila culture supernatant and after treatment of rPlaA with rProA nicely matches a protein, which is cleaved within the disulphide loop region (Figs 1, 3 and 5D,E). After reducing SDS-PAGE, the disulphide bond would be opened and the cleaved C-terminal part would be released corresponding to a ~ 28 kDa fragment visible after Western blotting. A truncated disulphide loop in PlaC has been shown to render the protein more active, especially with respect to GCAT activity independently of ProA 30 .
Therefore, we deleted 20 amino acids from the disulphide loop of PlaA and analyzed lipolytic activities of purified rPlaA versus rPlaAdel248-267. Indeed, lipase and LPLA activities of the loop deletion protein increased steeply compared to full length rPlaA (Fig. 6A). When culture supernatant of the L. pneumophila plaACD mutant or rProA was added to the preparations, the degree of activation was not further enhanced indicating that in the absence of the disulphide loop region, PlaA LPLA activity is independent from ProA-based activation (Fig. 6B,C). In accordance with the results above and in contrast to steep activation of LPLA activity, GCAT activity was not detectable for the loop truncation protein, showing that GCAT activity of PlaA is diminished by loop deletion (Fig. 6D). For the loop deletion protein no substantial processing after ProA addition, as it was observed for rPlaA ( Fig. 5D,E), was detected with the PlaA or Strep tag antibodies. However, a limited N-terminal processing releasing the N-terminal Strep-Tag was observed (Fig. 6E,F). This means that deletion of the disulphide loop leads to comparable activities as seen for full length PlaA processed by ProA. We therefore localized the ProA cleavage site within PlaA to the amino acid region between aspartate 248 and leucine 267.
PlaA is cleaved by ProA between glutamate 266 and leucine 267. Mass spectrometry analysis of rPlaA incubated with rProA indeed confirmed that PlaA is cleaved by ProA between cysteine 243 and cysteine 272. The peptides identified indicated that all potential cleavage sites are within the disulphide loop region. The most probable ProA-cleavage site within PlaA was in front of leucine 267 because the respective peptide displayed the highest intensity (Fig. 7, Table. S1). Mutagenesis of glutamate 266 and leucine 267 to asparagine was not sufficient to prevent the cleavage of PlaA by ProA (Fig. S4). This indicates that also less preferential sites may be cleaved by ProA which then also results in protein activation.

Discussion
We previously described that a GDSL enzyme of L. pneumophila, PlaC, is activated by a ProA-dependent disulphide loop cleavage. After cleavage, the two fragments of the protein, each containing catalytic essential amino acids, are kept together by a disulphide bond (Fig. 1). PlaC cleavage by ProA predominantly activates GCAT and also PLA activity of the enzyme. Mutants of PlaC where the disulphide loop was shortened from 44 to 31, 24, 22, or 6 amino acids show ProA-independent activation of GCAT, suggesting a loop-mediated enzyme inhibition/ activation mechanism. This ProA-mediated regulation might serve to protect the bacterium from a potential harmful activity of PlaC within the cell 30 .
A variety of enzymes with distinct activities using similar but not identical activation strategies have been described before. For example, Pseudomonas exotoxin A is a ADP-ribosylase, which is activated by proteolytic cleavage within a disulphide loop and subsequent reduction of the disulphide bridge, thereby releasing the active domain 39,40 . Another example is Shiga toxin of enterohemorrhagic E. coli, where the C-terminal region of subunit A contains a disulphide loop with the trypsin and furin recognition sequence R-V-A-R. Cleavage results in enzymatically active N-glycosidase and disulphide bond linked A1 and A2 fragments 41,42 .
In the case of PlaA, a ProA-dependent activation via disulphide loop processing was found and we propose the following model of PlaA activation. PlaA and ProA are both present as preproenzymes within the cytoplasm. Both are transported through the inner bacterial membrane via the Sec system and concomitant signal peptide cleavage. In the periplasm, a disulphide bond may be formed within PlaA (internal preform). Cell-associated quantity and activity of PlaA are very low (Figs 3 and 2C) compared to those in the external space suggesting fast export of the protein. After subsequent export of PlaA and ProA via the type II secretion system Lsp, ProA acquires its mature form via autoactivation by means of N-terminal propeptide cleavage 43 . PlaA (external preform) is then processed by ProA resulting in the external mature form of PlaA with increased LPLA activity and lost GCAT activity for transfer of palmitic acid (Fig. 8).
LPLA activity steeply increased when PlaA was processed by ProA or alternatively when the disulphide loop was shortened from 28 to 8 amino acids. We previously described that the loop region of PlaC shared substantial homology to PlaC-like enzymes from Legionella but not to the corresponding PlaA-like enzymes 30 . Similarly, the loop region in PlaA is only conserved in PlaA-like enzymes of L. pneumophila. Mass spectrometry analysis revealed that PlaA is cleaved by ProA within the loop region, but mutagenesis of the most probable cleavage site was not sufficient to prevent the digestion by ProA. The same was observed for PlaC, where several ProA-cleavage sites were found within the disulphide loop region, but mutagenesis studies of potential target sites indicated that the protease either is not very selective or that structural features are important 30 .
Although the proteolytic processing is very similar between PlaC and PlaA they are differentially regulated by ProA. Unprocessed PlaA possesses LPLA activity and has specific GCAT activity, only accepting lysophospholipids as fatty acid donors. Unprocessed PlaC has PLA activity and corresponding GCAT activity, which transfers palmitic acid and short chain fatty acids from diacylphospholipids to cholesterol. While PLA and GCAT activity of PlaC are both enhanced by proteolytic processing, PlaA processing leads to enhanced LPLA activity but diminished GCAT activity. We conclude that external proPlaA may predominantly act as GCAT at earlier time points of growth prior to ProA secretion. However, at later time points after ProA-mediated processing, PlaA shifts its activity to a very potent LPLA (Figure 8). Activated PlaA may then contribute to phagosomal membrane disintegration as described before when the protective type IVB secreted Dot/Icm effector SdhA is not present 37 .
Most interestingly, the type IVB secreted effector SdhA counteracts LCV disintegration by type II-secreted PlaA 37 . The mechanism by which this is achieved and whether the interaction is direct or indirect is not known so far. Nevertheless, keeping the vacuole membrane intact is a very important process to prevent immune detection of L. pneumophila 37,44 . It is possible that PlaA acts from the inside of the phagosome whereas the Dot/Icm injected SdhA rather acts from the outside. One can hypothesize that PlaA, although not able to drive bacterial exit from the LCV in the presence of SdhA, prepares the membrane for facilitated exit at a later time point under the influence of additional factors. Further, PlaA might help to modify the LCV membrane in a way that it is more prone to fuse with other lipid compartments or may facilitate expansion of the phagosomal membrane when the bacteria are replicating.
But the question remains: How does an LPLA activity, directed towards lysophospholipids but not diacylphospholipids, disintegrate the LCV membrane? The type II-secreted enzyme ProA was previously found in the LCV lumen 36 ; therefore it is possible that other type II secreted enzymes, such as PlaA or PlaC, localize to the same compartment. Provided that the LCV membrane resembles the membrane of the endoplasmic reticulum or the plasma membrane, its major lipid components would be PC, phosphatidylethanolamine, phosphatidylserine, and phosphatidylinositol 45 . In this case an LPLA alone should not substantially disintegrate the membrane of the compartment. Nevertheless, several conditions may facilitate such a process: specifically when 1) in addition to ProA another yet unknown activating factor switches LPLA to PLA activity, 2) the LCV contains a substantial amount of lysophospholipids or another yet not identified substrate of PlaA, and 3) PlaA acts together with another enzyme showing PLA activity. The latter point seems most plausible. PlaC or another currently not identified PLA could potentially catalyze the first step resulting in the release of a fatty acid from a diacylphospholipid thereby producing the lysophospholipid substrate for PlaA. PlaC was found in L. pneumophila OMVs but not PlaA 34 . If a twostep hydrolysis of the LCV membrane is conceivable by PlaC and PlaA, fusion of PlaC-containing OMVs might strongly promote the disintegration process 34 . PlaA may then take over the second step and hydrolyze the lysophospholipid and thereby may accelerate the process substantially. Furthermore, enhanced LPLA activity could be useful for the elimination of the signal transducer LPC 46 . LPC is known as a factor for enhancement of neutrophil bactericidal activity and can lead to reduced bacterial viability 47,48 . Lower LPC levels as a result of increased LPLA activity might therefore support bacterial survival and integrity.

Methods
Bacterial strains and growth conditions. L. pneumophila sg1 strain Corby was used for all experiments, including knock out mutant construction, as well as for recombinant gene expression in E. coli strains DH5α and BL21. The used strains are listed in Table 1 and were grown on buffered charcoal yeast extract (BCYE) agar and in buffered yeast extract (BYE) broth (both for Legionella), or on Luria Bertani (LB) agar and in LB broth (both for E. coli) as described previously 28,49,50 . DNA techniques and sequence analysis. E. coli BL21 was used for the propagation of recombinant plasmid DNA with backbones of the following vectors: pGP172 51 and pet28a(+) (Novagen). Genomic and plasmid DNA were prepared, amplified, and sequenced according to standard protocols. Primers were obtained from Eurofins MWG Operon. Restriction enzymes were purchased from New England Biolabs. Foreign DNA was introduced into E. coli strains by electroporation with a Life Technologies cell porator according to the manufacturer's specifications as described earlier 28 . Nucleotide and translated protein sequences were analyzed using the Geneious 10.0.05 52 , the NCBI website (http://www.ncbi.nlm.nih.gov/), DiANNA 1.1 (http://clavius.bc.edu/~clotelab/DiANNA/), SignalP 4.1 (http://www.cbs.dtu.dk/services/SignalP/) and ExPASy (http://www.expasy.ch).

Cloning of L. pneumophila plaA and plaD into pGP172 for recombinant expression in E. coli. L.
pneumophila Corby plaA (lpc1811) and plaD (lpc 0558) were cloned into the pGP172 vector resulting in pCL119 and pSB2 (N-term. Strep-Tag). To create PlaA disulphide loop deletion mutant constructs, pCL119 was amplified with primers listed in Table 2 and ligated; resulting in pCL126. To introduce mutations into the plaA gene, the pCL119 vector was mutated by means of the QuikChange Site-Directed Mutagenesis Kit (Stratagene) with the primers listed in Table 2, resulting in pCL121-pCL123, pCL127 and pMH41.
L. pneumophila GDSL enzyme knock out mutant construction. Plasmids pER11 and pCL13, containing the plaC and plaD genes, disrupted or partially replaced by a Gent r and Hyg r cassette, respectively, were used to introduce Gent r and Hyg r gene insertion mutation into the chromosome of strain L. pneumophila Corby plaA-mutant 30 via allelic exchange. To generate pER11, L. pneumophila 130b plaC was cloned into the pGEM-Tez vector yielding in pBH1 28 . After pBH1 restriction by means of HindIII and BamHI, the Gent r -gene cassette was cloned into the pBH1 vector yielding in pER11 (primers see Table 2). To generate pCL13, L. pneumophila Corby plaD was cloned into the pBCKS vector yielding in pCL7 (primers see Table 2). After the digestion of pCL7 with HpaI, the Hyg r -gene cassette was cloned into the vector yielding in pCL13 (primers see Table 2). Knockout mutants were generated by natural transformation and homologous recombination 29,53 . PCR was used to examine Gent r and Hyg r Legionella for the respective mutation.

Preparations of culture supernatants and cell lysates. For assessment of hydrolytic activities, bacteria
were harvested at the end of exponential growth (if not stated otherwise at an OD 600 of about 4). Culture supernatants were obtained by centrifugation for 5 min at 5000 g and in some instances were subsequently concentrated by ultra-filtration with an exclusion size of 10 kDa. Cell lysates were produced as described previously 29,54 . Culture supernatants and cell lysates were tested immediately for enzymatic activities.
Purification of recombinant ProA. Recombinant periplasmic ProA was isolated by osmotic shock 55 from E. coli BL21(pCL15) and subsequent anion exchange chromatography as described before 30 . For GCAT activity the reaction mix was doubled and cholesterol was added. All lipids, including standards for thin layer chromatography (TLC) were obtained from Sigma Chemicals or Avanti Polar Lipids. Detection of reaction products, such as free fatty acids and sterol esters, was performed by lipid extraction and subsequent TLC or by quantitative detection of free fatty acids as described earlier 28,29 .  Western Blot analysis. Samples were separated on a 12.5% SDS-polyacrylamide gel 57 . Proteins were transferred to a nitrocellulose membrane. Primary polyclonal rabbit antibodies against PlaA, PlaD (generated by BioGenes against recombinant expressed and purified PlaA and PlaD) and PlaC 30 were added at a dilution of 1:1000. α-rabbit linked to G-horseradish peroxidase (Sigma) was used as secondary antibody diluted 1:10,000. For detection of the Strep-Tag the monoclonal mouse StrepMAB-Classic antibody (IBA) was used at a dilution of 1:5000. Proteins were detected by chemiluminescence (ECL Kit; Amersham). Data analysis. The mass spectra were analyzed using MaxQuant (Version 1.5.1.2) 59 . At first, parent ion masses were recalibrated using the 'software lock mass' option 60 before the MS² spectra were searched using the Andromeda algorithm against the ProA and PlaA sequences as well as 245 entries of the cRAP database (http:// www.thegpm.org/crap/). Spectra were searched with a tolerance of 4.5 ppm in MS1 and 20 ppm in HCD MS² mode, using semi-tryptic specificity (KR not P) and allowing up to one missed cleavage site. Cysteine carbamidomethylation was set as a fixed modification and methionine oxidation as a variable modification. The false discovery rate was set to 1% for peptide and 1% protein identifications. Identifications were transferred between samples using the 'match between run' option within a match window of 0.  Data Availability. All data generated or analysed during this study are included in this published article or in the supplementary information (Figs S1-4 and Table S1).