Broad-Spectrum Regulation of Nonreceptor Tyrosine Kinases by the Bacterial ADP-Ribosyltransferase EspJ

ABSTRACT Tyrosine phosphorylation is key for signal transduction from exogenous stimuli, including the defense against pathogens. Conversely, pathogens can subvert protein phosphorylation to control host immune responses and facilitate invasion and dissemination. The bacterial effectors EspJ and SeoC are injected into host cells through a type III secretion system by enteropathogenic and enterohemorrhagic Escherichia coli (EPEC and EHEC, respectively), Citrobacter rodentium, and Salmonella enterica, where they inhibit Src kinase by coupled amidation and ADP-ribosylation. C. rodentium, which is used to model EPEC and EHEC infections in humans, is a mouse pathogen triggering colonic crypt hyperplasia (CCH) and colitis. Enumeration of bacterial shedding and CCH confirmed that EspJ affects neither tolerance nor resistance to infection. However, comparison of the proteomes of intestinal epithelial cells isolated from mice infected with wild-type C. rodentium or C. rodentium encoding catalytically inactive EspJ revealed that EspJ-induced ADP-ribosylation regulates multiple nonreceptor tyrosine kinases in vivo. Investigation of the substrate repertoire of EspJ revealed that in HeLa and A549 cells, Src and Csk were significantly targeted; in polarized Caco2 cells, EspJ targeted Src and Csk and the Src family kinase (SFK) Yes1, while in differentiated Thp1 cells, EspJ modified Csk, the SFKs Hck and Lyn, the Tec family kinases Tec and Btk, and the adapter tyrosine kinase Syk. Furthermore, Abl (HeLa and Caco2) and Lyn (Caco2) were enriched specifically in the EspJ-containing samples. Biochemical assays revealed that EspJ, the only bacterial ADP-ribosyltransferase that targets mammalian kinases, controls immune responses and the Src/Csk signaling axis.


RESULTS
EspJ regulates the host immune response in vivo. Previous studies of EspJ were performed mainly by infecting cultured cells for a few hours (12,13). However, in vivo, A/E pathogens primarily interact with intestinal epithelial cells (IECs) over several days or weeks. Indeed, by quantifying proteomic responses of IECs isolated from mice 8 days postinfection with wild-type (WT) C. rodentium, we have recently shown that it affects cellular bioenergetics and cholesterol homeostasis (14).
We first analyzed the infection dynamics after oral gavage of mice with WT or ΔespJ or ΔART (ΔespJ mutant complemented on the genome with espJ R79A catalytic mutation) mutant C. rodentium. Each strain showed similar colonization dynamics (Fig. 1a) and induced comparable levels of CCH (Fig. 1b to e), as measured by colonic crypt length and Ki-67 staining (a marker for cell proliferation). To investigate the role of EspJ in vivo, EspJ ADP-Ribosylates Nonreceptor Tyrosine Kinases ® we then performed proteomics analysis of IECs purified from mice at the peak of shedding, 8 days postchallenge with WT or ΔART mutant C. rodentium. We used isobaric peptide labeling (tandem mass tag [TMT]) and MS3 quantification, as recently reported (14), to determine protein abundance changes in IECs isolated from mock-infected mice or mice infected with WT or ΔART C. rodentium (four, five, and six mice per group, respectively). We identified 7,400 mouse proteins and quantified them by shotgun proteomics analysis of isobarically labeled IEC proteins (false-discovery rate [FDR]-corrected P value, Ͻ0.01). Compared to the protein abundances in uninfected IECs, those of 3,393 protein abundances were changed by Ϯ1.5-fold upon WT infection and those of 3,164 were changed upon ΔART mutant infection, with 2,650 of these changes observed in both infections (see Fig. S1a in the supplemental material). We performed KEGG pathway gene ontology (GO) term analysis of the proteins with Ϯ1.5-fold abundance changes upon WT or ΔART infection. The pathways regulated by WT and ΔART infections were very similar and reflected observations in our previous report (14) (Fig. S1b and c). In comparison to WT infection, 457 proteins were increased Ͼ1.5-fold in the absence of active EspJ (ΔART infection), suggesting that these proteins are induced during WT infection but directly or indirectly suppressed by EspJ catalysis (Fig. 2a; Table S3).
ClueGO biological process analysis of these 457 proteins suggests that EspJ has immunoregulatory roles, modulating proteins linked to the defense response to other organisms, regulation of immune cell migration and adhesion, cytokine responses, phagocytosis, and manipulation of the actin cytoskeleton (Fig. 2b). ClueGO cell component analysis revealed that these proteins are localized in and around the cell periphery; vesicles/phagosome, membrane, cytoskeleton/cortex, cell adhesions, and extracellular matrix (Fig. 2c). This is consistent with the fact that Src, the known EspJ target, is membrane tethered and has significant roles in the innate immune response and the actin cytoskeleton (15). However, considering that EspJ ADP-ribosylates a universally conserved kinase domain residue (Src E310) (12) and its broad effect during infection (Fig. 2), we hypothesized that its catalytic activity may impact multiple kinases.
EspJ regulates numerous tyrosine kinases in vivo. We used the kinase enrichment analysis (KEA) function within the Expression2Kinases (X2K) software to predict upstream regulatory kinases from the group of 457 proteins repressed by EspJ (16). KEA draws from kinase perturbation studies within numerous protein and gene expression databases to enable the prediction of upstream regulatory kinases responsible for changes in protein abundances. Twelve of the 15 most probable kinases regulated by EspJ were tyrosine kinases; 10 of these are NRTKs (Fig. 2d), of which 7 were present in the IEC proteome. Importantly, none of these kinases were significantly predicted from 10 randomly generated sets of 457 proteins ( Fig. 2d; Table S4). When ΔART mutantinfected and WT-infected IECs were compared, the abundance of Hck increased by Ͼ2-fold, but it was the only kinase from this group to change by Ͼ1.5-fold (Table 1). In conclusion, the in vivo proteomics data are consistent with the hypothesis that EspJ regulates multiple tyrosine kinases, including members of the Src, Fak, Jak, and Syk NRTK subfamilies ( Fig. 2d; Table S4).
EspJ can ADP-ribosylate a suite of NRTKs. To test the hypothesis that EspJ regulates the activity of multiple kinases, we investigated its substrate repertoire in numerous cell lysates. Recombinant EspJ EHEC and 2-ethynyl-adenosine-NAD (eNAD) were incubated with cell lysates before the conjugation of biotin onto resulting ADP-ribosylated proteins (17) and their enrichment with NeutrAvidin resin. Proteins were quantified by label-free quantitation (LFQ) mass spectrometry (MS) (PRIDE project PXD008533). Proteins enriched by the presence of EspJ were identified by using a two-sided t test and with an FDR-corrected P value of Ͻ0.01 and a fudge factor (s 0 ) value of 1.
In HeLa (human cervical epithelial) and A549 (human alveolar epithelial) cell lysates, Src and Csk were significantly enriched ( Fig. 3a and b). In polarized Caco2 (human epithelial colorectal) cell lysate, Src and Csk were again enriched, along with the SFK Yes1 (Fig. 3c). In differentiated Thp1 cells (human monocytes), lysate EspJ modified the largest range of targets, i.e., Csk, the SFKs Hck and Lyn, the Tec family kinases Tec and Btk, and the adapter tyrosine kinase Syk. Failure to enrich any of these kinases after incubation with the inactive maltose binding protein (MBP)-EspJ D187A showed that their modification depends upon a functional EspJ ART domain (Fig. 4). Closer analysis reveals that Abl (HeLa and Caco2 lysates) and Lyn (Caco2 lysate) are enriched specifically in the EspJ-containing samples but below the significance cut-off after imputation for missing values ( Fig. S2 and Table S5). This shows that EspJ can ADP-ribosylate a range of NRTKs from multiple cell lineages but suggests specificity within the tyrosine ). (a) A ratio of ΔART mutant to WT protein abundances was created and ranked according to those that increased the most in ΔART mutant-versus WT-infected IECs (left). The enlarged image shows 457 proteins with abundance increases of at least 1.5-fold in ΔART mutant compared to WT infections; which were assigned to 10 GO term groups for biological process (b) and cellular components (c). Analyses are shown with the number of GO term-associated proteins in grey bars and the negative log P value (with Bonferroni correction) indicated by black crosses. Abbreviations: (Ϫ) Reg., negative regulation; Inorg./ox. response, response to inorganic substance/oxidative stress; Exc., extracellular; FA, focal adhesion. (d) The 15 most likely upstream regulatory kinases targeted by EspJ to produce the observed changes in protein abundance. Grey bars show the kinase P value for the group of EspJ-modulated proteins, and black bars show the mean P value of the kinase across X2K runs from 10 random lists of 457 proteins from the total of 7,400. Check marks (✓) highlight the tyrosine kinases within this prediction and whether the kinase was identified in the IEC proteome. See also Fig. S1 and S2.
EspJ ADP-Ribosylates Nonreceptor Tyrosine Kinases ® kinases due to the absence of other key NRTKs. Further, despite the proteomic identification of RTKs and serine/threonine kinases in these cell lines, none were confidently enriched (Table S5). The SH2 domain is common to the Src, Tec, Syk, and Csk subfamilies but is not found in the FAK and JAK subfamilies. To test whether SH2 plays a role in kinase recognition, green fluorescent protein (GFP)-tagged Src WT and SH2 domain R175K mutant Src unable to bind phosphotyrosines (Src RK ) (18) were immunoprecipitated from HeLa cells before incubation with MBP-EspJ EHEC and NADbiotin. EspJ was able to ADP-ribosylate the Src SH2 domain mutant at an efficiency similar to that of WT Src, suggesting that EspJ may select its targets via alternative mechanisms ( Fig. 5a and c).
EspJ shows specificity within its NRTK targets. The proteomics assay highlighted ubiquitous SFKs Src and Yes1, but not Fyn, as EspJ targets. To test this specificity, we assayed the ADP-ribosylation of a panel of immunoprecipitated, GFP-tagged kinases incubated with EspJ and NAD-biotin as described above. EspJ could ADP-ribosylate all of the kinases tested (Src, Yes1, Fyn, Csk, and Abl) (Fig. 5b). Semiquantitative densitometry was used to infer the ADP-ribosylation efficiency from the ratio of ADPribosylated protein (anti-biotin antibody Western blot assay) to total protein (anti-GFP TABLE 1 Kinases identified in IEC proteomes a a Shown are the kinases identified in the proteomes of uninfected IECs and IECs infected with WT or ΔART mutant C. rodentium, their respective log 2 abundances, and the fold differences between ΔART mutant and WT infection conditions. In addition, whether they were predicted to be regulated by EspJ in vivo by KEA or ADP-ribosylated by EspJ in cell lysate is noted. Protein log 2 abundances range from blue (low) to white (medium) to red (high). UI, uninfected.

FIG 3
In vitro chemical proteomics analysis suggests that EspJ can ADP-ribosylate multiple tyrosine kinases. Lysates from HeLa (a), A549 (b), polarized Caco2 (c), and differentiated Thp1 (d) cells were incubated with EspJ EHEC and eNAD. ADP-ribosylated proteins were tagged with biotin by click chemistry before enrichment with NeutrAvidin resin and analysis by LFQ MS. Red dots represent proteins that were significantly enriched when abundances in the presence versus the absence of EspJ were compared. Data sets were subjected to a two-sided t test, and black lines display the significance cutoff with an FDR-corrected P value of Ͻ0.01 and an s 0 value of 1. See also Fig. S2.
EspJ ADP-Ribosylates Nonreceptor Tyrosine Kinases ® antibody Western blot assay) across three to five experiments. This suggested that the SFK Fyn was modified at less than half the efficiency of its family members Src and Yes1 and the non-SFK Abl kinase (Fig. 5C). Interestingly, while Csk was the only kinase to be ADP-ribosylated by EspJ in all four cell lysates (Fig. 3), it was the weakest GFP-tagged substrate in this assay ( Fig. 5b and c).
EspJ inhibits Csk by ADP-ribosylation. As EspJ can target both the SFKs and their inhibitor Csk, we further investigated the EspJ-Csk relationship to understand the interplay among EspJ, SFKs, and Csk. First, as we have previously shown that the EspJ homologues SeoC and SboC can ADP-ribosylate Src, we confirmed their ability to ADP-ribosylate Csk (Fig. 6a). Moreover, while the catalytic mutant (D187A) EspJ EHEC protein was unable to ADP-ribosylate Csk by using biotin-NAD, WT EspJ EHEC modified Csk in a time-dependent manner (Fig. 6b). Finally, mutation of Csk E236, which is analogous to Src E310, prevented ADP-ribosylation, indicating that EspJ EHEC targets the same residue in Src and Csk (Fig. 6c).
As the differential effect of Csk activation or inhibition could have drastic impacts on the regulation of Src kinase, we next tested the effect of Csk ADP-ribosylation by EspJ. ADP-ribosylation of Csk with native NAD was performed for 2 h prior to the addition of kinase buffer and either the synthetic tyrosine kinase substrate poly(Glu 4 , Tyr) or Src SH1-K295M/Y416A/E310A (K295M, catalytic mutant; Y416A, nonautophosphorylated; E310A, cannot be ADP-ribosylated by EspJ). Phosphotyrosine Western blotting revealed that while Csk could still phosphorylate Src/poly(Glu 4 , Tyr) after incubation with NAD and D187A mutant EspJ EHEC or TssF1 (an uncharacterized Pseudomonas aeruginosa protein used as an additional control), WT EspJ EHEC completely abolished Csk kinase activity ( Fig. 6d and e). This confirms that EspJ can ADP-ribosylate and inhibit both the SFKs and their inhibitor Csk.

DISCUSSION
Studying the role of T3SS effectors in vivo conventionally relies on the elucidation of pathological, immunologic, or cellular phenotypes. However, as the contributions of  many individual effectors to disease are subtle, their roles during in vivo infection remain largely unknown. This holds true for EspJ, as an espJ mutant colonizes mice as efficiently as the parental WT strain. In this paper, we show, for the first time, that comparing the proteomes of IECs extracted from mice infected with WT and espJ mutant C. rodentium reveals the in vivo processes affected by EspJ. By applying this approach to other effector mutants, we should be able to illuminate the processes they affect at a molecular level, which will shed new light on pathogen-host interactions.
In this study, we show that EspJ can modify a range of kinases and this is suggested to regulate the host immune response in vivo. Previously, Src kinase was the only known substrate for EspJ. Here we show that kinases from the Src, Abl, Csk, Tec, and Syk NRTK families are preferentially ADP-ribosylated by EspJ. Further, while predictions from in vivo proteomics hinted that EspJ modulates JAK and FAK, these were not ADP-ribosylated in cell lysate. While expression levels may have contributed to these observations, several JAKs and FAKs are ubiquitously expressed and thus were likely available for ADP-ribosylation by EspJ. Furthermore, our data suggest that within the ubiquitously expressed SFKs, EspJ has a preference for Src and Yes1 over Fyn. Thus, EspJ displays an intricate specificity despite ADP-ribosylating a universally conserved catalytic kinase residue, Src E310. The molecular basis of target specificity is currently not known.
The Src, Tec, and Syk NRTK families are crucial to the host immune response to pathogens. This includes roles in phagocytosis and inflammatory responses, both of which were predicted as EspJ-regulated processes in vivo (15,(19)(20)(21)(22). A multitude of findings link these kinases to the transduction of signaling from Toll-like receptors (TLR), such as the inflammatory response to bacterial lipopolysaccharide after detection by TLR4 (23), and Syk has been linked to signal transduction after pathogen recognition by IECs (21). The new EspJ substrates that we have identified were either previously observed in IECs or present in the IEC proteome from this study. As IECs are emerging as nonprofessional immune cells critical to the defense against pathogens (24,25), their inhibition is likely central to the suppression of the host immune response by EspJ in vivo.
Our study showed that EspJ can also ADP-ribosylate Csk at the conserved catalytic glutamic acid, ablating its kinase activity. The EspJ homologues SeoC and SboC were also able to ADP-ribosylate Csk, suggesting that the broad regulation of NRTKs may be a conserved strategy employed by the A/E pathogens and Salmonella. As Csk is only recruited to SFKs at the membrane upon their activation (26), by inhibiting SFKs, EspJ could potentially prevent the recruitment of Csk. If EspJ did interact with Src and Csk simultaneously, it has the potential to dynamically regulate both proteins, which may be necessary for subtle control of processes such as the dynamics of the EPEC actin pedestals. EspJ inhibition of Csk may also impact other roles for Csk, including the regulation of G proteins or the phosphatases SHP-1 and CD45 (27)(28)(29), though further studies are required to understand the global effect of EspJ-Csk interactions. Importantly, while Csk was found to be an EspJ target in lysates of all of the cell line tested, it was modified with less efficiency than the other NRTKs following protein purification. Whether this reduced efficiency is caused by the GFP tag or represents a physiological difference has yet to be determined.
Manipulation of tyrosine kinase signaling is rife during bacterial infections. Tyrosine phosphorylation of EPEC Tir is mediated by Abl and SFKs. Depending on the combination of phosphorylation, either Nck or the phosphatases SHP-1 and -2 may be recruited, stimulating downstream pedestal formation (30,31) or inhibiting the production of proinflammatory cytokines, respectively (32). Phosphorylation of Helicobacter pylori CagA by Src and Abl results in drastic cell elongation and increased motility EspJ ADP-Ribosylates Nonreceptor Tyrosine Kinases ® (33). Shigella recruits Src and Abl to induce the cytoskeletal dynamics necessary for invasion (34), and Btk is required for Shigella flexneri actin tail formation and motility (35). Contrastingly, bacteria may use phosphatases to inhibit NRTKs. Salmonella SptP is able to dephosphorylate Syk, (36), and Yersinia YopH can dephosphorylate multiple proteins, including NRTKs and targets thereof, controlling internalization, integrins, and immunoreceptor signaling (37).
While EspJ is the only bacterial ART that targets mammalian kinases, other T3SS ARTs can suppress the host immune response, such as HopF2 from the plant pathogen Pseudomonas syringae, which ADP-ribosylates mitogen-activated protein kinase kinases. The Pseudomonas aeruginosa ARTs ExoS and ExoT can inhibit phagocytosis by ADP-ribosylation of a distinct set of host proteins despite sharing 76% sequence identity (38). The EspJ homologues show sequence identity as low as 57% (13). Accordingly, these ARTs may have broader target repertoires, though deeper proteomics analysis is required to fully elucidate their specificity and downstream functions.

Animals.
Animal experiments were performed in accordance with the Animals Scientific Procedures Act 1986 and approved by the local Ethical Review Committee and UK Home Office guidelines. Experiments were designed in agreement with the ARRIVE guidelines (39) for the reporting and execution of animal experiments, including sample randomization and blinding. Pathogen-free female C57BL/6 mice (18 to 20 g, six per group; Charles River, Inc., United Kingdom) were housed in HEPAfiltered cages with sterile bedding (processed corncobs grade 6) and nesting (LBS Serving Technology) and free access to sterilized food (LBS Serving Technology) and water.
DNA manipulation: plasmid construction and site-directed mutagenesis. pCB6-GFP plasmids containing Src (WT and R175K mutant), Yes1, Fyn, and Abl were kindly provided by Michael Way (Francis Crick Institute, London). Csk was subcloned into pCB6 and pET28 by restriction enzyme digestion, ligation, and transformation. Site-directed mutagenesis for EspJ D187A and Csk E236Q was performed by nonoverlapping inverse PCR, followed by product phosphorylation and ligation with T4 DNA ligase. For the primers and restriction enzymes used, see Table S1, and for the plasmids used, see Table S2. All enzymes were from NEB.
IP. HeLa cells were seeded into six-well plates at 1.5 ϫ 10 5 /well and transfected 24 h later with 2.5 g of plasmid DNA and 5 l of Lipofectamine 2000 (Thermo Fisher Scientific) per well. Fourteen hour later, cells were washed twice with phosphate-buffered saline (PBS) and lysed with kinase immunoprecipitation (IP) buffer on ice before sonication. Lysates were clarified at a relative centrifugal force (RCF) of 20,000 for 10 min and precleared for 15 min with Dynabeads protein G (Thermo Fisher Scientific). Incubations were performed at 4°C with rotation. Supernatants were incubated with rabbit anti-GFP antibody 9F9.F9 (Abcam, Inc.)-conjugated Dynabeads for 1 h. Beads were sequentially washed with PBS-0.5% Triton X-100, PBS-0.05% Tween 20, 150 mM NaCl-20 mM Tris (pH 8), and kinase IP buffer before ADP-ribosylation by recombinant MBP-EspH EHEC .
In vitro enzyme reactions. Four micrograms of Csk was added to 4 g of purified MBP-EspJ EHEC or to EspJ homologues bound to 50 l of amylose resin. Immunoprecipitated tyrosine kinases were incubated with 2 g of purified MBP-EspJ EHEC . The above ADP-ribosylation reactions with 6-biotin-17-NAD ϩ (NAD-biotin; AMSBIO) were performed for 1 h at room temperature in ADP-ribosylation buffer ( Table 3) and terminated with Laemmli buffer and boiling.
For proteomics screening of EspJ targets, 400 g of HeLa, A549, Caco2, or Thp1 cell lysate in ADP-ribosylation buffer was incubated with 20 g of MBP-EspJ EHEC and eNAD (Jena Biosciences CLK-043) for 3 h at room temperature.
Samples were diluted with PBS to 0.2% SDS-PBS and incubated with 30 l of slurry or prewashed NeutrAvidin resin (Thermo Fisher Scientific) per sample for 1 h. Resin was washed three times with 1%  Digested peptide supernatants were combined with supernatants from subsequent washes with 80 l of 50 mM ammonium bicarbonate and 80 l of 0.1% trifluoroacetic acid (TFA). Peptides were loaded onto methanol-activated and pre-equilibrated home-made StageTips consisting of three layers of SDB-XC (Empore) solid-phase extraction medium, desalted with H 2 O, and eluted with 79% acetonitrile (ACN) before being vacuum dried and stored at Ϫ80°C. For analysis, samples were resuspended in 15 l of 0.5% TFA-2% ACN and transferred to LC-MS vials.
LFQ MS-in vitro ADP-ribosylation. (i) MS. For a detailed description of MS and data processing, see Text S1 in the supplemental material. Analysis was performed with an EASY-Spray LC column coupled to a Q Exactive mass spectrometer via an EASY-Spray Source (Thermo Fischer Scientific) in a datadependent acquisition mode (data were processed using MaxQuant version 1.5.7.4).
(ii) Volcano plot generation. Protein groups were analyzed with Perseus (version 1.5.6.0). Potential contaminants, "reverse," "identified by site," and proteins with only one unique peptide were removed and data were logarithmized (log 2 ). Replicates were grouped and filtered for proteins with at least three valid values across three (Caco2, THP-1) or four (A549, HeLa) replicates in one group. Missing values were imputed with a downshifted normal distribution (1.8 downshift, 0.3 width) for each sample to allow statistical analysis by a two-sided t test within the volcano plot function (permutation-based FDRcorrected P value of Ͻ0.01, s 0 value of 1) ( Table S5).
Infection of mice with C. rodentium. Oral gavage was used to inoculate mice with 200 l of PBS (uninfected) or C. rodentium (ca. 5 ϫ 10 9 CFU). The inoculum was determined by counting CFU after plating on LB agar supplemented with nalidixic acid, and colonization was monitored by counting CFU per gram of stool sample on days 3, 6, 7, and 8.
Tissue staining and measurement. A 0.5-cm section of the distal colon was washed with PBS, fixed in 1 ml of buffered formalin, paraffin embedded, and sectioned at 5 m. Sections were stained with hematoxylin and eosin (H&E) or treated with sodium citrate antigen demasking solution prior to immunofluorescence staining (antibodies, Table 2). CCH was assessed by measuring at least 20 crypt lengths in samples from at least four mice per group.
Enterocyte extraction. Eight days postinfection, a 4-cm section of the terminal colon was cut lengthwise and incubated for 45 min at 37°C in 4 ml of enterocyte dissociation buffer. Enterocytes were harvested by centrifugation at an RCF of 2,000, washed twice in PBS, and stored at Ϫ80°C prior to labeling.
Digestion and TMT labeling. Enterocytes were dissolved in 100 l of 0.1 M triethylammonium bicarbonate (TEAB)-0.1% SDS and lysed by pulse probe sonication. Eighty micrograms of protein per sample was SpeedVac dried and then resuspended in 100 l of 4% SDS-100 mM TEAB-15 mM TCEP, assisted with an ultrasonic bath. After reduction at 56°C for 20 min, samples were cooled to 25°C and alkylated with iodoacetamide (IAA) for 30 min. Proteins were purified by 20% trichloroacetic acid precipitation, followed by one wash with ice-cold acetone before resuspension in 100 mM TEAB and digestion with 3 g of trypsin (MS grade; Pierce) at 37°C for 18 h. Forty micrograms of protein digest was labeled with 0.4 mg of TMT10plex as instructed by the manufacturer (Thermo Fisher Scientific).

IEC proteome MS. (i) MS.
For a detailed description of MS and data processing, see Text S1 in the supplemental material. Briefly, samples were fractionated with a U3000 high-performance liquid chromatography system (Thermo Fisher Scientific) coupled to an XBridge ethylene-bridged hybrid C 18 column (Waters). Peptides were injected onto the Orbitrap Fusion Tribrid mass spectrometer coupled to a U3000 RSLCnano ultrahigh-performance liquid chromatography system (Thermo Fisher Scientific) loading onto a PepMap C 18 trap and separated on a PepMap C 18 column. Data acquisition was done by the SPS10-MS3 method.
(ii) Bioinformatics analysis. Protein groups were processed with Perseus (version 1.5.6.0). Absolute intensities were logarithmized (log 2 ), and those with only one unique peptide were removed. Ratios between conditions were created, and proteins with a fold change of Ͼ1.5 were analyzed further. Venn diagrams were created in BioVenn (41), and heat maps were created in Perseus.
GO terms were analyzed with the ClueGO Cytoscape plug-in (version 3.6.0) (42,43). The minimum and maximum GO levels were set to 3 and 7, respectively. GO terms with a P value of Ͻ0.01 were selected with a minimum of five proteins and 5% of the GO term proteins. GO term grouping and fusion were utilized, and the most significant (Bonferroni-corrected P value) GO term was used for visual representation.
The kinase prediction function (KEA) of X2K (16) was used in isolation to predict upstream regulatory kinases from the group of 457 proteins upregulated in the ΔART versus WT samples or from 10 randomly generated sets of 457 proteins from the total of 7,400 proteins identified.
Availability of data. The data sets obtained in this study can be found in PRIDE project PXD008533.

ACKNOWLEDGMENTS
We thank Valerie Crepin, Izabela Glegola-Madejska, and Agnes Sågfors for technical in vivo assistance. We thank Julia Morales Sanfrutos and Lisa Haigh for maintenance and operation of the Imperial College London MS facilities. We thank Michael Way for providing pCB6 plasmids encoding tyrosine kinases and Eleni Manoli for providing TssF1 protein.
D.P. was supported by a BBSRC studentship. L.Y. and J.S.C. were funded by a core grant from the Wellcome Trust (098051) to the Sanger Institute. G.F. has been supported by grants from the MRC and the Wellcome Trust.
D.J.P. performed each experiment and analysis and wrote the manuscript with G.F. L.Y. and K.H. prepared and ran MS samples. C.N.B., E.C.S., E.W.T., and J.S.C. contributed crucial discussion for the study and manuscript preparation.