Quantitative Phosphoproteomics of Cytotoxic T Cells to Reveal Protein Kinase D 2 Regulated Networks*

The focus of the present study was to characterize the phosphoproteome of cytotoxic T cells and to explore the role of the serine threonine kinase PKD2 (Protein Kinase D2) in the phosphorylation networks of this key lymphocyte population. We used Stable Isotope Labeling of Amino acids in Culture (SILAC) combined with phosphopeptide enrichment and quantitative mass-spectrometry to determine the impact of PKD2 loss on the cytotoxic T cells phosphoproteome. We identified 15,871 phosphorylations on 3505 proteins in cytotoxic T cells. 450 phosphosites on 281 proteins were down-regulated and 300 phosphosites on 196 proteins were up-regulated in PKD2 null cytotoxic T cells. These data give valuable new insights about the protein phosphorylation networks operational in effector T cells and reveal that PKD2 regulates directly and indirectly about 5% of the cytotoxic T-cell phosphoproteome. PKD2 candidate substrates identified in this study include proteins involved in two distinct biological functions: regulation of protein sorting and intracellular vesicle trafficking, and control of chromatin structure, transcription, and translation. In other cell types, PKD substrates include class II histone deacetylases such as HDAC7 and actin regulatory proteins such as Slingshot. The current data show these are not PKD substrates in primary T cells revealing that the functional role of PKD isoforms is different in different cell lineages.

The focus of the present study was to characterize the phosphoproteome of cytotoxic T cells and to explore the role of the serine threonine kinase PKD2 (Protein Kinase D2) in the phosphorylation networks of this key lymphocyte population. We used Stable Isotope Labeling of Amino acids in Culture (SILAC) combined with phosphopeptide enrichment and quantitative mass-spectrometry to determine the impact of PKD2 loss on the cytotoxic T cells phosphoproteome. We identified 15 The mammalian serine/threonine protein kinase D (PKD) 1 family comprises three different but closely related serine kinases, PKD1, PKD2, and PKD3 all of which have a highly conserved N-terminal regulatory domain containing two cysteine-rich diacylglycerol (DAG) binding domains (1). T lymphocytes express high levels of PKD2 and this kinase is selectively activated by the T-cell antigen receptor (TCR). The activation of PKD2 is initiated by DAG binding to the PKD N terminus but is also critically dependent on Protein kinase C (PKC)-mediated phosphorylation of two serine residues (Ser707 and Ser711) within the activation loop of the PKD2 catalytic domain (2,3). The importance of PKD2 for T-cell function has been probed by experiments in mice that lack expression of catalytically active PKD2. These studies have shown that PKD2 is important for effector cytokine production after T-cell antigen receptor engagement and also for optimal induction of T-cell dependent antibody responses (4,5). PKD2 thus has a key role in adult mice to control the function of T cells during adaptive immune responses.
The importance of PKD2 for primary T-cell function makes it critical to understand how PKD2 controls protein phosphorylation pathways. In this context, experiments with constitutively active and dominant negative PKD mutants in tissue culture cell lines have identified a number of candidate PKD substrates. These include the protein phosphatase Slingshot (6,7), the Ras effector Rin1 (8), phosphatidylinositol-4 kinase III beta (9), lipid and sterol transfer proteins such as CERT and OSBP (10,11). There are also experiments that have identified a key role for PKDs in regulating the phosphorylation and subcellular localization of the class II histone deacetylases (HDACs). For example, in PKD null DT40 B cell lymphoma cells the B cell antigen receptor cannot induce the phosphorylation and nuclear exclusion of the class II HDACs, HDAC5 and 7 (12). However, it remains to be determined whether the documented PKD substrates are universal PKD substrates in different cell lineages. In this context, the intracellular localization of PKD isoforms varies in different cells (13), and PKDs have also been shown to traffic between different cellular locations in response to specific stimuli (2,14). PKD function is dependent on its localization and cell context presumably reflecting that the localization of PKDs plays a key role determining the nature of PKD substrates in different cell populations (15).
Recently, mass-spectrometry based quantitative phosphoproteomics has been used to explore serine/threonine kinase controlled signaling pathways in T cells (16 -18). In this regard, SILAC labeling combined with quantitative mass-spectrometry has recently been used to examine the impact of overexpressing active and/or kinase dead PKD1 mutants in HEK293 cells treated with nocodazole, a microtubule-depolymerizing reagent that disrupts the Golgi complex and activates PKD1 (19). This has identified a number of PKD1 substrates in HEK293 cells. PKD1 and PKD2 are highly homologous kinases but it remains to be determined whether the PKD1 substrates identified in nocodazole-treated HEK293 cells are relevant to signaling pathways controlled by endogenous PKD2 in antigen receptor activated primary T cells.
Accordingly, in the present study we used SILAC labeling combined with phosphopeptide enrichment and mass-spectrometry quantification to compare the phosphoproteome of antigen receptor activated wild type and PKD2 deficient cytotoxic T cells (CTLs). Our experiments identify and quantify more than 15,000 site-specific phosphorylations in antigen receptor activated CTLs and thus provide a unique data source about the signaling networks operational in these cells. The loss of PKD2 impacts on about 5% of these phosphorylations and reveals that PKD2 has both positive and negative regulatory roles in regulating protein phosphorylation networks in T cells.

EXPERIMENTAL PROCEDURES
Mice, Cell Culture, and SILAC Labeling-P14 T-cell receptor transgenic mice (P14-TCR) PKD2 null mice (4,5), and wild-type littermates were bred and maintained under specific pathogen-free conditions in the Wellcome Trust Biocenter at the University of Dundee in compliance with U.K. Home Office Animals (Scientific Procedures) Act 1986 guidelines as previously describe (17,20). P14 CTL were generated and labeled in SILAC media as previously described (17). Briefly, splenocytes were activated for 2 days with the P14-TCR cognate ligand (peptide gp33-41 from Lymphocytic Choriomeningitis Virus, LCMV). Then, cells were cultured for 4 days in SILAC medium (Dundee Cell Products, Dundee, UK), L-proline 200 mg/l, L-arginine 84 mg/l, pre-supplemented with 300 mg/l L-glutamate, 10% dialyzed FCS with a 10kDa cutoff (Invitrogen, Carlsbad, CA), 50 units/ml penicillin-G, 50 g/ml streptomycin and 50 M ␤-mercaptoethanol, and 20 ng/ml IL-2 (Proleukin, Novartis, Basel, Switzerland). The following arginine and lysine isotopes were used: The SILAC labeling was performed in three biological replicates, where the P14 wild-type CTLs cells comprised the "light" condition and PKD2 null CTLs were labeled with "heavy" amino acids in two experiments (experiments 1 and 2), and a label switch was performed in the third experiment (experiment 3).
Eighteen fractions of each subcellular compartment were collected throughout the gradient and further enriched in phosphopeptides by immobilized metal affinity chromatography (IMAC, Phos-Select, Sigma) following manufacturer's instructions. Peptides were eluted with 200 l, 0.4 M NH4OH followed with 200 l, 0.2 M NH4OH/50% acetonitrile. A gel-loading tip was used to remove elution fractions and further remaining IMAC beads were removed by passing fractions through a ZipTip (Millipore). The supernatants from the IMAC (combining the 18 fractions for each subcellular compartment in nine samples) were used for additional TiO 2 pull down (Titanspheres, GL Sciences, Shinjuku, Japan).
Liquid Chromatography -Mass Spectrometry (LC-MS)-The peptide mixture was separated by nanoscale C18 reverse-phase liquid chromatography (Ultimate 3000 nLC (Dionex, Sunnyvale, CA) coupled online to a Linear Trap Quadrupole (LTQ)-Orbitrap mass spectrometer (LTQ-Orbitrap Velos; Thermo Fisher Scientific). The following buffers were used: HPLC Buffer A (2% acetonitrile and 0.01% formic acid), HPLC Buffer B (90% acetonitrile and 0.08% formic acid) and HPLC Buffer C (0.05% trifluoroacetic acid). Samples were injected in 1% formic acid, washed onto the column with HPLC Buffer C and eluted with a flow of 0.3 l/min under usage of the following buffer gradient: 5% B (0 -3 min), 5-35% B (3-68 min), 35-90% B (68 -70 min), 90% B (70 -80 min), 90 -5% B (80 -81 min), and equilibrated in 5% B (81-100 min). The eluting peptide solution was automatically (online) electrosprayed into the mass spectrometer using a nanoelectrospray ion source (Proxeon Biosystems, Odense, Denmark). The mass spectrometers were operated in positive ion mode and used in data-dependent acquisition modes. A full scan (FT-MS) was acquired at a target value of 1,000,000 ions with resolution r ϭ 60,000 over a mass range of 335-1800 amu (atomic mass unit). The ten most intense ions were selected for fragmentation in the LTQ Orbitrap Velos. Fragmentation in the LTQ was induced by collision-induced dissociation (CID) with a target value of 10,000 ions. For accurate mass measurement, the "lock mass" function (lock mass ϭ 445.120036 Da) was enabled for MS scan modes. To improve the fragmentation of phosphopeptides, the multistage activation algorithm in the Xcalibur software was enabled for each MS/MS spectrum using the neutral loss values of 48.99, 32.66, and 24.50 m/z units. Former target ions selected for MS/MS were dynamically excluded for 300 s. General mass spectrometric conditions were as follows: spray voltage, 1.0 -2.5 kV; no sheath and auxiliary gas flow; ion transfer tube temperature, 150 -180°C; and normalized collision energy (35%) using wide band activation mode for MS2. The isolation width was set to 2 amu for IT-MS/MS. Ion selection thresholds were 600 counts for MS2. An activation of q ϭ 0.25 and activation time of 30 ms were applied in MS2 acquisitions. The fill time for FTMS was set to 1000 ms and for ITMS to 150 ms. In this study, three independent biological replicates of P14 wild-type versus PKD2 knockout CTLs were analyzed, with two technical replicates for each biological sample (total 310 HPLC-MS/MS runs). In two experiments, the wild-type cells comprised the "light" condition, whereas the PKD2 knockout CTLs were labeled with "heavy" amino acids (experiments 1 and 2). In the third experiment, we performed a label switch (experiment 3).
Data Processing-For data analysis, we combined the raw data obtained from the two technical replicates, the two subcellular fractions (nucleus and cytosol), and the IMAC and TiO 2 phosphopeptide enrichment methods in each independent replicate. Data was processed using MaxQuant (21) version 1.3.0.5 which incorporates the Andromeda search engine (22). Proteins were mapped to the Uniprot mouse protein database ("Mouse Complete Proteome" retrieved on August 19, 2013). This version of the database contains 16,618 mouse complete proteome entries (UniProtKB/Swiss-Prot canonical and isoform sequence data). Search parameters specified an MS tolerance of 20 ppm, an MS/MS tolerance at 0.5 Da and full trypsin specificity, allowing for up to two missed cleavages. Carbamidomethylation of cysteine was set as fixed modification and oxidation of methionines, N-terminal protein acetylation, and phosphorylation of serine, threonine, and tyrosine were set as variable modifications. Peptides were required to be at least six amino acids in length with false discovery rates (FDRs) of 0.01 calculated at the level of peptides, proteins, and modification sites based on the number of hits against the reversed sequence database.
To make our data accessible to the scientific community, the MS proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository (23) with the data set identifier PXD001076. The annotated_spectra.zip file contains .pdf files with annotated MS/MS spectra for all PTM containing peptides described in the manuscript. Separate MS/MS spectra for the highest identification and localization scores have been deposited. To facilitate the search for the annotated spectra, an additional excel file (Phospho (STY)Sites_an-notated_spectra.xlsx file) containing a list of all reported sites and the file names of their corresponding .pdf files have been deposited. Prior to statistical analysis, the outputs from MaxQuant were filtered to remove known contaminants and reverse sequences. The distribution of SILAC ratios was normalized within MaxQuant at the peptide level so that the median of log 2 ratios is zero (21) (supplemental Table S1).
Bioinformatics and Statistical Tools-For functional pathway analysis, the different subsets of proteins identified in the data set were subjected to functional analysis using DAVID bioinformatics resources (24). Gene ontology terms for biological processes (BP) and molecular functions (MF) charts were obtained using default statistical parameters (threshold: count 2, ease 0.1). Proteins with protein kinase activity and transcription factor activity were filtered based on GOTERM_MF_FAT (GO:0004672 and GO:0003700, respectively).
For the kinase motif distribution, Linear Motif analysis tool in Perseus v1.4.1.3 software was used. Significance of PKD2 motif enrichment was assessed by Chi-square test using Graphpad software (http://www.graphpad.com/).

RESULTS
The CTL Phosphoproteome-The present screen used SI-LAC-based quantitative phosphoproteomic analysis to characterize the kinase-substrate signaling networks present in antigen receptor activated cytotoxic T cells (CTL). SILAC labeling requires that cells undergo multiple cell doublings to ensure sufficient label penetrance. Accordingly, for SILAC experiments in CD8 T cells we used a well-characterized model of CD8 CTL differentiation (17). In this model, antigen primed CD8 ϩ T cells from P14 TCR transgenic mice are cultured in interleukin 2 (IL-2) to produce a homogenous population of fully differentiated effector CTL (20,26,27). This model reproduces the in vivo situation where sustained IL-2 signaling promotes the production of terminally differentiated effector CTLs required for virus clearance (28 -30). The impact of loss of the serine-threonine kinase PKD2 on CTL signaling networks was also explored. Hence, in these experiments wild-type or PKD2 knockout P14-TCR CTLs were differentially labeled with one of two different isotope combinations of lysine (K) and arginine (R), R0K0 ("light") and R10K8 ("heavy"). After 4 days of culture in SILAC media, wild-type and PKD2 knockout CTLs were mixed at equal cell number and triggered via their TCR with cognate peptide for 5 min to maximally activate PKD2 in the wild-type population. Cells were then lysed, digested with trypsin and phosphopeptides were enriched by HILIC fractionation followed by IMAC and TiO 2 affinity chromatography. The workflow of our experiments is displayed in Fig. 1A. The phosphopeptide-enriched fractions were analyzed in a LTQ-Orbitrap Velos for mass spectrometry (MS) data collection. All raw mass spectrometry data from three biological replicates were processed using the Max-Quant software.
A key aim of this study was to map the repertoire of protein phosphorylations in CTLs. The collective analysis of the current experiments identified and quantified 15,871 unique phosphorylation sites in CTLs on 3505 distinct proteins. The full list of all unique phosphosites identified in our analysis can be found in supplemental Table S1. The ratio of phosphorylations on Serine, Threonine, and Tyrosine residues was 81/ 17/2, comparable to what has been observed in other phosphoproteomic studies (31,32). Importantly, the total number of phosphosites identified in the individual biological replicates was similar (Fig. 1B), and ϳ58% of the unique phosphosites (9205 on 2767 proteins) were identified in all three replicates (Fig. 1C). We have previously described a phosphoproteomic analysis of CTLs but these earlier experiments only identified 2000 phosphorylations in CTLs (17). The higher coverage obtained in the present set of experiments offers a more complete view of CTL signaling networks.
We used DAVID Bioinformatics Resources 6.7 (24) to perform functional pathway analysis on the phosphoproteins and found that the CTL phosphoproteome is significantly overrepresented by proteins that control gene transcription and chromatin and by kinases and phosphatases that control protein phosphorylation (Fig. 1D). There was also overrepresentation in the CTL phosphoproteome of molecules that regulate macromolecular catabolic processes, notably proteins that control the ubiquitylation and sumoylation of proteins (Fig. 1D). The mouse genome encodes over 500 protein kinases (33). The present data detected phosphopeptides from ϳ200 protein kinases in CTLs (supplemental Table S2). We have also used Linear Motif analysis tool in Perseus software to assign the kinases most likely to phosphorylate the 15,871 phosphosites identified in CTLs. This work indicated activity of a minimum of 54 serine/threonine kinases and eight tyrosine kinases in CTLs (Fig. 1E). One limitation of this latter analysis is that is difficult to assign substrates individually to closely related kinase isoforms that may share substrate specificity. For example, a proportion of the identified phosphosites in CTLs was predicted to be phosphorylated by Protein Kinase A (PKA) or Protein Kinase C (PKC) (Fig. 1E). These are members of the AGC family of kinases and it is very likely that substrates assigned to PKA or PKC may in fact be phosphorylated by other AGC kinases. In this context, the present data reveals the full repertoire of AGC kinase isoforms expressed in CTLs. For example, five of the possible nine Protein Kinase C isoforms were detected in CTLs, PKC␣ ␤, ␦, PKC, and PKC (Table I). The data presented in Table I show that AGC kinases in CTL also include two of the four RSK isoforms, RSK1 and RSK2; two of the three Akt isoforms, Akt1 and Akt2; one of the three serum glucocorticoid kinases, SGK3, and one of the two possible S6K isoforms, S6K1. In a similar way, the kinase prediction analysis also indicated activity of the adenosine monophosphate (AMP)-activated protein kinase in CTL (Fig.  1E). It is known that CTL are dependent for their survival on LKB1 (34), which phosphorylates and activates AMPK␣1, AMPK␣2, and 13 other AMPK-related kinases (AMPK-RK) (35). The deletion of AMPK␣1 does not phenocopy the loss of LKB1 in T cells indicating that it must be other AMPK family members that mediate LKB1 actions (36). The present data identify the candidates to mediate LKB1 function in T cells, because CTLs express multiple AMPK family isoforms apart from AMPK␣1 notably SNRK; SIK1 and 3, and MARK2, 3, and 4 ( Table I). The data presented in Table I and supplemental  Table S2 identifying the specific kinase isoforms expressed in CTLs offer invaluable knowledge in the context of the selection of isoform-specific inhibitors of kinase activity and the development of new inhibitors targeting T-cell function.
The relevance of kinase-substrate networks present in CTLs relies in their ability to dictate the transcriptional program controlling T-cell effector function during an immune response. Accordingly, the most overrepresented function among phosphorylated proteins in CTLs is the regulation of transcription (Fig. 1D). Of the 160 proteins with annotated transcription factor activity identified in our data set (supplemental Table S3), we found relevant phosphorylation sites that regulate transcriptional activity, subcellular localization, and/or protein stabilization of essential transcription factors for CTL development such as the STAT family members STAT3, STAT4, and STAT5, Foxo1, Foxo3, or Myc (Table II). More importantly, we identify phosphorylation sites in key transcription factors such as Eomes, T-bet, Hif1␣, Arnt, Tfeb, Srebf1, and Srebf2 or Irf4 whose function has not been yet characterized (Table II). Notably, several of the phosphosites found for these transcription factors are predicted to be phosphorylated by kinases known to be active in antigen stimulated T cells such as Erk1/2, GSK3, or PKC family (Table II and  supplemental Table S3). These data thus open the door for novel targeted approaches exploring the role of protein phosphorylation in regulating CTL transcriptional program.
Impact of PKD2 Deficiency on the Cytotoxic T-cell Phosphoproteome-A second aim of the present screen was to use SILAC-based quantitative phosphoproteomic analysis to explore the role of PKD2 in CTLs. The present data reveal that antigen activated CTLs have high levels of active PKD2. We found eight unique phosphosites derived from PKD2 in wildtype CTLs including peptides corresponding to phosphorylated residue Ser711 (Table III). This phosphorylation is mediated by PKCs and is critical for PKD2 catalytic activity (4). When active, PKD2 autophosphorylates on Ser873 (37); the presence of phosphorylated PKD2-Ser873 in CTLs thus con- firms the activity of this kinase in these cells (Table III). Importantly, the SILAC ratio for all the phosphosites found for PKD2 was down-regulated in PKD2 deficient CTLs, particularly the C-terminal phosphosites (Table III). Previous studies have indicated that T cells do not express PKD1 and only express low levels of PKD3 (4). The current analysis found six phosphopeptides for PKD3, and another four whose sequence can be assigned to either PKD3 or PKD1. In addition, the data shows that there was no change in the SILAC ratio for PKD3 phosphosites in PKD2 deficient CTLs (Table III). Thus, the loss of PKD2 catalytic function in CTLs is not compensated by increased expression or activity of other PKD isoforms.
PKD2 Regulated Phosphorylations in CTL-There was some impact of PKD2 loss on the CTL phosphoproteome. Hence, of the 15,871 unique phosphosites identified and quantified in CTLs, 450 phosphosites on 281 distinct proteins were down-regulated and 300 phosphosites on 196 proteins were up-regulated in PKD2 null CTLs (Fig. 2B). The threshold for change was set to a z-core of 2 for phosphorylation ratios in PKD2 null versus wild-type CTLs, using the averaged value of the three biological replicates (1.8-fold change). We discarded phosphorylation events that displayed an inconsistent trend in regulation among the three biological replicates (i.e. down-regulated in one replicate and up-regulated in another). Based on this threshold, the data showed that PKD2 directly and indirectly regulates about 5% of the phosphorylations of cytotoxic T cells, on proteins that are mainly involved in transcription, chromosome reorganization and more importantly in the context of cytotoxic T cells, in leukocyte activation and hemopoiesis (Fig. 2C).
Among the down-regulated sites implicated in leukocyte activation we found five phosphosites derived from CD5 (Fig.  2D), a cell surface glycoprotein whose expression is normally increased after activation in T cells (40). This could reflect loss of CD5 phosphorylation or loss of CD5 expression. In this respect we have established that PKD2 controls expression of CD5 during T-cell development (15) but its role in controlling CD5 expression in peripheral T cells is unknown. Accordingly, we used flow cytometry to examine if CD5 expression was impaired in PKD2 deficient CTLs. Data in Fig. 2E show that PKD2 deficient CTLs express lower levels of CD5 than their wild-type counterparts.
We also found two down-regulated phosphosites derived from the F-actin severing protein Cofilin, Ser3, and Ser41, in the TCR activated PKD2 null CTL (Fig. 2F). This result was initially surprising as previous studies have shown that TCR triggering, which activates PKD2, normally down-regulates Cofilin-Ser3 phosphorylation (17). Indeed, Western blot experiments confirmed that TCR triggering reduces Cofilin-Ser3 phosphorylation in CTLs (Fig. 2G). Nevertheless, there was strikingly less phosphorylated Cofilin-Ser3 in the both nonstimulated and TCR stimulated PKD2 null T cells compared with wild-type cells (Fig. 2G). These data orthogonally validated the reduced Cofilin-Ser3 phosphorylation in the TCR activated PKD2 null T cells observed in the mass spectrometry experiments (Fig. 2F, G). The reduced Cofilin phosphorylation in PKD2 null T cells argues that PKD might control the activity of kinase pathways upstream Cofilin phosphorylation. In this respect, previous studies have reported a PKD requirement for the phosphorylation of Ser3 on Cofilin via PKD activation of PAK4/LIMK (41,42).
To search for PKD2 substrate candidates within the data set we used the following criteria: the SILAC ratio must be down-regulated in the PKD2 null cells, and the sequence of the down-regulated phosphosite must correspond to a PKD2 consensus motif. The consensus sequence for PKD2 phosphorylation described in the literature is (L/V/I)x(R/K)xx(S*/T*) (43)(44)(45). In addition to the described consensus site, PKD2 has an auto-phosphorylation site at residue Ser873 (Ser916 in PKD1) (37). This PKD2 autophosphorylation motif has slightly modified sequence compared with the conventional motif, (L/I)xx(R/K)x(S*/T*), but because it is a well described substrate for PKD2 we have included this sequence in our search for PKD2 phosphorylation sites. The analysis of the whole data set determined that among the 15,871 phosphosites found in our study, 940 contained a consensus motif for PKD2 phosphorylation (6% of the data set) (supplemental Table S1). Strikingly, 73 out of the 450 phosphosites (16%) with decreased phosphorylation in PKD2 deficient CTLs had a PKD2 consensus motif, whereas only 26 out of the 300 phosphosites (9%) with increased phosphorylation matched the PKD2 consensus motif (Fig. 2H). Thus among the down-regulated phosphosites in PKD2 knockout CTLs, we had a significant enrichment of phosphosites containing a consensus phosphorylation site for PKD2 (p Ͻ 0.0001, Fig. 2I). Detailed information about modified position and ratios of the 73 phosphosites on 69 distinct proteins down-regulated in our data set that had a PKD2 consensus motif can be found in FIG. 2. Impact of loss of PKD2 in the CTL phosphoproteome. A, Histogram shows the SILAC ratio distribution of the data set, using the averaged SILAC ratio of the three biological replicates (PKD2 knockout versus wild-type, AVG KO/WT, log 2 value). B, Graph shows SILAC ratio distribution (AVG KO/WT, log 2 value) plotted against the signal intensity (sum of intensities of the three biological replicates, log 10 ) for all identified phosphopeptides. Dark dots and inset numbers indicate phosphosites (p-sites) with a z-score of 2 (1.8-fold change). C, Graph supplemental Table S4. It was notable that 377 of the 450 phosphosites with decreased phosphorylation in PKD2 deficient CTLs, including Cofilin1-Ser3 (Fig. 2E), did not have a consensus PKD2 site and hence are unlikely to be directly phosphorylated by PKD2 (Fig. 2B, 2F). In addition, loss of PKD2 increased the phosphorylation of 300 phosphosites (Fig. 2B) and this is clearly an indirect consequence of PKD2 loss. Table VI lists the 69 proteins phosphorylated on a PKD2 consensus sequence whose phosphorylation was decreased in PKD2 null T cells. It also lists the biological function ascribed to these proteins. The decreased phosphorylation of these proteins in PKD2 null cells could reflect that they are direct substrates for PKD2. However, it is possible that the expression of the protein is decreased. To attempt discriminate between these possibilities, we analyzed all the phosphosites found in our study for the proteins presented in Table  VI. For 41 proteins we could find evidence that although they had lost phosphorylation on PKD consensus sequences there were other phosphorylation sites that were unchanged, indicating that the protein expression was not decreased. For 22 proteins only one to two phosphosites were found and hence we cannot make any conclusion about their expression. However, for another four proteins, HDAC7, Osbl3, Sgk223, and Specc1, there was decreased phosphorylation of all detected peptides derived from these proteins (supplemental Table  S1). This could indicate that there is lower expression of these proteins in PKD2 deficient CTLs. This second group included histone deacetylase HDAC7 (Table VI), which is a well characterized PKD substrate in many cells including B cells (12,46). Moreover, in CTLs, HDAC7 phosphorylation on PKD consensus sites is essential for HDAC7 nuclear exclusion (17). HDAC7 is thus a very strong candidate for a PKD2 substrate in CTLs. However, Fig. 3A shows that the loss HDAC7 phosphorylation in PKD2 null CTLs occurred on multiple sites and was not restricted to the PKD substrate sequences Ser156 and Ser182 (47,48). We therefore explored the possibility that PKD2 knockout CTLs may have decreased expression of HDAC7. The Western blot in Fig. 3B addresses this question and shows that PKD2 knockout CTLs express lower levels of HDAC7 compared with wild-type CTLs. The decrease in HDAC7 phosphorylation in PKD2 null CTL thus reflects that these cells have reduced expression of this molecule. In this respect, nonphosphorylated HDAC7 would accumulate in the nucleus whereas phosphorylated HDAC7 is in the cytosol (17). Confocal microscopy comparing the intracellular location of HDAC7 in wild-type and PKD2 deficient CTLs clearly showed that HDAC7 localization is mainly cytosolic in both wild-type and PKD2-deficient CTLs (Fig. 3C). HDAC7 phosphorylation on PKD consensus sites is also essential for CTL to express the IL-2Ralpha chain (CD25) (17). Fig. 3D shows that PKD2 deficient cells express normal levels of CD25. These data collectively argue that HDAC7 is not a PKD2 substrate in cytotoxic T cells.  The data presented for HDAC7 is thus a good demonstration of how decreased phosphorylation of a protein in a particular cell may be because of the fact its expression is decreased. Nevertheless, in the current data set there were at least 41 proteins that had decreased phosphorylation on PKD substrate sequences yet phosphorylations on other sites were unchanged, making these strong candidates to be PKD2 substrates (Table VII). Six of these proteins, Hip1r, Map4, Rdbp, Ssr3, Snx2, and Pcbp1, have been shown to be in vitro substrates for PKD1 (19) (Tables VI and VII). We also identified Rin3 (Ras and Rab interactor 3) as a PKD2 substrate (Tables  VI and VII) and previous studies have identified Rin1 as a PKD1 substrate in fibroblasts (49). It was, however, important that our studies identified novel PKD2 substrate candidates in CTLs such the E3 ubiquitin ligase c-Cbl (Cbl) (50), the Lysine (K)-specific demethylase 2a (Kdm2a) (51), the transcription factor NFAT3 (52) and the Phosphorylated adaptor for RNA export (Phax), a regulator of U snRNA nuclear export (53). DISCUSSION The present study explores the phosphoproteome of antigen receptor activated cytotoxic T lymphocytes and identifies 15,871 unique phosphorylation sites on 3505 distinct proteins in this critical lymphocyte population. This is a unique data set about the kinase-substrate signaling networks in CTL that allows new insights about the repertoire of protein kinases that are expressed and active in T cells. One new vision derived from this work is that the CTL phosphoproteome was overrepresented by proteins that regulate macromolecular catabolic processes, notably proteins that control the ubiquitylation and sumoylation. In this respect, the reversible ubiquitin or sumo modification of proteins is critical for intracellular signal transduction pathways (54,55). Moreover, it has been recognized for several years that cross talk between phosphorylation and ubiquitylation/sumoylation signaling occurs (56,57). For example, phosphorylation of various E3 ligases can either positively or negatively regulate their deubiquitylation activity (56). The phosphorylation of a protein can create a docking site for a particular E3 ligase whereas in other cases, phosphorylation controls the intracellular localization of E3 ligases and their substrates (56). The full extent of the cross talk between protein phosphorylation and ubiquitylation is not understood in any cell system. However, the overrepresentation in the CTL phosphoproteome of molecules linked to the control of ubiquitylation and sumoylation signaling is a striking indication that the intersection between different post-transcriptional modifications may be critical to control CTL function.
The study also presents an exhaustive analysis of the impact of the loss of a single kinase, PKD2, in CTLs using an unbiased phosphoproteomic approach. This global analysis allows the characterization of direct and indirect PKD2-regu-  (58), Tbc1d5 (59), and Rin3 (60), or Sorting nexin 2 (Snx2) (59), which contains a phosphoinositide binding domain. Notably, the relocalization of intracellular vesicles containing signaling molecules has been shown to be required for antigen receptor signal propagation in T cells (61,62). Future experiments will determine if PKD2-regulated phosphorylations influence the amplitude, location, and duration of T-cell signaling by controlling intracellular vesicle trafficking. Interestingly, among the PKD2 substrate candidates we also find GEF-H1 (Arhgef2), a microtubule-associated guanine nucleotide exchange factor whose function has been recently shown to be crucial for antiviral host defenses (63). In the context of phosphorylation and ubiquitylation crosstalk mentioned above, our study identified the E2 ubiquitin conju-gating enzyme Ube2o which has a crucial role in endosomal protein trafficking (64) as novel PKD2 substrate in CTLs. Moreover the identification of the E3 ligase c-Cbl as a PKD2 substrate is also interesting, because c-Cbl is known to regulate the function and intracellular trafficking of several molecules implicated in antigen receptor signaling (50). Future studies into the role of PKD2 in regulating endosomal protein trafficking in CTLs may thus be interest. Previous studies have shown that PKD2 has a critical role to control the T-cell transcriptional program. Antigen stimulation in CD8 T cells induces changes in the expression levels of ϳ2600 annotated genes and PKD2 both positively and negatively regulates expression of 5% this transcriptional program (5). Here we have identified PKD2 substrate candidates implicated in regulation of transcription and translation such as the Lysine (K)-specific demethylase 2a (Kdm2a) (51), the transcription factor NFAT3 (52) and Phax (53), a protein that controls mRNA export from the nucleus. However, the diversity of the PKD2 controlled phosphoproteome identified herein affords an explanation for the broad role of PKD2 as a regulator of the T-cell transcriptional program. Thus, there is no evidence for a simple linear pathway of phosphorylation of a single substrate that ex- plains PKD2 action. Rather, the impact of PKD2 loss will thus result from a complex interplay between the direct and indirect effects of PKD2 loss on the phosphoproteome that are exposed by the present work.
A small number of the PKD2 substrates identified herein have been shown to be substrates for PKD family kinases in other cell lineages (Hip1r, Map4, Rdbp, Ssr3, Snx2, and Pcbp1). However, it is of equal importance to note that some of the key PKD substrates identified in experiments in fibroblasts, endothelial, and epithelial cells are not substrates in T cells. HDAC7 is thus a PKD substrate in many cells but not in CTLs, the protein phosphatase Slingshot (SSH1) is phosphor-ylated on Ser929 by PKD in HeLa and breast adenocarcinoma cell lines (6,7) whereas the present data set revealed that SSH1 was phosphorylated on Ser929 in CTLs but this phosphorylation was not reduced on PKD2 null CTLs (supplemental Table S1). In the context of intracellular protein transport and Golgi function, other well described PKD substrates are the phosphatidylinositol-4 kinase III beta; a regulator of Golgi vesicle fission and protein secretion (9) and the lipid and sterol transfer proteins CERT and OSBP (10,11). The present data identified eight unique phosphosites on phosphatidylinositol-4 kinase III beta (Pi4kb) including the Ser294, the PKD site previously characterized by Hausser and col- leagues in fibroblast (9). We found that there was no reproducible loss of the phosphorylation of Phosphatidylinositol-4 kinase III beta in PKD2 null cells. Similarly, the ceramide transfer protein CERT (Col4a3bp) has been shown to be phosphorylated on Ser132 by PKD, thereby reducing its ceramide transfer activity and placing PKD as a regulator of lipid homeostasis. The phosphorylation of CERT of Ser132 was not reproducibly lost in PKD2 null CTLs making this an unlikely PKD2 substrate in CTLs. Collectively, the data presented here show that PKD family kinases have both common and unique functions in different cell lineages. One explanation for this variability in substrates for different PKD isoforms could be that PKDs show cell lineage specific patterns of intracellular localization. For example in HEK293 and Hela cells PKD isoforms localize to the trans-Golgi network, where PKD phosphorylates substrates such as phosphatidylinositol-4 kinase III beta (9), CERT, and OSBP (10,11). PKD1 has been also shown to be recruited to the lamellipodium in HEK293 cells and pancreatic tumor cell lines, where PKD phosphorylates substrates such as SSH1 (6,7), cortactin (65), or Rin1 (8). In contrast, active PKD is mainly cytosolic in T cells (2). PKD is briefly recruited to the plasma membrane after antigen receptor stimulation but rapidly returns to the cytosol where it remains active (2). Thus, the cytosolic localization of PKD2 in T cells may prevent the access to substrates in particular cell compartments such as the Golgi network or the plasma membrane.