Dynamic Interaction- and Phospho-Proteomics Reveal Lck as a Major Signaling Hub of CD147 in T Cells

CD147 is a ubiquitously expressed type 1 transmembrane glycoprotein belonging to the Ig superfamily. It is therefore not surprising that it is involved in a plethora of cell functions such as cell proliferation (1–15), cell death (11, 12, 16–22), stress response (17, 23), chemosensitivity (10, 17, 21, 24, 25), chemotaxis (9, 26–34), migration (1, 2, 10–13, 18, 33, 35–41), adhesion (1, 6, 14, 33, 35, 42–45), and metabolism (3, 5, 16). Concomitantly, the effect of CD147 on these cellular functions has implications in the regulation of T cell activity (1, 4, 6–9, 14, 15, 34, 46–48), cancer (5, 13, 49–51), and metastasis (37–41, 52), and makes CD147 an ideal disease marker and therapy target (49, 53).

In using CD147 as a therapy target, it is essential to understand the possible signaling outcomes, and several attempts have been made to understand CD147-dependent molecular mechanisms. The interaction of CD147 with transporters, such as the monocarboxylate transporters 1 and 4 (MCT1 and MCT4) (3, 5, 16) or the amino acid transporter CD98 H chain (CD98) (54, 55), has been shown to affect cell growth and survival by modulating intracellular lactate or amino acid concentrations and stimulating PI3K/PKB (3, 55). The activation of PI3K/PKB and subsequent activation of the MAPK pathway upon engagement of CD147 with cyclophilins has also been observed to support cell proliferation (56, 57), stress protection (23–25), and chemotactic processes (9, 26–32, 43, 58). Moreover, activation of PI3K (59) and ERK (36) were found to be prerequisites for the CD147/β1-integrin–dependent modulation of FAK activation and the adhesive capacities of cells. Activation of FAK and MAPKs and the cytoskeletal regulators vinculin and paxillin were shown to be involved in subsequent CD147-dependent cytoskeletal changes in actin, microtubule, and vimentin filaments (60, 61). In addition to cytoskeletal processes, CD147 has also been observed to protect cells from apoptosis by regulating caspase-3 (19, 20, 22), BIM (20, 22), Bcl-2 (20), and XIAP (21), an effect that is at least partially dependent on MAPK activation (22). A possible link between CD147 and these pro- and anti-apoptotic regulators might lie in the CD147-dependent activation of FAK and Src, which enhances heat shock 70 kDa protein 5 (HSPA5) promoter activity via the phosphorylation of TFII-I (17).

In addition, CD147 also plays a crucial role in T cell biology: it has been shown to affect thymocyte development (62) and T cell proliferation (1, 4, 6–8, 14, 15), which was ascribed to modulated expression levels of IL-2 (7) and its receptor CD25 (1, 7, 47, 63). Further, two studies have reported reduced phosphorylation of early TCR signaling components and reduced calcium mobilization upon CD147 silencing (34) or CD147 Ab treatment (7), whereas in another study enhanced JNK and p21-activated kinase 1 activation and thus higher NFAT transcriptional activity in CD147-silenced cells has been reported (48). We showed recently that interaction of CD147 with the plasma membrane calcium ATPase isoform 4 (PMCA4) bypasses TCR proximal signaling and inhibits IL-2 expression (64). Nevertheless, the understanding of the molecular functionality of CD147 is still incomplete. In particular, not all the direct interaction partners and first-line signal transducers are understood, which might generate pitfalls for several therapeutic strategies. In addition, a huge bias has been generated in earlier studies, because major findings are limited and have focused on signaling routes for which specific Abs for Western blot analysis were available. The present study combines two proteomic approaches and analysis of the in silico protein network database to find hidden molecules in the CD147-dependent signaling system and to scrutinize the major signaling hubs.

CD147 mAb MEM-M6/1 was provided by Vaclav Horejsi (Institute of Molecular Genetics, Academy of Sciences of the Czech Republic, Prague, Czech Republic); mAb C305 to the TCR was a gift from Arthur Weiss (University of California, San Francisco, CA). The rabbit anti-phospho-Src pan-Tyr416, the rabbit anti–phospho–lymphocyte-specific protein tyrosine kinase (Lck) Tyr505, the rabbit anti-phospho–protein kinase C (PKC) pan-βII Ser660, the rabbit anti-phospho-PKC α/βII Thr638/641, the rabbit anti-phospho-PKC δ/θ Ser643/676, the mouse anti-rabbit mAb (clone 4H1), and the rabbit anti-GAPDH mAb (clone 14c10) were purchased from Cell Signaling Technology (Danvers, MA). Anti-Lck polyclonal rabbit Ab (clone H-95) was from Santa Cruz Biotechnology (Dallas, TX). The goat anti-rabbit IgG-HRP conjugates were purchased from Cell Signaling Technology and Rockland (Limerick, PA) and the goat anti-mouse IgG-HRP conjugate was from Sigma-Aldrich (St. Louis, MO).

PMA and ionomycin calcium salt (ionomycin) from Streptomyces conglobatus were purchased from Sigma-Aldrich. The protease inhibitor mixture Complete, Mini, EDTA-free and the phosphatase inhibitor mixture PhosSTOP were from Roche (Basel, Switzerland). N-Dodecyl β-d-maltoside (lauryl-maltoside) and Nonidet P-40 were from Thermo Fisher Scientific (Waltham, MA).

The human leukemic T cell line Jurkat E6.1 and the human embryonic kidney cell line HEK 293T were obtained from the American Type Culture Collection (Manassas, VA). Cells were cultivated as previously described (65). For stable isotope labeling with amino acids in cell culture (SILAC), cells were fed with RPMI 1640 containing [¹³C₆]l-arginine/D₄ l-lysine (R6K4) or [¹³C₆,¹⁵N₄]l-arginine/[¹³C₆,¹⁵N₂]l-lysine (R10K8), respectively (Cambridge Isotope Laboratories, Tewksbury, MA), and supplemented with 10% dialyzed FCS (Invitrogen, Carlsbad, CA).

The lentiviral packaging plasmids psPAX and pMD2.G were from Addgene (Cambridge, MA); the pLKO.1-puro-nonmammalian short hairpin RNA (shRNA) control (shControl) was supplied by the German Science Center for Genome Research (Berlin, Germany). The pLKO.1-puro-CD147_333 (shCD147) containing an shRNA construct specific for human CD147 and the pBMN-IRES-GFP-HACD147etc containing the RNA interference (RNAi)-resistant full-length CD147 with a hemagglutinin (HA)-tag (HACD147etc) were cloned as described previously (64).

Viral transduction was used for gene delivery, as documented elsewhere (66).

Jurkat T cells were stimulated with either 16.2 nM PMA plus 1 μM ionomycin or a 1:100 diluted hybridoma supernatant containing mAb C305 against the T cell Ag receptor at 37°C in 5% CO₂ and a humidified atmosphere.

Cells were lysed in lysis buffer (50 mM HEPES, 150 mM NaCl, 1 μM PMSF, 1 mM sodium orthovanadate, 50 mM NaF, 0.5% lauryl-maltoside, and protease inhibitor mixture). The lysate was incubated with agarose coated with anti-HA mAb (Sigma-Aldrich). The proteins were eluted with urea buffer (6 M urea, 2 M thiourea, 10 mM HEPES, pH 8), reduced with 1 mM DTT, alkylated with 5.5 mM iodoacetamide and digested in-solution for 3 h with LysC (1 μg/50 μg of protein). After four times dilution with 50 mM ammonium bicarbonate, the proteins were digested overnight with 1 μg trypsin/50 μg protein (Promega, Fitchburg, WI) at room temperature. The next day samples were desalted on a C18 stage tip.

For the phosphopeptide enrichment experiments, cells were lysed in extraction buffer (4% SDS, 0.1 M Tris/HCl, pH 7.6, protease and phosphatase inhibitor mixture), and the lysate was boiled for 5 min at 95°C and sonicated. The protein concentration was estimated by measuring the tryptophan content with the Infinite 200 (Tecan, Männedorf, Switzerland) at 280 nm and the same protein amounts of the shControl and shCD147 lysate were pooled. Proteins (5–6 mg) were loaded with 8 M urea on to an Amicon Ultracell 30k by centrifugation for 30 min at 2500 × g. The proteins were then washed three times with 8 M urea and incubated with 0.05 M iodoacetamide in 8 M urea for 20 min. Thereafter, the proteins were washed twice with 8 M urea and twice with digest buffer (10% acetonitrile, 0.1 M Tris/HCl, pH 8.5). Proteins were digested with 40 μg trypsin in digest buffer overnight at 37°C in a wet chamber. The next day, peptides were collected by centrifugation and the remaining protein in the Amicon cell was subjected to further digestion with 20 μg trypsin in digest buffer for 150 min before residual peptides were collected again by centrifugation. Lastly, 0.1% trifluoroacetic acid was added to the eluate and the peptide concentration was determined at 280 nm by the NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific).

The phosphopeptides were enriched with titanium oxide (TiO₂) beads as described by Sharma et al. (67). In short, phosphopeptides were isolated in six sequential isolation steps with the TiO₂ beads using the supernatant of the preceding pull-down for the subsequent pull-down as illustrated in Fig. 2A. For the first to fourth pull-downs, a peptide-to-bead mass ratio of 1:0.25 was used and for the fifth and sixth pull-downs the ratio was 1:4.

The peptides were desalted on a C18 stage tip. A chromatographic separation was performed using an Agilent 1200 nanoflow system (Thermo Electron, Bremen, Germany), coupled with a reversed-phase ReproSil-Pur C18-AQ 3 μm resin (Dr. Maisch, Ammerbuch-Entringen, Germany) in a 15-cm silica emitter (75 μm inner diameter; Proxeon Biosystems, Odense, Denmark). The flow rate of injection was set to 500 nl per min and peptides were eluted with a flow rate of 250 nl per min using a 100 min gradient of 2–40% acetonitrile in 0.5% acetic acid. Peptides were then analyzed with the LTQ Orbitrap mass spectrometer (Thermo Electron) containing an electrospray ion source (Proxeon Biosystems). The settings for the precursor ion analysis were: m/z 300–1800, a resolution of 60,000, and an ion accumulation to a target value of 1,000,000. The 10 most abundant ions were further fragmented and recorded in the ion trap and then dynamically excluded for 60 s. A lock mass option was enabled.

The data were processed with MaxQuant version 1.3.0.5 (68) with the integrated Andromeda search engine (69) and the UNIPROT protein index database for Homo sapiens with common contaminants added. Search parameters set to ≤3 allowed missed cleavages for the affinity-purification mass spectrometry (AP-MS) experiments, and ≤2 allowed missed cleavages for the phosphopeptide analysis, with cystein carbamidomethylation as the fixed modification and N-acetyl protein, oxidized methionine, [¹³C₆]l-arginine, D₄ l-lysine, [¹³C₆,¹⁵N₄]l-arginine, and [¹³C₆,¹⁵N₂]l-lysine as variable modifications. For the AP-MS experiments, the mass tolerance was set to 7 ppm for precursor ion peaks and to 0.5 kDa for the product ion peaks; for the phosphopeptide analysis, the mass tolerance was 6 and 20 ppm for the precursor and product ion peaks respectively. The false discovery rate was determined by searching a reverse database and the false discovery rate of 0.01 on peptide and protein level. A minimum peptide length of five residues was set for the AP-MS experiments and a minimum of six residues for the phosphopeptide analysis and one unique and a second identified peptide was set as the prerequisite for protein identification. The match between runs option with a time window of 2 min between replicates was enabled. Quantification, random error issues, and estimates of uncertainty were performed as detailed in Cox and Mann (68).

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (70) with the dataset identifier PXD002055 (https://www.ebi.ac.uk/pride/archive/).

The procedure described by Pfisterer et al. (71) was used for Western blot analysis. For analysis of protein phosphorylation, the cells were stimulated as described above or left untreated for the indicated time, then quickly washed with ice-cold stopping buffer (20 mM Tris-HCl, pH 7.5, 140 mM NaCl, 0.1 mM sodium orthovanadate), and lysed in lysis buffer (20 mM Tris-HCl, pH 7.5, 140 mM NaCl, 1 μM PMSF, 1 mM sodium orthovanadate, 20 mM NaF, 1% (v/v) Nonidet P-40, protease inhibitor mixture) on ice for 30 min. Lysates were precleared by brief centrifugation in a table-top centrifuge for 20 s and the protein concentration was determined using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). Samples of 10 μg were separated by SDS-PAGE, blotted onto polyvinylidene difluoride membranes, blocked, and probed with the respective phospho-specific Abs, and processed further as already described (71). To prevent protein dephosphorylation, the blocking buffer, Ab solutions, and the wash buffer were supplemented with 0.1 mM sodium orthovanadate. After the detection of the chemiluminescent signal, Abs were stripped off the membrane with acidic stripping buffer [0.2 M glycine, 0.1% (w/v) SDS, 1% (v/v) Tween-20, pH 2.2], washed extensively, then blocked and reprobed using mAbs recognizing phosphorylation-independent epitopes.

Volcano plots based on unpaired Student t tests were generated for statistical evaluation of interaction partners identified in AP-MS experiments from three biological replicates with the Perseus software package 1.5.2.6 (http://ww.perseus-framework.org/). The plots show the difference (the fold-intensity difference as LogN) of two conditions on the x-axis and the p value (as −log10) on the y-axis. Western blotting data were analyzed by two-way ANOVA with Bonferroni posttest using Prism 5 (GraphPad Software, La Jolla, CA). Significance was accepted at p < 0.05.

To examine how the microenvironment of CD147 changes during T cell stimulation, we employed AP-MS in the presence or absence of T cell stimulating agents. We retrovirally transferred HACD147etc as we described earlier (64) into CD147-silenced Jurkat T cells. Then we treated the cells with T cell stimulating agents for the indicated time points. The anti-HA Sepharose was used for affinity purification from HACD147etc-expressing cells, and wild-type Jurkat T cells were used as negative control. Purified proteins were subjected to LysC and trypsin digest in-solution and analyzed using electrospray ionization-liquid chromatography–tandem mass spectrometry. Volcano plots based on unpaired Student t tests showed significant (p ≤ 0.05) and strongly enriched (difference >2) CD147 interaction partners that have been previously described by us and others: MCT1 (54, 72, 73), CD98 (43, 54), MHC class 1A (HLA-A), the Na⁺/K⁺ ATPase α-1 subunit (ATP1A1) (54), PMCA4, and moesin (64). Moreover, calnexin, β-2-microglobulin (B2M), the guanine nucleotide binding protein (G protein) β polypeptide 1 (GNB1), and valosin-containing protein (VCP) were identified being part of the CD147 microenvironment. Upon stimulation, we found significant interaction of CD147 with the protein phosphatase tyrosine receptor type C (CD45), the G protein α inhibiting activity polypeptide 2 (GNAI2), HSPA5, Lck, VAT1, the integrin-associated signal transducer CD47 (CD47), and the Ras family small GTP binding protein (RAP1B) (Fig. 1). The numeric data are provided in Supplemental Table I. In Supplemental Fig. 1 we compare the intensity values of the coisolated molecules of the different stimulation conditions with each other side-by-side. Most interactions of CD147 were intensified after stimulation, and PMA plus ionomycin enforced the interactions with all interaction partners, except ATP1A1, calnexin, and moesin: ATP1A1 showed reduced interaction with CD147 upon stimulation with both mAb C305 and PMA plus ionomycin, and calnexin and moesin were reduced associated upon short-term stimulation with the mAb C305.

To elucidate CD147-dependent signaling pathways in unstimulated and stimulated Jurkat T cells, we followed the strategy illustrated in Fig. 2A. In short, Jurkat T cells were labeled using SILAC, either with R10K8 or with R6K4. After 10 d, the label incorporation tested by electrospray ionization-liquid chromatography–tandem mass spectrometry was above 95% for 99% of the identified peptides (Fig. 2B). Next, shRNA constructs targeting CD147 or a nontarget control construct were lentivirally transferred to the heavy R10K8-labeled or the light R6K4-labeled cells, respectively, and after 1 wk the silencing efficiency was estimated by Western blot analysis (Fig. 2C).

Then the CD147-silenced and shControl Jurkat T cells were either left unstimulated or separately stimulated for 1 min with TCR mAb C305, or for 5 and 30 min with PMA plus ionomycin. After lysis, equal protein amounts from the shControl and shCD147 samples were pooled and subsequently digested with trypsin. Phosphopeptides were isolated, prefractionated using the sequential phosphopeptide enrichment method with TiO₂ beads (67), and analyzed by electrospray ionization-liquid chromatography–tandem mass spectrometry. We thereby identified 1374 different phosphopeptide pairs of CD147-silenced cells and shControl cells from which a SILAC ratio was generated. Using the Perseus software package 1.5.2.6, we correlated the SILAC ratios of two independent experiments and assigned 76 phosphopeptides from 57 proteins that were consistently enriched or decreased by at least one-third in response to CD147 silencing (Fig. 3). The numeric data are provided in Supplemental Table II.

To determine which of the phosphopeptides were differentially phosphorylated in response to T cell stimulating agents and which changes are stimulation independent, we used the Perseus software package 1.5.2.6 Euclidean clustering function to generate a heat map showing the ratios of the 76 phosphopeptides (Fig. 4). Most CD147-dependent changes in phosphorylation were unaffected by T cell stimulation and thus stimulation independent. For instance, over all different stimulations, phosphosites associated with cell cycle regulations, such as T373, S788, S811, T821, T823, and T826 in the retinoblastoma 1 (RB1) protein and T14 in cyclin-dependent kinase 1 (CDK1) were phosphorylated to a greater degree in CD147-silenced cells. Cytoskeleton-associated molecules, such as the dedicator of cytokinesis 2 (DOCK2) at S1784, erythrocyte membrane protein band 4.1 (EPB41) at S85, microtubule-associated protein 1A (MAP1A) at S2056, and septin 2 (SEPT2) at S218 were less phosphorylated upon CD147 silencing. Similarly, Lck, which we found to associate with CD147 upon stimulation (Fig. 1), and protein kinase C β (PKCβ) were persistently less phosphorylated at Y394 and S660/T641, respectively.

In contrast to the stimulation-independent changes, we also detected CD147-dependent phosphopeptides only at specific time points of stimulation (Fig. 4, highlighted in gray). These dynamic stimulation-dependent phosphopeptides were further analyzed for peptide intensity levels and those with intensities >10⁶ are indicated with asterisks: phosphorylation of reticulon 4 (RTN4) at S184 was only enhanced in unstimulated CD147-silenced cells and was not detected any more upon stimulation. Dynamic phosphorylations at S291 in sialophorin (CD43) and at S2358 in spectrin β nonerythrocytic 1 (SPTBN1) were gradually enhanced by PMA plus ionomycin stimulation in CD147-silenced cells. Similarly, the phosphosite T679 on the Rho/Rac guanine nucleotide exchange factor 2 (ARHGEF2) was increased after 30 min PMA plus ionomycin treatment. Phosphorylation at S575 in the stromal interaction molecule 1 (STIM1) was decreased in unstimulated CD147-silenced cells and was not detected upon short-term stimulation. The decreased phosphorylations at S1938 in myosin-10 (MYH10) and at T241 in the ribosomal protein S6 (RPS6) were only reproducibly detected upon 5 min PMA plus ionomycin stimulation.

To confirm CD147-dependent changes in protein phosphorylation by an independent approach, we stimulated CD147-silenced and shControl Jurkat T cells with PMA plus ionomycin for 5 and 30 min (or left them untreated), and analyzed some of the detected phosphosites by Western blotting using specific Abs (Fig. 5). We consistently detected lower phosphorylation of Lck at Y394 and of PKCβ at S660 in CD147-silenced cells (Fig. 5). Similarly, phosphorylation of PKC α/βII at T638/641 was lower in shCD147 cells. We also tested phosphorylation of PKCθ, which is an important player in TCR-downstream signaling events and is involved in CD43 phosphorylation (74). However, using the phospho-PKCδ/θ Ab recognizing S643/676, we found no significant effect caused by CD147 silencing on this phosphorylation site, though the level of phosphorylation was also mildly decreased compared with the levels found in shControl cells (Fig. 5). Moreover, it should be noted that stimulation with PMA did not affect phosphorylation of PKCβ at S660 and T641 or PKCθ at S676 in general, although the activity of both of these PKC isoforms was described to be induced by PMA (75). Interestingly, we found decreased phosphorylation at the inhibitory Y505 site of Lck as well as reduced protein levels of total Lck in CD147-silenced cells. Further, the protein levels of RB1 were rather increased (Fig. 5), which might give rise to the higher levels of its phosphorylated versions. Thus, some of the differentially phosphorylated peptides in CD147-silenced versus shControl cells might be caused by downregulation or upregulation of the specific proteins in the CD147-silenced cells.

To get a comprehensive picture of the interaction and signaling routes of CD147, we linked the microenvironment of CD147 with the CD147-dependent phosphoproteome using STRING (76) analysis (Fig. 6A). Direct connections of proven protein-protein interactions indicated by magenta lines between interaction partners and signaling targets of CD147 were found with the following molecules: RAP1B with the RAP1 GTPase activating protein 2 (RAP1GAP2); RAP1B and GNAI2 with the leucine-rich repeats and WD repeat domain containing protein 1 (LRWD1); RAP1B, GNB1, Lck and HSPA5 with p21-activated kinase 2 (PAK2); GNB1 and Lck with PKCβ; HSPA5 with SPTBN1 and MYH10; moesin with CD43; VCP and HSPA5 with CDK1; VCP with the ubiquitin fusion degradation protein 1 homolog (UFD1L); PMCA4 with STIM1 and ATP1A1 with cofilin 1 (CFL1). Thus, the G proteins RAP1B and GNB1, the kinases PKCβ, PAK2, Lck, and CDK1, and the chaperone HSPA5 appear as major signaling hubs connecting several of the identified interaction partners and signaling targets of CD147.

On the basis of STRING database gene ontology biological process (GOBP) term enrichment analysis (shown in the table of Fig. 6B and indicated in Fig. 6A by the colored circles), we found the terms associated with wound healing as the highest enriched terms in the combined analysis of CD147 interaction partners and CD147-dependent phosphoproteins. Other enriched GOBP terms include cytoskeleton-, immune system–, stress response–, phosphorylation- and protein modification–, defense response– (particularly to virus), and TNF production–associated terms. The links between proteins in the same GOBP term group indicate possible CD147-dependent signaling routes for the respective biological processes. For instance, interaction partners GNAI2, RAP1B, GNB1, Lck, CD47, HSPA5, CD98, MCT1, and PMCA4, and phosphoproteins PKCβ, MYH10, CD43, CFL1, and STIM1 of CD147 are all linked according to the STRING database to the GOBP term wound healing (orange). Proteins enriched in the cytoskeleton-associated terms (green) are the CD147 interaction partners CD47, Lck, CD45, B2M, moesin, CD98, and MCT1, and the CD147-dependent phosphoproteins reticulon 4 (RTN4), DOCK2, MYH10, PAK2, SPTBN1, RPS6, the bridging integrator 2 (BIN2), CDK1, CD43, the nuclear receptor corepressor 1 (NCOR1), and CFL1. The STRING database also found connections to the GOBP terms associated with the immune system (violet) for the CD147 interaction partners CD47, Lck, CD45, B2M, and HLA-A, and the CD147-dependent phosphoproteins MAP1A, the platelet and T cell activation Ag 1 (CD226), the SAM and SH3 domain containing protein 3 (SASH3), the apoptotic chromatin condensation inducer 1 (ACIN1), RPS6, DOCK2, PAK2, PKCβ, the heterogeneous nuclear ribonucleoprotein K (HNRNPK), CDK1, RB1, CD43, NCOR1, and CFL1. Proteins linked with stress response-associated terms (gray) were the CD147 interaction partners RAP1B, GNAI2, CD47, HSPA5, Lck, CD45, B2M, HLA-A, VCP, CD98, MCT1, and PMCA4, and the CD147-dependent phosphoproteins MAP1A, ARHGEF2, ACIN1, the transformer 2 β homolog (Drosophila) (TRA2B), the sperm associated Ag 9 (SPAG9), RCSD domain containing protein 1 (RCSD1), RAD18 E3 ubiquitin protein ligase, CD226, the cerebral cavernous malformations 2 protein (CCM2), DOCK2, the programmed cell death 4 (PDCD4), MYH10, PAK2, PKCβ, CDK1, CD43, NCOR1, STIM1, and CFL1. Phosphorylation- and protein modification–associated terms (red) were found with the CD147 interaction partners RAP1B, GNAI2, GNB1, HSPA5, Lck, CD45, B2M, VCP, CD98, and PMCA4, and the CD147-dependent phosphoproteins ARHGEF2, Sorbin and SH3 domain containing protein 3 (SORBS3), PDCD4, PAK2, RPS6, HNRNPK, CDK1, SPAG9, RB1, CD43, and NCOR1. Proteins assigned to the defense response (blue) include the interaction partners Lck, HLA-A, B2M, and CD47, and the CD147 phosphoproteins MAP1A, ACIN1, PAK2, DOCK2, CDK1, CD43, and CD226. The TNF production-associated terms (yellow) involve the CD147-dependent phosphoproteins ARHGEF2, SASH3, nucleolin, and CD43.

A number of reports and data indicate that the Ig family member CD147 fulfills a plethora of functions in immune cells, particularly in T cells; functions that are typically hijacked by cancer cells. However, until now few direct links between function and physically or functionally associated signaling partners of CD147 have been recognized. Using AP-MS and various T cell stimulation strategies, we found the previously identified interaction partners ATP1A1 (54), CD98 (43, 54), HLA-A (54), MCT1 (54, 72, 73), moesin, and PMCA4 (64), furthermore novel interaction partners in B2M, calnexin, GNB1, and VCP, and CD45, CD47, GNAI2, HSPA5, Lck, RAP1B, and VAT1 as novel interaction partners that dynamically interact with CD147 upon T cell activation. With the exception of HSP5A, all interactions were enforced or decreased with T cell stimulation: in particular, the increased interaction between PMCA4 and CD147 is interesting, as we reported recently that via this interaction CD147 inhibits IL-2 production and the calcium-exporting function of PMCA4 (64). Interestingly, we also found a CD147- and stimulation-dependent phosphorylation site in STIM1. Based on our findings and as CD147 translocates to the T cell synapse (7), and STIM1 and PMCA4 were shown to regulate localized calcium fluxes there (77, 78), we hypothesize that the dynamic interaction with CD147 plays a role in the timely regulation of the localized activity of PMCA4. Additionally, both CD147 and PMCA4 possess an ezrin/radixin/moesin binding site, the potential association site with the CD147 interaction partner moesin, and PMCA4 also possesses cytoskeleton regulatory activity (79–81). Thus, CD147 might directly and indirectly affect cytoskeletal mechanisms and localized signaling.

Using phosphopeptide enrichment in CD147-silenced cells, we found 76 CD147-dependent phosphosites. The majority of the detected CD147-dependent phosphorylation sites are not functionally characterized and phospho-specific Abs are not yet available; however, there are some exceptions: the phosphorylation of Lck at Y394 is essential for its kinase activity (82), and T641 and S660 of PKCβ are necessary for its kinase activity and subcellular localization respectively (83, 84). All three phosphosites were constantly reduced in CD147-silenced T cells, and thus we conclude that CD147 promotes the activation of PKCβ and Lck. The impact of CD147 on phosphorylation of Lck at Y394 and on Src at Y416 was also reported recently by others (34, 85). We further confirm these data using Western blotting by finding the hypophosphorylated Lck Y394 residue in CD147-silenced cells. Moreover, we also found that the phosphorylation of Y505 and the total Lck levels were decreased in CD147-silenced cells. These new data about Lck and PKC support our previous report (64), and that from Guo et al. (34), showing that early TCR signaling components are not upregulated in CD147-silenced Jurkat T cells but are rather reduced. We speculate that the stimulation-induced interaction of CD147 with both Lck and Lck-regulating phosphatase CD45 might have localized functional consequences in the phosphorylation status and thus activity of Lck.

Another functionally studied phosphosite is T679 on ARHGEF2, an activating phosphorylation (86) leading to binding of Rho and inducing cytoskeletal rearrangements (87). ARHGEF2 is associated with microtubules and is released and activated after their depolymerization, inducing stress fiber formation via the RhoA/ROCK/MLC signaling module (88). In contrast to the previously discussed phosphosites, the phosphorylation of ARHGEF2 at T679 depends on PMA plus ionomycin stimulation, just as for the functionally unknown phosphorylation sites S291 in CD43 or S2358 in SPTBN1. These three phosphorylation events show similar dynamics; therefore, we speculate that they act in concert to transduce CD147-dependent signals to the cytoskeleton, thereby modulating the localization of signaling complexes during T cell synapse formation. Another CD147-dependent and PMA plus ionomycin stimulation–dependent phosphorylation was found in the site S1938 in MYH10. This site, in concert with neighboring sites, was recently described to regulate polarization of migrating cells (89) and might be also important for CD147-dependent cytoskeletal reorganizations in response to T cell stimulation.

In addition to phosphorylations regulating cytoskeletal mechanisms, we also found CD147-dependent phosphorylations of cell growth regulating molecules. We observed CD147-dependent phosphorylation of RPS6 at serines 235 and 236, which are sites regulating cap-dependent translation (90). Moreover, we found dynamic CD147-dependent phosphorylation of RPS6 at the functionally uncharacterized site T241. Further there were increased CD147-dependent phosphorylations of cell cycle regulating sites T14 in CDK1 and T373, S788, S811, T821, T823, T826, and S855 in RB1. Thus, these phosphosites might be important in another major function of CD147: the regulation of cell growth.

GOBP term analysis by the STRING database revealed a strong enrichment of wound healing–, immune system–, stress response–, phosphorylation- and protein modification–, defense response–, particularly virus defense response–, and TNF production–associated terms in the combined dataset of identified CD147 interaction partners and CD147-dependent phosphoproteins. CD147 is known to play a role in all these biological processes and the reported analysis for known and predicted protein-protein interactions describes possible CD147-dependent signaling routes in these processes. A role for CD147 in the wound-healing process has been already described, but the interaction of CD147 with matrix metalloproteinases has been made primarily responsible for this function (91, 92). Based on the STRING database analysis and the lack of matrix metalloproteinases in the AP-MS approach, we hypothesize that CD147 is involved in the wound-healing process through additional interaction partners.

The most important signaling hubs in the CD147 signal network appear to be the G proteins RAP1B and GNB1, the kinases PKCβ, PAK2, Lck, and CDK1, and the chaperone HSPA5. Notably, among those, Lck might be the central hub because it was detected as a CD147 interaction partner, on CD147 silencing showed not only reduced phosphorylation on both its activating site Y394 and inhibitory site Y505 but also reduced protein expression, was strongly molecularly interlinked by the STRING database, and was associated with nearly all of the GOBP terms enriched in the CD147 microenvironment and signaling network. Thus, Lck might be of great importance in CD147 signaling and a good candidate for further detailed analysis.

In summary, we connected known interaction partners of CD147 and several novel ones that dynamically interact upon T cell stimulation as well as proteins whose phosphorylation depend on CD147 into a molecular network. We also found signaling groups by GOBP term analysis and correlated trends in association and phosphorylation dynamics to generate evidence for their functional association. Thereby we provide a great deal of interesting CD147-dependent proteins and phosphorylation sites, most of those to date undescribed, as starting point for further detailed analysis.

We thank Herbert Schiller, Rochelle D’Souza, and Kirti Sharma (Department of Proteomics and Signal Transduction, Max Planck Institute of Biochemistry, Martinsried, Germany) for valuable technical advice and discussion; Reinhard Fässler (Department of Molecular Medicine, Max Planck Institute of Biochemistry, Martinsried, Germany) for experimental and intellectual support with mass spectrometric analysis; Vaclav Horejsi (Institute of Molecular Genetics, Academy of Sciences of the Czech Republic, Prague, Czech Republic) for providing the CD147 mAb MEM-M6/1; Arthur Weiss (University of California, San Francisco, CA) for providing the TCR mAb C305; Sheila Stewart (Washington University School of Medicine, St. Louis, MO) for providing the lentiviral shRNA expression vector pLKO.1-puro; Giulio Superti-Furga (Center for Molecular Medicine, Austrian Academy of Sciences, Vienna, Austria) for providing the pfMSCV Strep3xHA plasmid; and Gary Nolan (Stanford University School of Medicine, Stanford, CA) for providing the retroviral expression vector pBMN-IRES-GFP.

The authors have no financial conflicts of interest.