Molecular Basis of the Interaction of the Human Protein Tyrosine Phosphatase Non-receptor Type 4 (PTPN4) with the Mitogen-activated Protein Kinase p38γ*

The human protein tyrosine phosphatase non-receptor type 4 (PTPN4) prevents cell death induction in neuroblastoma and glioblastoma cell lines in a PDZ·PDZ binding motifs-dependent manner, but the cellular partners of PTPN4 involved in cell protection are unknown. Here, we described the mitogen-activated protein kinase p38γ as a cellular partner of PTPN4. The main contribution to the p38γ·PTPN4 complex formation is the tight interaction between the C terminus of p38γ and the PDZ domain of PTPN4. We solved the crystal structure of the PDZ domain of PTPN4 bound to the p38γ C terminus. We identified the molecular basis of recognition of the C-terminal sequence of p38γ that displays the highest affinity among all endogenous partners of PTPN4. We showed that the p38γ C terminus is also an efficient inducer of cell death after its intracellular delivery. In addition to recruiting the kinase, the binding of the C-terminal sequence of p38γ to PTPN4 abolishes the catalytic autoinhibition of PTPN4 and thus activates the phosphatase, which can efficiently dephosphorylate the activation loop of p38γ. We presume that the p38γ·PTPN4 interaction promotes cellular signaling, preventing cell death induction.

The human protein tyrosine phosphatase non-receptor type 4 (PTPN4) prevents cell death induction in neuroblastoma and glioblastoma cell lines in a PDZ⅐PDZ binding motifs-dependent manner, but the cellular partners of PTPN4 involved in cell protection are unknown. Here, we described the mitogen-activated protein kinase p38␥ as a cellular partner of PTPN4. The main contribution to the p38␥⅐PTPN4 complex formation is the tight interaction between the C terminus of p38␥ and the PDZ domain of PTPN4. We solved the crystal structure of the PDZ domain of PTPN4 bound to the p38␥ C terminus. We identified the molecular basis of recognition of the C-terminal sequence of p38␥ that displays the highest affinity among all endogenous partners of PTPN4. We showed that the p38␥ C terminus is also an efficient inducer of cell death after its intracellular delivery. In addition to recruiting the kinase, the binding of the C-terminal sequence of p38␥ to PTPN4 abolishes the catalytic autoinhibition of PTPN4 and thus activates the phosphatase, which can efficiently dephosphorylate the activation loop of p38␥. We presume that the p38␥⅐PTPN4 interaction promotes cellular signaling, preventing cell death induction.
PTPN4 5 is a non-receptor-tyrosine phosphatase (PTP) with functions in T cell signaling, learning, spatial memory, and cerebellar synaptic plasticity (1)(2)(3). Its overexpression reduces cell proliferation in COS-7 cells and suppresses CrkI-mediated cell growth and mobility in HEK293T cells (4,5). PTPN4 also regulates neuronal cell homeostasis by protecting neurons against apoptosis (5,6). PTPN4 is a large modular protein containing three domains, an N-terminal FERM domain, a PDZ domain, and a C-terminal catalytic tyrosine phosphatase domain (7). PDZ domains are protein-protein interaction domains, which play a central role in cell signaling by favoring spatial contacts between enzymes and their substrates or, more generally, by assembling and/or regulating protein networks (8,9). Thus, disrupting the interactions between PDZ domains and PDZ binding motifs (PBM) can trigger profound alterations in cell signaling pathways (10 -12). Such disruptions of PDZ⅐PBM complexes are used by viruses to hijack cell signaling pathways to their own advantage and can also be obtained by treating cells with an excess of cell penetrating peptide encoding a PBM (13). Indeed, we showed that rabies virus (RABV) peptides encoding a PBM can target the PDZ domain of PTPN4 (PTPN4-PDZ) and antagonize the function of PTPN4, leading to cell death (6). We optimized such pro-apoptotic peptides and designed the Cyto8-RETEV sequence, that is so far the most affine ligand of PTPN4-PDZ and the best inducer of cell death (15).
Four endogenous partners of PTPN4 have been reported: the glutamate receptor ␦2 and ⑀ subunits, the T-cell receptor subunit, the CrkI oncoprotein, and the TRIF (TIR domain-containing adaptor-inducing interferon-B)-related adaptor molecule TRAM (3,5,16,17). Nevertheless, the natural ligands of PTPN4 involved in the regulation of cell homeostasis remain unknown. Hou et al. (12) have identified p38␥, a mitogen-activated protein kinase (MAPK), as an endogenous partner of PTPN3, the closest homologous protein of PTPN4. MAPKs, including extracellular signal-regulated kinases (ERK), c-Jun N-terminal kinases, and p38s, are a major class of kinases involved in signaling cascades that mediate and integrate signals for a coordinated cellular response such as proliferation, transformation, and cell death (18). p38␥ is the only MAPK that encodes a PBM (KETPL) at its C terminus allowing interactions with PDZ domain-containing proteins. The RAS-dependent oncogenic activity of p38␥ results from the direct interaction between the PBM of p38␥ and the PDZ domain of PTPN3 (PTPN3-PDZ), which allows the catalytic domains of PTPN3 to dephosphory- late a phosphotyrosine residue on p38␥ (12). PTPN3 and PTPN4 are members of the NT5 non-receptor-tyrosine phosphatase group. They contain the same modular organization and have a high sequence identity of 71 and 65% for the PDZ and phosphatase domains, respectively. Thus, we surmised that p38␥ could be a ligand of PTPN4-PDZ involved in PTPN4 cell signaling. Interestingly, silencing PTPN4 or p38␥ genes is phenotypically equivalent and results in massive cell death (19). A similar phenotype is also observed when PTPN4-PDZ is targeted by peptides encoding a PBM after their intracellular delivery (15).
Here, we show that full-length p38␥ and PTPN4-PDZ interact in vitro in GST pulldown assays. The C-terminal sequence of p38␥ (p38␥-KETPL), encompassing the PBM, interacts tightly with PTPN4-PDZ and has the highest affinity of all endogenous PTPN4 ligands. NMR and x-ray structural data of the PTPN4-PDZ and p38␥-KETPL complex and of the PTPN4-PDZ and optimized peptide Cyto8-RETEV complex lead us to identify the structural basis for PBM recognition by PTPN4-PDZ. We also highlight the predominant role of this PDZ⅐PBM association in the molecular interactions between the fulllength p38␥ and the physiologically active PTPN4 two-domain (PDZ-PTP) encompassing the PDZ and the phosphatase domains. As a functional consequence, the binding of p38␥-KETPL to PTPN4-PDZ lifts the auto-inhibited state of PTPN4 and activates the phosphatase. Finally, the peptide p38␥-KETPL efficiently promotes death of human glioblastoma upon intracellular delivery.
GST Pulldown Assay-GST fusion construct containing the PDZ domain of PTPN4 (GST-PTPN4-PDZ) was expressed and purified as previously described (15) without TEV cleavage. GST was purified after an additional TEV cleavage step of GST-PTPN4-PDZ. Purified GST-PTPN4-PDZ or GST was individually incubated with glutathione-agarose beads for 1 h at 4°C with mild shaking. The beads were pelleted by centrifugation and washed 3 times with binding buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5 mM TCEP). Immobilized GST or GST-PTPN4-PDZ was incubated with mild shaking for 1 h at 4°C with an equivalent ratio (v/v) of beads:lysate of human neuroblastoma cells (SH-SY5Y cells), which were prepared as previously described (6). Glutathione-agarose was washed four times and resuspended in a SDS-PAGE sample buffer. After boiling, the bound proteins were analyzed using 12% SDS-PAGE followed by Western blotting using anti-p38␥ antibody (Cell Signaling, #2307) and Rabbit TrueBlot monoclonal antibody (Rockland). The proteins were visualized by enhanced chemiluminescence according to the manufacturer's instructions (Amersham Biosciences).
Crystallization, Data Collection, and Structure Determination-The Cyto8-RETEV and p38␥-KETPL peptides used for co-crystallizations were added in excess to form Ͼ95% of the complex with the protein. PDZ domain⅐ligand complexes for crystallization were generated by mixing PTPN4-PDZ domain and the appropriate peptide at a ratio of 1:2 and 1:4 for PTPN4-PDZ⅐Cyto8-RETEV and PTPN4-PDZ⅐p38␥-KETPL, respectively, and incubating the mixture for 30 min on ice before crystallization experiments. Crystals were then flash-cooled in liquid nitrogen using Paratone-paraffin (50%/50%) oil as the cryoprotectant. X-ray diffraction data were collected on beamline PROXIMA-1 at Synchrotron SOLEIL (St. Aubin, France). The data were processed with XDS (21), SCALA (22), and other programs from the CCP4 suite (23).
The structures were solved by molecular replacement with PHASER (24) using the search atomic model of PTPN4-PDZ (PDB access code 3NFK). The locations of the bound peptides were determined from a F o Ϫ F c difference electron density map. Models were rebuilt using COOT (25), and refinement was performed using BUSTER (26). The overall assessment of model quality was performed using MolProbity (27). The crystal parameters, data collection statistics, and final refinement statistics are shown in Table 1. All structural figures were generated with the PyMOL Molecular Graphics System, version 1.5.0.4 (Schrödinger).
The asymmetric unit of PTPN4-PDZ⅐Cyto8-RETEV and PTPN4-PDZ⅐p38␥-KETPL crystal structures contains four subunits of protein⅐peptide complex. Each subunit corresponds to the complex formed by PTPN4-PDZ protein (chains A, B, C, and D) with the peptides (chains E, F, G, and H), respectively, A-E, B-F, C-G, and D-H. The four PTPN4-PDZ subunits are disulfide-bonded by pair: chains A to B and chains C to D for PTPN4-PDZ⅐Cyto8-RETEV structure; chains A to D and chains B to C for PTPN4-PDZ⅐p38␥-KETPL structure. However, the monomeric state of the PDZ domain in complex with these peptides in solution was previously confirmed, implying that the intermolecular disulfide bonds are due to crystalline conditions (15) and (20).
For the PTPN4-PDZ⅐Cyto8-RETEV structure, only 6 C-terminal residues of the 13 residues of Cyto8-RETEV were well defined in the electronic density. For the PTPN4-PDZ⅐p38␥-KETPL structure, in two subunits (chains A-E and B-F), only six C-terminal residues of p38␥-KETPL were well defined into the electronic density, whereas due to crystal packing in the other subunits (chains C-G and D-H), all of the 11 residues of p38␥-KETPL in chain G and H were well ordered. The four PDZ domains from each subunit of the asymmetric unit superimpose very well with a root mean square deviation on C␣ of 0.17 Å and 0.33 Å for the PTPN4-PDZ⅐Cyto8-RETEV and PTPN4-PDZ⅐p38␥-KETPL complexes, respectively, indicating that crystal packing has no influence in the overall structure of the PDZ domain. In the following text, we described the PTPN4-PDZ⅐Cyto8-RETEV complex formed by the C and G chains and the PTPN4-PDZ⅐p38␥-KETPL complex formed by the D and H chains.
Peptide-induced Cell Death-Peptides were conjugated to the HIV-1 Tat domain and labeled with a FITC molecule as previously described (6). Human grade III U373MG glioblastoma cells (HTB 17; ATCC) (250,000 cells per well in 1 ml of culture medium) were treated with FITC-conjugated peptides at a concentration of 25 mM. Three hours post treatment, cells were assayed for peptide entry (FITC-positive cells), and cell death (propidium iodide-positive cells) was by flow cytometry (FACSCalibur, BD Biosciences) using the Cell Quest Pro software as described (15). M (1:1.2 molar ratio). All the 1 H, 15 N HSQC spectra recorded on these samples to map the PTPN4-PDZ⅐p38␥ interaction were acquired at 25°C on a Varian Inova 600-MHz spectrometer equipped with a cryoprobe, with 128 (60 ms) and 1024 (95 ms) complex points (acquisition times) in 15 N and 1 H, respectively.
The NMR titration experiment to measure the PTPN4-PDZ⅐p38␥-KETPL peptide affinity at 25°C was performed as previously described (6). Briefly, the unlabeled p38␥-KETPL peptide (stock solution of 6.5 mM) was added stepwise in a sample initially containing 280 l of 15  Phosphorylation of p38␥-To generate phosphorylated p38␥ for enzymatic assays, a p38␥ sample at 100 M was mixed at a molar ratio of 1:4000 of MKK6-EE⅐p38␥ in a buffer composed of 250 mM Tris-HCl, pH 7.5, 125 mM NaCl, 0.4 mM TCEP, 5 mM MgCl 2 , 5 mM MnCl2, 2 mM ATP. The mixture was incubated overnight at 22°C. After incubation, the samples were purified by an ion-exchange column (HiPrep Q HP) and then eluted with a NaCl gradient. p38␥ phosphorylation was controlled by electrospray ionization mass spectrometry.
Phosphatase Assay and Kinetic Analysis-Phosphatase activity with phosphopeptides and p-nitrophenyl phosphate was measured as previously described (20). Reactions were measured with various phosphopeptide pTGpY concentrations ranging from 1.  phospho-p38␥ or TGpY peptide in the final volume and in buffer conditions identical to that of the phosphopeptide assay.

A Direct Interaction between PTPN4-PDZ and Endogenous p38␥ Is Detected by GST Pulldown Assays Using Neuroblastoma Cell
Extracts-To identify a potential interaction between endogenous p38␥ and PTPN4 through their PBM and PDZ, respectively, we performed pulldown analyses using human neuroblastoma SH-SY5Y cell lysates. The presence of p38␥ was initially verified in cell lysate (Fig. 1A). Purified GST-PTPN4-PDZ and GST alone were incubated with the soluble fraction of SH-SY5Y cell lysates. As shown in Fig. 1A, endogenous p38␥ binds to GST-PTPN4-PDZ but not to GST alone (Fig. 1A). The direct interaction was also confirmed with purified recombinant p38␥ mixed with glutathione-agarose bound to GST-PTPN4-PDZ fusion proteins (Fig. 1A). Thus, the results dem-onstrate the direct interaction between endogenous p38␥ and PTPN4-PDZ in cell lysates and suggest that p38␥ is a physiological partner of PTPN4 through PDZ-mediated interaction.
PBM of p38␥ Has a High Affinity for PTPN4-PDZ-The affinity of the PBM of p38␥ (p38␥-KETPL peptide) for PTPN4-PDZ was measured by NMR titration. 1 H, 15 N chemical shift perturbations of the PDZ domain were followed as a function of increasing concentrations of the p38␥-KETPL peptide to determine the dissociation constant (K d ). The p38␥-KETPL peptide binds to PTPN4-PDZ with a K d of 1.58 Ϯ 0.97 M ( Table 2). This affinity is very close to that of the Cyto8-RETEV sequence (K d of 1.0 M), which we have previously designed on structural basis to optimize PTPN4-PDZ binding (15). The C-terminal sequence of p38␥ has the strongest affinity for PTPN4-PDZ compared with the two endogenous partners, GluN2A and GluD2, with K d values of 42 M and 128 M, respectively.
PBM of p38␥ and the Optimized Peptide Have a Similar Binding Mode to PTPN4-PDZ-To investigate the interaction of PTPN4-PDZ with the endogenous ligand p38␥-KETPL or the optimized ligand Cyto8-RETEV, the crystal structures of the two complexes were solved by molecular replacement at 2.35 Å and 2.09 Å resolutions, respectively ( Table 1). The structure factors and coordinates for the structures PTPN4/Cyto8-RETEV and PTPN4/p38␥-KETPL have been deposited in the Protein data Bank under accession codes 5EYZ and 5EZ0, respectively.
The structures of PTPN4-PDZ⅐Cyto8-RETEV and PTPN4-PDZ⅐p38␥-KETPL are very similar with a root mean square deviation of 0.22 Å. PTPN4-PDZ adopts the typical PDZ fold comprising five ␤-strands and two ␣-helices ( Fig. 2A) and possesses the interaction network specific to class I PDZ domains that recognize the consensus peptide sequence X(S/T) X⌽ COOH , where X is any residue, and ⌽ is a hydrophobic residue. The last six C-terminal residues encompassing the PBM of Cyto8-RETEV (-GRETEV -COOH ) and p38␥-KETPL (-SKETPL -COOH ) are well ordered in the crystal. The peptides were inserted into the PDZ-binding pocket in the conventional mode (Fig. 2). The interactions of the residues at the positions 0, Ϫ2, and Ϫ3 of both peptides with PTPN4-PDZ are quite similar to the bonding patterns already observed in the complex between PTPN4-PDZ and the viral peptide Cyto13-att (-GETRL -COOH ) (x-ray structure PDB code 3NFK; Ref. 15). The peptide Cyto13-att corresponds to the last 13 residues of the G envelope protein from an attenuated RABV strain. Starting from the Cyto13-att sequence that binds PTPN4-PDZ with a K d of 157 M, the Cyto8-RETEV sequence has been optimized focusing on the residues at positions Ϫ1 (Arg 3 Glu) and Ϫ4 (Gly 3 Arg). These positions are not involved in the network of interactions between the viral sequence and PTPN4-PDZ (15). Here, we confirm our previous postulate that Glu at position Ϫ1 and Arg at position Ϫ4 in Cyto8-RETEV likely induces crucial polar contacts with the surrounding residues of PTPN4-PDZ and consequently improve the affinity for PTPN4-PDZ.
The cationic side chain at position Ϫ1 in the viral peptide Cyto13-att (-GETRL -COOH ) is unfavorable in the positively charged environment of PTPN4-PDZ consisting of residues Arg-527, Arg-546, and Lys-587. The acidic residue (Glu) in Cyto8-RETEV and the neutral residue (Pro) in p38␥-KETPL at position Ϫ1 are fully exposed to the solvent. However, these residues are H-bonded to PTPN4-PDZ involving the oxygen atom of their peptidic bond and the H of the Arg-527 from the PTPN4-PDZ binding site GLGF loop (Fig. 2, B and C). The basic side chains at position Ϫ4, Arg for Cyto8-RETEV and Lys for p38␥-KETPL, form a H-bond with the C␥ carboxyl oxygen of the Asp-580 located at the bottom of the ␣-helix 2. These additional interactions could mainly explain the increase in binding free energy of complex formation with Cyto13-att compared with Cyto8-RETEV or p38␥-KETPL (⌬⌬G 0 ϭ RTln(K D (Cyto13-att)/K D (Cyto8-RETEV)) ϭ 2.8 kcal⅐mol Ϫ1 or ⌬⌬G 0 ϭ RTln(K D (Cyto13-att)/K D (p38␥-KETPL)) ϭ 2.7 kcal⅐mol Ϫ1 ) corresponding to a 160-fold and 100-fold difference in affinity, respectively. Thus, similar binding modes of p38␥-KETPL and Cyto8-RETEV with PTPN4-PDZ are found for the key positions of recognition between PBM and PDZ, explaining the similar strong affinity of p38␥-KETPL and Cyto8-RETEV for PTPN4-PDZ.

PBM of p38␥ Is an Efficient Inducer of Cell Death in Human Glioblastoma
Cell Line Expressing PTPN4 -The affinities of PBM peptides for PTPN4-PDZ are closely correlated to their killing efficiency after their intracellular delivery (15). We conceived the best-suited Cyto8-RETEV peptide that binds with the highest affinity to PTPN4-PDZ and consequently the best inducer of cell death. In the same way, the killing efficiency of the Tat-conjugated peptide of the C-terminal PBM of p38␥ was estimated in U373MG (Fig. 1B), a human glioblastoma cell line expressing PTPN4. After a 3-h exposure of U373MG to the peptide, a delay sufficient for maximal peptide entry, the induction of cell death was measured by propidium iodide assay. A 2-fold increase of glioblastoma cell death was observed using p38␥-KETPL compared with Cyto8-RETEV with a similar capacity to enter the cells for both peptides. Thus, p38␥-KETPL is an efficient inducer of cell death in glioblastoma.
Mapping the PTPN4-PDZ⅐p38␥ Interaction in Vitro-To investigate the molecular basis of the interaction between p38␥ and PTPN4, we probed in vitro the binding of PTPN4-PDZ to full-length p38␥ by NMR chemical shift perturbation experiments on both sides of the partners. 1 H, 15 N HSQC spectra were recorded on 15 N-labeled PTPN4-PDZ in its free form, in the presence of unlabeled p38␥-KETPL peptide and in the presence of unlabeled p38␥ full-length protein.
The addition of p38␥-KETPL peptide induces strong 1 H, 15 N chemical shift changes on PTPN4-PDZ (Fig. 3A). Residues on strands ␤2, ␤3, and helix ␣2 are the most affected by the binding (⌬␦ avg Ͼ 0.17 ppm) (Fig. 3, B and C), suggesting a conventional  anchoring mode of the PBM into the PDZ groove. Overall, the same chemical shift perturbations are observed by the addition of full-length p38␥ (Fig. 3, A and B). Four resonance peaks were severely affected by exchange broadening and not detected in the spectrum and corresponded to residues localized near the binding groove. Thus, the same surface was mapped on PTPN4-PDZ by the binding of the short p38␥-KETPL peptide or of the full-length p38␥ protein (Fig. 3C). We compared 1 H, 15 N HSQC spectra recorded on 15 N-labeled full-length p38␥ in its free form and in the presence of unlabeled PTPN4-PDZ to map the residues of the kinase perturbed by the interaction with PTPN4-PDZ (Fig. 4). Backbone assignment of p38␥ was performed on a 13 C, 2 H, 15 N-labeled sample using a set of three-dimensional heteronuclear NMR experiments recorded at 950 MHz. Over the 349 expected nonproline residues, 273 amide signals were visible in the HSQC spectrum and only 154 could be assigned (44.1%) (Fig. 4) (BMRB (Biological Magnetic Resonance Bank) accession number 26758). Note that similar results were previously obtained for the homologous MAP kinase p38␣ (28) for which 64% of amide signals were initially assigned using a set of selectively 15 N-labeled samples and distance information from a paramagnetic adenosine derivate. Later, the sequence-specific backbone assignment was improved to 82% (29). Our partial assignment of p38␥ is sufficient to unambiguously identify residues perturbed by the binding of PTPN4-PDZ as the C terminus residues from Gly-358 to Leu-367 COOH . In the presence of PTPN4-PDZ, the latter were essentially affected by exchange broadening, and this line broadening increased in a gradual manner toward the C terminus (from Gly-358 to Leu-367), up to the complete disappearance of Thr-365 and Leu-367 resonances (Fig. 4). These results indicate that full-length p38␥  15 N HSQC spectra recorded on free PTPN4-PDZ (black), on PTPN4-PDZ in complex with p38␥-KETPL peptide (blue), and on PTPN4-PDZ in complex with full-length p38␥ (red). The large chemical shift changes are indicated by dotted blue lines. Assignment is indicated for the most affected residues (⌬␦ avg Ͼ 0.17 ppm; see in B). B, 1 H, 15 N-weighted average chemical shift differences ⌬␦ avg between PTPN4-PDZ and PTPN4-PDZ⅐p38␥-KETPL complex (blue) and between PTPN4-PDZ and PTPN4-PDZ⅐p38␥ complex (red), calculated as ⌬␦ avg ϭ (⌬␦H 2 ϩ (⌬␦N ϫ 0.159) 2 ) 1/2 . The red asterisks denote residues for which no resonance peaks were detected in the PTPN4-PDZ⅐p38␥ complex due to exchange broadening. Secondary structure elements are indicated at the top. C, structure of PTPN4-PDZ (gray ribbon and space-fill representations) bound to p38␥-KETPL peptide (blue ribbon) showing the same binding surface of p38␥-KETPL peptide or full-length p38␥ onto PTPN4-PDZ (in violet), as identified by the large NMR chemical shift perturbations in B: ⌬␦ avg Ͼ 0.17 ppm. The p38␥ residues, the most affected by PTPN4-PDZ binding (Fig. 4), are shown in blue (Gly-358 -Leu-367 COOH ). The full-length p38␥ until the Ϫ7 position of the PBM is schematic.
binds PTPN4-PDZ via its C terminus PBM motif in a conventional PBM⅐PDZ manner and that no additional p38␥ regions are involved in the binding to PTPN4-PDZ as monitored by NMR.
The PBM of p38␥ Is the Major Contributor to the PTPN4⅐p38␥ Interaction-We established that p38␥ interacts tightly with PTPN4-PDZ via its PBM. Next, we studied the contribution of this PBM⅐PDZ interaction to the association between the PDZ-PTP WT two-domain of PTPN4 and the fulllength p38␥ by ITC. We noticed that p38␥-KETPL binds PDZ-PTP WT with an affinity (1.54 Ϯ 0.24 M) similar to that observed by NMR for PTPN4-PDZ (1.58 Ϯ 0.97 M). This indicates that the PTP domain does not affect the PDZ ligand binding. Full-length p38␥ binds PDZ-PTP WT with an affinity of 4.1 Ϯ 0.3 M that is slightly higher than the one calculated for the PDZ-PTP WT ⅐p38␥-KETPL association. In both cases, the interactions are mainly driven by an overall enthalpy contribution and with an unfavorable entropic penalty for the fulllength p38␥, most likely due to a decrease of conformational freedom of the two proteins in the presence of the kinase domain (Table 2). Then, we investigated the contribution of the PBM recognition on the PDZ-PTP WT ⅐p38␥ interaction. The addition of the p38␥-KETPL peptide in large excess (i.e. 95% of saturation of the PDZ binding site) abrogates the interaction between PTPN4-PDZ and p38␥ (Table 2, Fig. 5). The PBM⅐PDZ association is, therefore, the major contributor to the PDZ-PTP WT ⅐p38␥ interaction. However, a residual heat [⌬H Ϸ Ϫ4.1 kcal⅐mol Ϫ1 ] was observed at the end of the p38␥ titration or during the competition experiment. This enthalpy does not correspond to the heat of dilution (Fig. 5B, inset). This event could be related to a weak binding of PDZ/PBM off-site, which could involve the phosphatase domain and the kinase domain.
The Inhibition of the Catalytic Activity of PTPN4 by Its PDZ Domain Is Released when the PDZ Binding Site Is Occupied by p38␥-KETPL-We recently demonstrated that the PDZ domain of PTPN4 down-regulates the phosphatase activity through intramolecular interactions and that the mere binding of a PDZ ligand releases this catalytic inhibition (20). Here, we investigated whether the binding of p38␥-KETPL to the PDZ domain could also affect this catalytic regulation and in turn modulate the phosphatase activity. As we previously reported, the PDZ domain inhibits the catalytic activity of PTPN4 through a mixed inhibition. Indeed, the turnover number (k cat ) values decreased by 3.3-fold from linker-PTP to PDZ-PTP WT ,  15 N HSQC spectra recorded on free p38␥ (black) and on p38␥ in complex with PTPN4-PDZ (red). Assignment of p38␥ is indicated in black. The residues perturbed by PTPN4-PDZ binding (labeled in blue) are located at the C terminus (Gly-358 -Leu-367 COOH ) and are affected in a gradual manner toward the C terminus as schematically indicated at the top.
whereas the Michaelis constant (K m ) value decreased by a 2.2fold using p-nitrophenyl phosphate as substrate ( Table 3). The kinetic analysis of PDZ-PTP WT was then carried out in the presence of p38␥-KETPL peptide at a concentration sufficient enough to ensure the maximal effect of the PDZ ligand on the phosphatase activity. Both the k cat and K m values significantly increased in the presence of p38␥-KETPL, showing that the release of the inhibition operates with partial compensation of the effect on the two rate constants ( Table 3). The k cat and K m values are 2.3-fold and 2.2-fold higher, respectively, in the presence of p38␥-KETPL. Altogether, these data indicate that the PBM of p38␥ has the ability to modulate the phosphatase activity of PTPN4.
In addition, PTPN4 was assessed for its capacity to dephosphorylate the full-length p38␥. To this aim, p38␥ was produced and fully phosphorylated on the activation loop (TGpY) using in vitro MKK6 kinase. We measured the dephosphorylation of full-length phospho-p38␥ by the PTPN4 linker-PTP and PDZ-PTP WT in the presence or absence of an excess of p38␥-KETPL peptide (final concentration 40 M) at 30 M phospho-p38␥ (Fig. 6A). We performed the same experiments with the TGpY peptide as a substrate to compare the activation coming from the presence of the PBM on the full-length p38␥. We observed that the full-length p38␥ kinase was efficiently dephosphorylated by PTPN4 linker-PTP and PDZ-PTP WT with or without an excess of p38␥-KETPL peptide (Fig. 6A). The initial rates of linker-PTP and PDZ-PTP WT with p38␥-KETPL were similar, whereas the initial rate of PDZ-PTP WT alone was slightly lower (Fig. 6A). The relative amplitude between PDZ-PTP WT alone and PDZ-PTP WT with excess of peptide is significantly reduced with the full-length p38␥ as a substrate (ratio of initial rates of PDZ-PTP WT ϩKETPL/PDZ-PTP WT of 1.31) (Fig. 6A) compared with the one with the TGpY peptide as a substrate (ratio of 2.60) (Fig. 6B). This strongly suggests that the presence of the PBM on the full-length p38␥ already releases the inhibition of the PDZ domain of PTPN4 by the PDZ⅐PBM interaction from the enzyme⅐substrate complex.

Discussion
PTPN4 functions as an anti-apoptotic protein. The interaction of PTPN4-PDZ with the PBM of cellular partners is critical for the protection of cells against cell death as PBM binding to PTPN4-PDZ jeopardizes cell homeostasis (Refs. 6 and 15 and this study). However, the cellular partners that directly interact with PTPN4 are poorly known. In this study we identified the MAP kinase p38␥ as a PTPN4-interacting protein. Both proteins are well expressed in human glioblastoma and neuroblastoma cells and depletion of p38␥ or PTPN4, both, lead to cell death. We show that p38␥ and PTPN4 interact in vitro and in neuroblastoma cell extracts. This interaction is mediated by the PDZ domain of PTPN4 and the PBM of p38␥. The p38␥ kinase PBM has the highest affinity among endogenous partners of PTPN4 described up to now with a micromolar affinity for the PDZ domain of PTPN4. Indeed, we previously showed that the C-terminal motifs of two known endogenous partners of PTPN4, the GluN2A subunit of the N-methyl-D-aspartate receptor and the GluD2 subunit of the orphan ionotropic glutamate receptor, bind to the PDZ domain of PTPN4 with k D values equal to 128 M and 42 M, respectively (15). The micromolar affinity observed for the association of the PBM of p38␥ with PTPN4-PDZ was also previously reached by our optimized PBM peptide, Cyto8-RETEV. This peptide was designed based on the x-ray structure of the PTPN4-PDZ in complex with a low affinity RABV PBM ligand (ϪGETRL -COOH ). Here, our x-ray structural study of PTPN4-PDZ in complex with the C-terminal sequence of p38␥ and the tailored peptide Cyto8-RETEV revealed the crucial structural determinants explaining the gain in affinity. Indeed, the two high affinity sequences contract additional interactions between their basic residue at position Ϫ4 of the PBM and the Asp-580 of PTPN4-PDZ and are devoid of unfavorable Arg at position Ϫ1 surrounded by positively charged residues of PTPN4 observed for the low affinity viral sequence. These common interactions stabilized the complexes and hence explained the high affinity of Cyto8-RETEV and p38␥-KETPL for the PDZ domain of PTPN4.
The C-terminal sequence of p38␥ alone is able to mainly recapitulate the thermodynamics properties observed in the full-length p38␥⅐PTPN4-PDZ-PTP complex as revealed by our isothermal calorimetric titrations. Full-length p38␥ binds PTPN4-PDZ via its C terminus PBM motif in a conventional type I PBM⅐PDZ manner, and no other regions of the kinase are involved in the complex formation, as we did not monitor by NMR any chemical shift perturbation of the kinase domain in the presence of PTPN4-PDZ. The PDZ⅐PBM interaction is, therefore, the major contributor to the p38␥⅐PTPN4 complex formation. Among p38 family proteins, p38␥ is the only member to encode a C-terminal PBM. This unique feature of this kinase allows its recruitment independently of the presence of a kinase interaction motif identified in nearly every MAPK regulatory and substrate protein (12). Interestingly, the two NT5 tyrosine phosphatase group members, PTPN3 and PTPN4, both, sharing the same domains organization, take advantage of this unique feature to recruit p38␥ through their PDZ domain (12). Thus, the recognition of the PBM of p38␥ by the PDZ domain of PTPs greatly contributes to the specificity of the interaction compared with other MAPK proteins.
In addition, our results show that recruitment of the PBM of p38␥ by the PTPN4 PDZ domain has a further functional consequence. The catalytic activity of both PTPN4 and PTPN3 is autoinhibited by their PDZ domain when the PDZ and PTP two-domain adopt a compact conformation (20,30). We demonstrate here that the binding of the PBM of p38␥ to PTPN4-PDZ lifts this inhibition and restores the PTPN4 phosphatase activity. We previously described that the PBM binding to the PDZ likely causes partial disruption of the transient interactions between the PDZ and PTP domains and the dynamic rearrangement between the two domains, resulting in the activation of the phosphatase. In the PTPN4⅐p38␥ complex, the PDZ domain of PTPN4 is not only crucial in the recruitment of the kinase, but it would also be involved in the control of the phosphatase activity and specifically favors the dephosphorylation of p38␥. After the first PDZ⅐PBM recognition step, the subsequent phosphatase activation could then serve to induce the dephosphorylation of the TGY activation loop of p38␥, that is crucial for the regulation of the kinase activity.
Viral proteins efficiently mimic cellular PBM such as the one of p38␥, affecting both the recruitment of substrate and the regulation of the activity in the case of PTPN4. Infection by an attenuated RABV strain perturbs the anti-apoptotic function of PTPN4 and induces cell death. The targeting of the PDZ domain of PTPN4 by the PBM of the viral G protein of the attenuated RABV strain, Cyto13-att, abrogates the protective role of PTPN4 and is responsible for the cell death caused by this RABV strain. Nevertheless, the G protein of attenuated RABV strain competes with an endogenous ligand whose identity is not yet known. The PTPN4⅐p38␥ association could be affected by the G protein during viral affection. The mere disturbing of the crucial PDZ⅐PBM p38␥ recruitment step might be sufficient to prevent the dephosphorylation of the kinase and thereby to disrupt its function. Thus, hijacking the

Kinetic parameters of the hydrolysis of p-nitrophenyl phosphate (pNPP) and phosphorylated TGY-p38␥ loop by various PTPN4 constructs
The initial reaction rates at each substrate concentration were independently measured at pH 7.5 and 25°C. The Michaelis constant (K m ) and the turnover number (k cat ) values were deduced by fitting the experimental data to the Michaelis-Menten equation. The data and error are representative of three independent experiments.  p38␥⅐PTPN4 signaling by the RABV G protein could be the main cause of the apoptosis phenotype of the attenuated RABV strain. Further studies are necessary to demonstrate this point. As shown by a few studies such as those of the HtrA2 and DegS proteases and ours on PTPN4 phosphatase, the PDZ domains and their ligand play an essential role in regulating enzymatic activity. Targeting PDZ⅐PBM interactions involving signaling proteins such as kinases and phosphatases is a promising therapeutic strategy to specifically perturb the homeostasis of tumor cells. By elucidating the molecular basis of PTPN4⅐p38␥ interaction, we are rationally designing new tumoricid molecules inhibiting the phosphatase by locking PDZ and PTP domains together.