Reverse Two-hybrid Screening Identifies Residues of JNK Required for Interaction with the Kinase Interaction Motif of JNK-interacting Protein-1*

The development of specific inhibitors for the c-Jun N-terminal kinase (JNK) family of mitogen-activated protein kinases (MAPKs) has been a recent research focus because of the association of JNK with cell death in conditions such as stroke and neurodegeneration. We have demonstrated previously the presence of critical inhibitory residues within an 11-mer peptide (TI-JIP) based on the sequence of JNK-interacting protein-1 (JIP-1). However, the corresponding region of JNK bound by this JIP-1-based peptide was unknown. To identify this region, we used a novel reverse two-hybrid approach with TI-JIP as bait. We screened a library of JNK1 mutants that had been generated by random PCR mutagenesis and found three mutants of JNK1 that failed to interact with TI-JIP. The mutations in JNK1 were L131R, R309W, and Y320H. Of these mutated residues, Leu-131 and Tyr-320 were located on a common face of the JNK protein close to other residues implicated previously in the interactions of MAPKs with substrates, phosphatases, and scaffolds. To test whether these JNK1 mutants were thus affected in their regulation, we evaluated their activation in mammalian cells in response to hyperosmolarity or cotransfection with a constitutively active upstream kinase or their direct phosphorylation by either MAPK kinase (MKK)4 or MKK7. In each situation, all three JNK mutants were not activated or phosphorylated to the same level as wild-type JNK. Therefore, the results of our unbiased reverse two-hybrid screening approach have identified residues of JNK responsible for binding JIP-1-based peptides as well as MKK4 or MKK7.

The c-Jun N-terminal kinase (JNK) 1 subfamily of mitogenactivated protein kinases (MAPKs) is activated after the expo-sure of cells to various stimuli including growth factors, cytokines, and cellular stresses (1)(2)(3)(4)(5). This activation of JNK requires the phosphorylation of Thr and Tyr residues in its activation loop by the upstream kinases MKK4 and MKK7 (6). Activated JNK then phosphorylates nuclear substrates including c-Jun, ATF-2, and Elk-1 and non-nuclear substrates such as Bcl-2 family members. This allows JNK to contribute to diverse biological processes including cell proliferation, differentiation, survival, and death (for review, see Ref. 7).
Selective JNK or JNK pathway inhibitors have been developed after the implication of JNK signaling in various disease states. Presently, there are three classes of JNK pathway inhibitors. The first of these is Cephalon Incorporated library compound 1347 (CEP-1347), which inhibits the mixed lineage kinases that function upstream in the JNK pathway (8). This compound inhibits JNK activation both in vitro and in vivo (9 -12) and is currently in phase II clinical trials for the treatment of Parkinson's disease. The second inhibitor is Signal Pharmaceuticals library compound 600125 (SP600125), which is a reversible ATP-competitive JNK inhibitor (13). However, the selectivity of this compound for JNK is questionable after recent specificity tests where SP600125 inhibited 13 other protein kinases with similar or greater potency (14).
The third class of JNK inhibitors includes peptides based on the kinase interaction motif (KIM) of the scaffold protein, JNKinteracting protein-1 (JIP-1). We demonstrated previously that an 11-mer peptide, TI-JIP, corresponding to residues 153-163 of murine JIP-1, potently inhibited JNK activity in vitro (15). In addition, cell-permeable peptides based on this region of JIP-1 effectively inhibited JNK activity and subsequent downstream events in vivo. For example, these peptides protected pancreatic ␤TC-3 cells against interleukin-1␤-induced apoptosis (16). In addition, neurons were protected from injury caused by excitotoxicity and cerebral ischemia (17), and the peptides offered protection from aminoglycoside and acoustic trauma-induced auditory hair cell death and hearing loss (18). The high specificity of these peptides for inhibiting JNK has also been demonstrated, with 10 M peptide not inhibiting the activity of 40 different kinases tested and 500 M peptide reducing substrate phosphorylation by 6 different kinases by no more than 1% (17).
Despite these successes in the use of JIP-1-based peptide inhibitors, the region of JNK targeted by these inhibitors has remained undefined. Because these peptides are based on the KIM of JIP-1, it is possible that they are bound by regions of JNK reported to interact with the KIMs of other JNK-interacting substrates, upstream activators, or phosphatases. One such region on JNK is the acidic patch known as the "common docking" (CD) domain. Previous studies have demonstrated that mutagenesis of residues in this region affects the interaction of the MAPKs ERK2 and p38 with their activators and substrates (19,20). In addition, the "ED" site in p38 and the corresponding "TT" site in ERK2 have been implicated in docking interactions with MAPK-activated protein kinases (21). However, p38 has been co-crystallized in complex with KIMbased peptides from substrate MEF2A and activator MKK3b (22). These peptides both bound to the same site in the Cterminal domain of p38 outside the active site and distinct from the CD domain implicated previously in docking site interactions (22). Extrapolating these data from p38 MAPK to JNK suggests that a homologous region of JNK might be responsible for mediating the interaction with the scaffold protein JIP-1 and therefore with the inhibitor TI-JIP. However, many of the residues lining the docking groove in p38 are conserved in ERK2 but not JNK, and most of the residues in p38 which were in contact with the peptides were variable among the different MAPK subfamilies (22). This suggests the possibility that a unique peptide docking groove on JNK may exist.
We have used reverse two-hybrid screening to investigate the region of JNK1 which interacts with JIP-1-based peptide inhibitors. This approach has allowed us to identify residues in JNK which, when mutated, interfere with binding to the TI-JIP peptide. By screening a randomly generated library of JNK mutants, our approach would not be biased by previous studies implicating residues in the interaction of MAPKs and their KIM-containing partners. Thus, we used the 11-mer TI-JIP peptide as bait, and this was screened against a library of JNK mutants created by random PCR mutagenesis. With reverse two-hybrid screening, we identified JNK mutants unable to interact with TI-JIP. Furthermore, when we tested individual point mutations, three were identified which disrupted the interaction between JNK and TI-JIP. Two of these mutations were located in regions distinct from the CD and ED sites identified previously on MAPKs but adjacent to regions thought to participate in hydrophobic interactions between MAPKs and KIMs of interacting partners. This suggested that the JNK mutants created in this study might also be unable to interact with the KIMs of other interacting partners. This hypothesis was confirmed when we found that the JNK mutants were not phosphorylated by either MKK4 or MKK7 or activated in cells expressing constitutively active MEKK1 or exposed to hyperosmotic shock. These results suggested that JIP-based inhibitor peptides interacted with a region of JNK which was similarly bound by the KIMs of other protein partners such as activators, substrates, and scaffolds. These conclusions have been reinforced with the recently published structure of JNK co-crystallized with TI-JIP (23).

EXPERIMENTAL PROCEDURES
Plasmid DNA Constructs-Oligonucleotides encoding TI-JIP were annealed to produce a double-stranded DNA fragment with overhanging ends compatible with EcoRI at the 5Ј-end and XhoI at the 3Ј-end. These were ligated into EcoRI/XhoI-digested pGILDA (Clontech), thus generating C-terminal fusion proteins with the LexA DNA binding domain. The human JNK1 sequence (24) was PCR amplified, digested with MfeI and XhoI, and ligated into EcoRI/XhoI-digested pJG4-5 (Clontech), thus generating C-terminal fusion proteins with the B42 transcriptional activation domain. DNA sequencing confirmed the identity of these constructs.
Construction of Mutant JNK Library Using Random PCR Mutagenesis-50-l PCRs contained 5 units of Taq polymerase (Roche Applied Science), 50 pmol of forward primer, 50 pmol of reverse primer, and 10 ng of template DNA in error-prone PCR buffer (final concentrations: 100 mM Tris-HCl, pH 8.3, 500 mM KCl, 70 mM MgCl 2, 0.1% (w/v) gelatin, 10% (v/v) dimethyl sulfoxide, 0.2 mM dATP, 0.2 mM dGTP, 1 mM dCTP, 1 mM dTTP). Four different mutagenesis reactions were performed, where MnCl 2 was added to final concentrations of 0.1, 0.2, or 0.3 mM prior to temperature cycling, or MnCl 2 was added to a final concentration of 0.3 mM after completion of 10 rounds of temperature cycling. Reactions were subjected to 30 cycles with the following conditions: 94°C for 1 min; 55°C for 1 min; 72°C for 3 min. The pooled PCR products were digested with MfeI/XhoI, ligated into EcoRI/XhoI-digested pJG4-5, transformed into ElectroTenBlue (Stratagene), electrocompetent Escherichia coli, and plated on LB agar containing 100 g/ml ampicillin. Plates were incubated at 30°C overnight and then 37°C for 3 h, yielding a total of 9 ϫ 10 6 colonies. The bacterial library was harvested and the DNA isolated using a Qiagen Maxiprep Kit before being transformed into the yeast strain PRT 48 2 (derived from SKY 48: MAT␣, trp1, ura3, his3, 6lexAop-LEU2, cIop-LYS2 (25)) in accordance with the Gietz high efficiency transformation protocol (26). Yeast were grown at 30°C for 4 days on synthetic complete medium lacking tryptophan and containing 2% glucose. The transformation yielded a library of 5 ϫ 10 5 colonies that were harvested and stored at Ϫ80°C in sterile yeast freezing buffer (65% (v/v) glycerol, 0.1 M MgSO 4 , 25 mM Tris-HCl, pH 8.0).
Immunoblotting for HA-tagged JNK Proteins-Yeast expressing JNK mutant constructs were grown on UHW Ϫ agar containing 0.05% (w/v) galactose and 2% (w/v) raffinose for 48 h at 30°C and then vortexed in 20 l of SDS-PAGE sample buffer and snap-frozen in liquid N 2 as described in Ref. 27. Samples were heated at 100°C for 5 min prior to separation by SDS-PAGE. Proteins were transferred to nitrocellulose by semidry electroblotting and probed for HA-tagged products using an anti-HA antibody (Roche Applied Science).
Recovery of Mutant JNK Constructs-Where yeast expressed a fulllength HA-tagged activation domain-JNK1 fusion protein (58 kDa), the JNK constructs were isolated from yeast by lyticase extraction and then electroporated into KC8 bacteria and purified further as described in Ref. 27.
Cell Transfection, Lysis, and Immunoblotting-COS cells were transfected with pCMV-FLAG-JNK1 (24) or equivalent mutant constructs and pEBG-MKK7␤1 (provided by A. Whitmarsh, University of Manchester) as specified in the figures using LipofectAMINE and PLUS reagent (Invitrogen) according to the manufacturer's instructions. After cell lysis as described in Ref. 15 and addition of 3ϫ SDSsample buffer, proteins were separated using SDS-PAGE. After protein transfer onto nitrocellulose, immunoblotting was performed using antiactive JNK (Promega), anti-FLAG M2 (Sigma), or anti-JNK1 (Santa Cruz) primary antibodies. Primary antibodies were bound by horseradish peroxidase-conjugated secondary antibodies (Pierce), and immunocomplexes were visualized using chemiluminescence.
Immunoprecipitation and Protein Kinase Assays-FLAG-JNK1 proteins were immunoprecipitated by the addition of anti-FLAG M2 and assayed for kinase activity as described in Ref. 15. For assays of JNK activation, the washed immunocomplexes were incubated with 30 l of reaction buffer containing 20 M ATP, 5 Ci of [␥-32 P]ATP, and 1 g of GST-MKK4(ED) at 30°C for 1 h, and then analyzed as described above. Where immunoprecipitated GST-MKK7␤1 was used to activate JNK, reactions were performed as for those with GST-MKK4(ED) but with 1 Ci of [␥-32 P]ATP and incubation at 30°C for 30 min.

RESULTS
Random PCR Mutagenesis Created a Library of JNK1 Mutants-Initially, we constructed a series of directed N-and C-terminal truncations of JNK1 as fusions with the C terminus of the B42 transcriptional activation domain, to identify a smaller region of JNK to be subjected to mutagenesis. However, these JNK1 mutants were poorly expressed in RFY 206/ PRT 49 diploids relative to the wild-type protein (data not shown), and therefore we proceeded to mutagenize the entire JNK1 sequence randomly. In optimizing the random PCR mutagenesis, we found that reactions containing 0.3 mM MnCl 2 resulted in the presence of up to 11 point mutations per fulllength JNK sequence. Therefore, we used four different mutagenic PCR conditions to generate a library of JNK sequences containing up to 11 point mutations per JNK sequence.
Reverse Two-hybrid Screening-We employed a reverse twohybrid method to screen the library of JNK mutants for those mutants that lost the ability to interact with TI-JIP ( Fig. 1). In this system, the PRT 480 yeast strain with the counterselectable URA3 reporter gene was transformed with pGILDA-TI-JIP. These yeast were mated to PRT 48 yeast transformed with the mutant JNK library in the pJG4-5 vector and grown in the presence of 5Ј-FOA, which is toxic to yeast when the URA3encoded enzyme is expressed. In the presence of galactose, the neutral carbon source raffinose, and a low concentration of glucose to reduce background survival (upper panels showing selective growth), bait and prey expression was induced, and A, where bait and prey interacted, as was the case for TI-JIP and the AD-JNK fusion protein, the URA3 reporter gene was transcribed. This converted 5Ј-FOA included in the yeast medium into a toxic product, and the yeast died. B, where TI-JIP was screened against a library of random JNK mutants (AD-JNK(MUT)), there were some mutants that retained interaction with TI-JIP and died and some noninteracting JNK mutants that survived because the URA3 reporter gene was not transcribed, and 5Ј-FOA was not converted into a toxic product. C, where bait and prey did not interact, as was the case for TI-JIP and the AD product of the empty prey vector (AD alone), the URA3 gene was not transcribed, and 5Ј-FOA was not converted into a toxic product; hence, the yeast survived. Illustrated are the optimized screening conditions that permitted maximal death of the positive control yeast (TI-JIP and AD-JNK) with minimal death of negative control yeast (TI-JIP and AD). The upper panels show yeast growth in the presence of galactose (0.08% Gal), raffinose (2% Raff), and a low concentration of glucose (0.05% Gluc), which induced bait and prey expression. The lower panels show yeast growth in the presence of glucose (2% Gluc), which repressed bait and prey expression and was indicative of the total number of yeast plated on the medium. yeast expressing interacting partners were sensitive to 5Ј-FOA. Therefore, in the presence of 5Ј-FOA, an interaction between TI-JIP and JNK resulted in cell death (Fig. 1A). In contrast, a lack of interaction between TI-JIP and either a noninteracting JNK mutant (Fig. 1B) or the activation domain encoded by the empty pJG4-5 vector (Fig. 1C) allowed yeast to survive treatment with 5Ј-FOA. More yeast colonies grew on the test plates (TI-JIP plus mutant JNK library, Fig. 1B) than the positive control plates (TI-JIP plus JNK, Fig. 1A), but this was less than the number on the negative control plates (TI-JIP plus pJG4-5, Fig. 1C), which would be expected when the mutant JNK library contained both noninteracting mutants and mutants that were phenotypically normal and retained the ability to interact with TI-JIP. Yeast were grown separately in the presence of glucose ( Fig. 1, lower panels showing total growth), which repressed bait and prey expression resulting in insensitivity to 5Ј-FOA and was indicative of the total number of viable yeast on the plates.
Analyzing Colonies That Survived the Reverse Two-hybrid Screening-Approximately 600 colonies were obtained after plating 600,000 diploids on the reverse screening plates. Screening by colony PCR using JNK-specific primers indicated that 200 of the 600 colonies contained a prey plasmid with a JNK insert. A representative selection of this screen is shown in Fig. 2A. Immunoblotting for the HA-tagged prey protein indicated that only 21 of the 200 interaction-deficient mutants expressed a full-length JNK protein (46 kDa) in fusion with the activation domain, AD (12 kDa) to produce the expected protein size of 58 kDa. A representative selection of this screen by immunoblotting showing six full-length JNK proteins and four truncated JNK proteins is shown in Fig. 2B. Yeast expressing full-length JNK proteins were then grown on medium lacking leucine in the presence of 0.08% (w/v) galactose and 2% (w/v) raffinose. This forward two-hybrid analysis screened for an interaction between TI-JIP and mutant JNK proteins, made evident by growth of yeast in the absence of leucine because of expression of the LEU2 reporter gene product. Of the 21 colonies expressing mutant JNK proteins, 5 were found to represent false positives because they interacted with TI-JIP under the conditions of the forward screen (results not shown). It is likely that these false positive yeast grew on the reverse screening plates despite interacting bait and prey proteins that would normally produce toxicity and death because of an evasion of the counterselection pathway such as the epigenetic shutdown of the URA3 reporter expression in the yeast. The 16 remaining interaction-deficient mutants, each containing from 2 to 11 mutations in the full-length JNK sequence, were analyzed by DNA sequencing. By translating the mutated JNK sequences, we found that in total 70 amino acids had been mutated, and some mutations were common to more than one mutant JNK sequence.
Summary of Mutation Data-From the reduced pool of 16 mutants, the frequency of mutations per region of secondary structure was calculated and normalized for the length of the structure (Fig. 3A), resulting in secondary "hot spots" (#1, #2). Human JNK1 and JNK3 demonstrate up to 96% sequence homology when their sequences are compared using an Entrez BLAST query, therefore the JNK1 mutations were mapped onto the surface of the JNK3 structure (PDB: 1JNK) using WebLab ViewerLite software, to depict their positions in the protein tertiary structure. Because the mutations mapped to various regions of the JNK structure, it was difficult to detect tertiary hot spots. Therefore, we reduced the mutant pool to those containing 5 or fewer point mutations per JNK molecule in an attempt to reduce background noise. This reduced the mutant pool from 16 to 6. Furthermore, this revealed some clustering of mutations on the surface of JNK, particularly in the C-terminal lobe of JNK (Fig. 3B). Using both the secondary and tertiary hot spot data along with residues that were altered in multiple mutants, we assigned regions for further investigation.
We chose nine individual JNK residues to target by point mutation. Using site-directed mutagenesis in accordance with the Stratagene QuikChange protocol, we altered single residues of JNK to represent the changes that occurred in mutants isolated by reverse two-hybrid screening. Specifically, the point mutations were L110H, D124Y, L131R, V219D, E261K, R309W, D313G, D314G, and Y320H, and the locations of the targeted residues are represented on the JNK structure shown in Fig. 4A. When these mutants were tested for interaction with TI-JIP by forward two-hybrid screening, a ␤-galactosidase overlay assay indicated that of the nine point mutants tested, only the L131R, R309W, and Y320H did not interact with TI-JIP (Fig. 4B). This was not simply the result of impaired protein expression of the mutants, because Western blotting indicated that full-length JNK proteins were expressed (Fig.  4B). Two independent yeast colonies were tested in the case of each mutation to confirm the results of the ␤-galactosidase overlay assay and Western blotting.
The residues Leu-131 and Tyr-320 were located near each other on a common face of the JNK protein ( Fig. 4A(ii)), whereas the Arg-309 mutation was located on another face of JNK (Fig. 4A(iii)). These amino acids were not buried within the core of the JNK protein, and therefore it is unlikely that their mutation affected the global folding or stability of the protein (29). Because these residues demonstrated some surface exposure, it was possible that they were involved in mediating the interaction between JNK and TI-JIP. In addition, because TI-JIP is a KIM-based peptide, it was possible that these mutations would disrupt the interaction between JNK and other KIM-containing proteins. To investigate this notion, we assessed the biochemistry of the L131R, R309W, and Y320H mutants of JNK in mammalian cells.
JNK1 Mutants Were Impaired in Their Ability to Phosphorylate c-Jun after Exposure to Activating Stimuli-We used site-directed mutagenesis to generate mutations in the pCMV-FLAG-JNK1 construct to produce JNK1 proteins with point mutations corresponding to L131R, R309W, and Y320H. COS cells were transfected with these constructs, and Western blotting performed on cell lysates revealed overexpression of FLAG-tagged JNK1 and all three FLAG-tagged mutants (Fig.  5A). In addition, the Y320H mutant consistently demonstrated reduced mobility after SDS-PAGE relative to the wild-type FIG. 4. Single point mutants define important residues on JNK for its interaction with TI-JIP. A, amino acids located in putative mutational hot spots were targeted for further investigation. Point mutants of JNK were constructed by site-directed mutagenesis to assess the relative contribution of different hot spots to the JNK-TI-JIP interaction. Four views of the protein are shown (i-iv) to illustrate all faces of the three-dimensional structure with the positions of mutated amino acids shown in black. B, a ␤-galactosidase overlay assay was performed to investigate the ability of JNK mutants to interact with TI-JIP. Of the nine point mutations tested, three point mutations (L131R, R309W, and Y320H) rendered JNK incapable of interaction with TI-JIP. Western blotting to detect the HA-tagged full-length JNK1 mutant proteins was performed to ensure that the lack of interaction did not arise from problems associated with protein expression. Two independent colonies were tested for each mutation to confirm the results of the overlay assay and Western blotting. for 30 min resulted in strong phosphorylation of c-Jun substrate in in vitro kinase assays using FLAG immunoprecipitation from lysates prepared from these cells. This phosphorylation was 5.5-fold greater than the c-Jun phosphorylation detected in assays of the corresponding unstimulated cells (Fig.  5B). When the JNK mutants were tested in parallel assays, each mutant was only activated by 1-2-fold by osmotic shock (Fig. 5B). To confirm these results, we next tested a more potent and specific stimulus of the JNK pathway.
When a constitutively active form of MEKK1 (CA-MEKK1) was cotransfected into COS cells with wild-type JNK, FLAG immunoprecipitates from these cell lysates showed a 240-fold increase in c-Jun phosphorylation in in vitro kinase assays relative to samples prepared from cells transfected with wildtype JNK alone (Fig. 5C). However, the corresponding samples with mutant JNKs again displayed a much lower increase in c-Jun phosphorylation (20 -70-fold) after cotransfection of cells with CA-MEKK1 (Fig. 5C). Therefore, the JNK mutants displayed an impaired ability to phosphorylate c-Jun in response to both of these activating stimuli.
JNK1 Mutants Were Not Activated by Either MKK4 or MKK7-The impaired c-Jun phosphorylation by the JNK mutants (Fig. 5, A and B) may have resulted from their impaired activation. To evaluate this issue, we directly investigated the phosphorylation of these mutants without relying on the subsequent phosphorylation of c-Jun. Mutant JNKs were immunoprecipitated from transfected cell lysates and incubated with a constitutively active form of MKK4 (GST-MKK4(ED)) in the presence of [␥-32 P]ATP. The presence of active MKK4 increased the phosphorylation of wild-type JNK relative to the autophosphorylation that occurred in the absence of any upstream activator protein (Fig. 6A). In contrast, the negligible amount of radioactive phosphate incorporated into any of the three JNK mutants was not increased by the presence of active MKK4 (Fig. 6A). In addition, there appeared to be some phosphorylation of GST-MKK4 in the assay, and it was evident that this was increased in the presence of wild-type JNK, but not in the presence of any of the JNK mutants (Fig. 6A).
While our study was nearing completion, it was reported that a double alanine mutant of JNK2 (E329A/E331A) did not interact with JIP-1, c-Jun, or MKK4 but retained the ability to be activated by MKK7 (30). Therefore, we tested the ability of the L131R, R309W, and Y320H JNK1 mutants to be activated by MKK7, by phosphoblotting and in vitro kinase assays. Phosphoblotting for dual phosphorylated JNK indicated that wildtype JNK1 was strongly phosphorylated by MKK7 in cotransfected cells (Fig. 6B, upper panel). However, cotransfection of MKK7 did not stimulate the phosphorylation of any of the JNK mutants relative to the mock-transfected control (Fig. 6B, upper panel). This was despite the overexpression of these JNK mutant proteins relative to endogenous JNK as indicated by blotting for total JNK1 (Fig. 6B, lower panel). Similar results were obtained from in vitro kinase assays, where wild-type JNK was strongly phosphorylated in the presence of MKK7, but there was no detectable phosphorylation of the JNK mutants in the presence of MKK7 (data not shown). Therefore, it appeared that the L131R, R309W, and Y320H JNK1 mutants were impaired in their activation by both MKK4 and MKK7, contributing to their impaired responses to hyperosmolarity and cotransfection with CA-MEKK1 (Fig. 5). DISCUSSION The JNK MAPK pathway is activated after the exposure of cells to a wide range of extracellular stimuli including stress, cytokines, and growth factors, but still the consequences of JNK activation remain controversial (for review, see Ref. 31). Our understanding of this pathway is being enhanced by mul-tiple parallel approaches including gene knock-outs and overexpression studies, as well as the closer evaluation of the biochemical features of members of this pathway. In addition to studies on the JNKs, or their upstream activators, increasing attention is being focused on the regulation of JNK signaling by the JIP family of scaffold proteins. Interestingly, JIPs can both increase (32)(33)(34) and decrease (15,16,35) signaling through the JNK cascade.
In this study, we have investigated further the binding interaction between JNK and the TI-JIP peptide, which represents the KIM of the JIP-1 scaffold protein. When we initiated our study, there were no published studies on the interaction interface of MAPKs for scaffold proteins or peptides derived from them. To perform an unbiased analysis of the residues of JNK which may mediate its interaction with the TI-JIP peptide, we used a reverse two-hybrid analysis. We screened a library of mutant JNK1 proteins and isolated full-length mutant JNKs that could not interact with TI-JIP. Robinson and colleagues (39) have previously combined random PCR mutagenesis and forward two-hybrid analysis to identify four residues on ERK2 that mediate its interaction with MEK1/2. Because reverse two-hybrid screening selects against an interaction but forward two-hybrid screening selects for an interaction, noninteractors can be more readily obtained with the reverse two-hybrid approach. Other groups have also successfully coupled PCR mutagenesis and reverse two-hybrid screening to characterize interactions between proteins other than MAPKs (36 -38). In our approach, we have implicated three residues in JNK1, Leu-131, Arg-309, and Tyr-320, which had not previously been identified as possible mediators of the interaction with the TI-JIP peptide.
To gain further understanding of how mutation of these three residues might impact JNK interaction with TI-JIP, we have compared their locations with previously identified binding sites on JNK. In a schematic representation of the structure of JNK1 in Fig. 7, the ribbon structure of JNK (Fig. 7A) provides orientation for the subsequent space-filling model diagrams. First, the positions of Leu-131 and Tyr-320 that we have now identified are shown in purple in Fig. 7B and in all subsequent representations of their positions on the JNK structure ( Figs. 7 and 8). In previous studies, a number of residues have been reported to mediate the interaction of MAPKs with their KIM-containing binding partners. In the following paragraphs, we discuss these regions. We begin with the acidic CD domain of MAPKs first characterized by Tanoue and colleagues (20) in their studies on ERK2 and its binding partners.
The CD domain has been described as a cluster of negatively charged amino acids located on the side opposite the active site in the structure of MAPKs (20). To characterize the CD domain, Tanoue and colleagues mutated two aspartate residues in ERK2 to asparagine residues and found that these changes disrupted binding to its KIM-containing partner MAPK phosphatase 3 (20). In human JNK1, the CD domain residue Asp-326 is conserved and is shown in Fig. 7C. Only one previous study has used mutagenesis to evaluate the residues of JNK which mediate binding with its KIMcontaining partners (30). Somewhat surprisingly, this showed that mutation of the JNK2 CD site residue (Glu-326 in JNK2) did not disrupt the interaction of JNK2 with JIP-1 (30). However, the mutation of two other acidic residues, Glu-329 and Glu-331, which are located close to Asp/Glu-326, confirmed these as important for the interaction between JNK2 and the JIP-1 scaffold protein (30). In particular, Glu-329 was critical for efficient binding between JNK2 and JIP-1, whereas Glu-331 made a more minor contribution (30). These residues are conserved in JNK1. When Glu-329 and Glu-331 are highlighted on the three-dimensional structure (Fig. 7C), Glu-329 is situated a short distance (12-14 Å) from two of the residues, Leu-131 and Tyr-320, that we have identified in our current study.
In addition to the classical CD site residues, other residues have been previously identified that are responsible for high affinity binding to KIM-containing partners (21,40). Additional negatively charged amino acids Glu-160 and Asp-161 near the CD domain in p38 MAPK form the ED site and mediate the interaction with MAPK activated protein kinase-3, a dominant target of wild-type p38 (21). The corresponding site in ERK2 has been referred to as the TT site (21), whereas in JNK1 the equivalent residues are Ser-161 and Asp-162 so that this may be described as the JNK SD site (Fig. 7D). Furthermore, the JNK1 residue Tyr-130 shares homology with a corresponding residue in ERK2 involved in the ERK2-MKP3 interaction (40), and it is located directly adjacent to Leu-131, identified in our current study. This site on JNK is situated on the same face of the kinase as the CD domain, and Leu-131 is situated directly below the SD site residues (Fig. 7D). Therefore, although Leu-131 and Tyr-320 that we have identified are distinct from both the CD site and the ED site residues identified by others, they are located close to these sites. Interestingly, when the ED site residue Asp-162 was mutated to asparagine, the interaction between JNK2 and JIP-1 was not disrupted (30). Furthermore, mutation of residues adjacent to FIG. 7. Location of JNK1 residues Leu-131 and Tyr-320 relative to other residues implicated in MAPK docking interactions. A, ribbon structure of JNK for comparison with space-filling models. B, space-filling structure with JNK1 residues Leu-131 and Tyr-320 highlighted in purple, which were implicated in the interaction between JNK1 and the TI-JIP inhibitor, based on the KIM of JIP-1. C, as in B, with the CD residue Asp-326 highlighted in green, along with residues Glu-329 and Glu-331, which mediate the interaction between JNK2 and JIP-1 (30). D, as in B, with ED site residues Ser-161 and Asp-162 highlighted in red along with Tyr-130, which was implicated in the ERK2-MKP3 interaction (40). E, as in B, with JNK1 residues 107-131 and 159 -165 highlighted in cyan, which correspond to residues in the related p38 MAPK which mediate hydrophobic contacts with KIM sequences present in interacting partners (22).
FIG. 8. Proximity of altered residues in JNK1 mutant proteins isolated by reverse two-hybrid screening, to JNK1 residues implicated in the interaction with TI-JIP by co-crystallization analysis. A, space-filling structure with JNK1 residues Leu-131 and Tyr-320 highlighted in purple, which were implicated in the interaction between JNK1 and the TI-JIP inhibitor. Residues that were implicated in the interaction between JNK1 and TI-JIP by co-crystallization analysis (23) are colored green. B, the hydrophobic pocket thought to interact with the L-X-L motif of TI-JIP is circled and consists of JNK1 residues Ala-113, Leu-115, Val-118, Met-121, Leu-123, Leu-131, Val-159, and Cys-163 (23). Furthermore, as described previously (23), the backbone carbonyl group of Ser-161 of JNK1 H-bonded with TI-JIP residue Leu-8; the N⑀ atom of Arg-127 of JNK1 H-bonded with TI-JIP residue Thr-7; JNK1 residues Tyr-130, Glu-126, and Trp-324 had Van der Waals contacts with TI-JIP residue Pro-5; JNK1 residue Val-323 made a weak interaction with TI-JIP residue Pro-2; and JNK1 residue Glu-329 interacted with TI-JIP residue Arg-4 with a bidentate salt bridge. Lastly, in our original screen, 14 amino acids appeared close to residues implicated in the JNK-TI-JIP interaction by co-crystallization. Two of these mutations were tested and found not to change the interaction with TI-JIP (indicated in red), whereas 12 remain to be tested (indicated in yellow).
the ED site including K160N/S161E or T164D did not disrupt the interaction between JNK2 and JIP-1 (30). Thus, any close proximity of amino acids to the identified binding residues in the CD or ED sites cannot be used solely to predict whether a residue mediates the interaction with a KIM-containing partner. Instead, it would appear that a more robust test is to subject the candidate residues to site-directed mutagenesis and then to evaluate the subsequent effect on binding interactions.
An alternative approach that does not rely on mutagenesis to identify binding regions but that has the power to provide robust information on residues important for binding requires the co-crystallization of a protein and the peptide ligand of interest. In this case of identifying KIM-binding residues, a MAPK would be co-crystallized with KIM-based peptides of interest. In the first example of this approach, p38 MAPK was co-crystallized in complex with KIM-based peptides from substrate MEF2A and activator MKK3b (22). This has provided further support that the site of interaction is in the C-terminal domain of this MAPK and that a number of hydrophobic interactions likely contribute to the interaction with the KIM peptides (22). The equivalent regions of human JNK1 comprise Val-107 to Leu-131 and Val-159 to Leu-165, and these are highlighted in Fig. 7E. Again, Leu-131 and Tyr-320 that we identified are near this region being directly within the cluster or adjacent to these residues, respectively. In addition, Ile-116 in p38 was reported to form hydrophobic contacts with the L-X-L motif present in the KIM consensus sequence (22), and the side chain of the corresponding JNK1 residue, Val-118, points toward Leu-131 and is in close proximity to this residue (3-5Å). Finally, the p38 residues Leu-113 and Leu-122 were also found to be in contact with bound KIM peptides (22). These residues are conserved in p38, ERK2, and JNK1/2, and in JNK1 their side chains are also in close proximity to Leu-131 (4 Å). This emphasizes the critical contribution that Leu-131 makes to this binding interface.
This conclusion that Leu-131 is important in the KIM binding interface of JNK is supported further by the data published on JNK1 in complex with TI-JIP just prior to the submission of our study (23). Consistent with the models presented in Fig. 7, only residues from the C-terminal domain of JNK1 were shown to be involved (23). In Fig. 8A we show residues that have now been identified in this structural analysis which are likely mediators of the interaction with TI-JIP. In Fig. 8B, a closer view of this region is represented. Although a detailed analysis of the structural information can only be made after the publication of the structural coordinates on the Protein Data Base (identification numbers 1UKH and 1UKI), a number of general points can be made by reference to the amino acids highlighted by Heo and colleagues (23). Overall this interface region of JNK1 is mostly nonpolar and this emphasizes the importance of hydrophobic interactions when JNK1 binds TI-JIP. The Leu residues in the L-X-L motif of the TI-JIP peptide, which are critical for mediating the JNK inhibition by TI-JIP (15), were extensively surrounded by many hydrophobic residues of JNK1 including Leu-131, identified in our current study (indicated in purple in Fig. 8), as well as Ala-113, Leu-115, Val-118, Met-121, Leu-123, Val-159, and Cys-163, indicated in green and circled in Fig. 8B (23). In addition, the critical Pro residue within TI-JIP (15) made van der Waals contacts with the side chains of Tyr-130, Glu-126, and Trp-324 (23). These residues, which are indicated in Fig. 8B, are located on the surface of JNK between Leu-131 and Tyr-320 highlighted in our current study. Furthermore, the critical Arg-4 residue within TI-JIP (15) was found to interact with Glu-329 with a bidentate salt bridge (23), and this residue was demonstrated to be critical for the efficient binding between JNK2 and JIP-1 (30).
Taking the published data together with the mutants we have identified (Figs. 7 and 8), it appears that Leu-131 and Tyr-320 are both likely to contribute to the interaction of JNK with TI-JIP. However, as we have shown in Fig. 4A(iii), the other residue we identified, Arg-309, is located on the opposite side of the JNK molecule. The distance between Arg-309 and the other two residues (Ͼ20 Å) suggests that all three residues cannot be directly involved in the interaction with TI-JIP by forming a single contiguous binding pocket. In contrast to the side chains of Leu-131 and Tyr-320, the side chain of Arg-309 is not surface-exposed. It is more likely therefore that the R309W mutant we identified alters the structure of the JNK protein.
Energy minimizations of the modeled structure of the JNK1 R309W mutant using the program DeepView do not predict instability of the protein, and indeed this mutant protein was expressed in mammalian cells to a level comparable with both the wild-type JNK1 and the other two mutants. Further structural analysis of this mutant will therefore be needed to evaluate whether an R309W change is critical for structural changes that subsequently alter the ability of JNK to bind KIM-containing partners. Support for the structural flexibility of JNK comes also from the recent study by Heo and colleagues (23) who also showed that the binding to the TI-JIP peptide subsequently disordered the ATP binding pocket of JNK.
The results in the study by Heo and colleagues additionally allow us now to return to the 16 full-length JNK mutants identified in our original screen. We can now reevaluate the mutations that we had first discarded in the refinement of the mutant pool (Fig. 3B) to include only those JNK clones containing 5 or fewer point mutants per JNK molecule. Of the 70 mutations, 12 were represented from two to four times, whereas 58 were represented only a single time in this mutant group. Of these 70 residues, 12 were located close to the residues identified by Heo et al. (23), and these are indicated in yellow in Fig. 8. Each of these changes will need further experimental confirmation of their contribution to the binding of TI-JIP. Importantly, the close proximity of Asp-124 and Val-219 (indicated in red in Fig. 8), which we tested for interaction by the D124Y and V219D mutations (Fig. 4B), did not appear sufficient to implicate these residues in the binding interaction. This again emphasizes that close proximity to the residues identified is not a simple definition of the importance of these residues in the interaction interface.
Lastly, although our analysis has focused primarily on the use of reverse two-hybrid analyses to investigate the binding site on JNK for TI-JIP, it is also possible that these JNK residues will also be involved in the interactions of JNK with other activators and/or substrates. Thus, Mooney and Whitmarsh (30) demonstrated that JNK2 mutants that did not bind JIP-1 were not activated by MKK4 and could not bind c-Jun. In our study the JNK1 mutants L131R, R309W, and Y320H were not activated efficiently by MKK4 or MKK7 (Fig. 6), suggesting that they are no longer able to recognize the KIMs of either of these upstream kinases. Given that the 11-mer peptide of JIP-1 is sufficient for interaction with JNK1 (15), it is also unlikely that our JNK1 mutants interact with full-length JIP-1. We have also shown that the TI-JIP peptide shows competitive inhibition with respect to the c-Jun substrate in protein kinase assays (41), and this has led us to question whether the residues we identified mediate interactions with c-Jun. Although we have attempted to assess how these changes of L131R, R309W, and Y320H would also impact on c-Jun binding by JNK1, we have been hampered by the autoactivation of c-Jun constructs in our yeast two-hybrid systems despite extensive titration of its expres-sion levels. 3 Furthermore, we have been unable to detect coprecipitation of wild-type JNK1 or the mutants we identified on recombinant GST-c-Jun(1-135) immobilized on glutathione-Sepharose by immunoblotting. 3 Such in vitro binding assays of JNK1-c-Jun interaction are likely hampered compared with the JNK2-c-Jun assessments made by Mooney and Whitmarsh (30) because JNK1 has ϳ25-fold lower affinity for c-Jun compared with JNK2 binding under the same conditions (2). Future analyses are required to evaluate how the residues implicated in all of the studies to date impact on binding to c-Jun as well as other transcription factor substrates such as Elk or ATF-2 (15) or other proteins known to interact with, and inhibit, JNK activity such as glutathione S-transferases (42), retinoblastoma protein (43), Hsp72 (44), or p57 Kip2 (45).
In summary, our study is the first to combine PCR mutagenesis with a reverse two-hybrid screening protocol to assess the interface between a MAPK and a binding partner. This reverse screening protocol should prove useful in future studies to identify interaction interfaces of other protein complexes in the absence of structural data. In our study of JNK and TI-JIP, the information gained using this approach together with recent information from co-crystallization analyses should further our understanding of the regulation of JNK proteins by the JIP-1 scaffold. Because of the inhibitory properties of TI-JIP, this information will also prove useful in the design of small chemical inhibitors of JNK with superior potency and specificity.