JNK1 (c-Jun N-terminal kinase 1) plays a crucial role in the regulation of obesity-induced insulin resistance and is implicated in the pathology of Type 2 diabetes. Its partner, JIP1 (JNK-interacting protein 1), serves a scaffolding function that facilitates JNK1 activation by MKK4 [MAPK (mitogen-activated protein kinase) kinase 4] and MKK7 (MAPK kinase 7). For example, reduced insulin resistance and JNK activation are observed in JIP1-deficient mice. On the basis of the in vivo efficacy of a cell-permeable JIP peptide, the JIP–JNK interaction appears to be a potential target for JNK inhibition. The goal of the present study was to identify small-molecule inhibitors that disrupt the JIP–JNK interaction to provide an alternative approach for JNK inhibition to ATP-competitive inhibitors. High-throughput screening was performed by utilizing a fluorescence polarization assay that measured the binding of JNK1 to the JIP peptide. Multiple chemical series were identified, revealing two categories of JIP/JNK inhibitors: ‘dual inhibitors’ that are ATP competitive and probably inhibit JIP–JNK binding allosterically, and ‘JIP-site binders’ that block binding through interaction with the JIP site. A series of polychloropyrimidines from the second category was characterized by biochemical methods and explored through medicinal-chemistry efforts. As predicted, these inhibitors also inhibited full-length JIP–JNK binding and were selective against a panel of 34 representative kinases, including ones in the MAPK family. Overall, this work demonstrates that small molecules can inhibit protein–protein interactions in vitro in the MAPK family effectively and provides strategies for similar approaches within other target families.
MAPKs (mitogen-activated protein kinases) play an essential role in the regulation of biological responses such as cell growth, proliferation, differentiation, programmed cell death and inflammatory reactions . The JNKs (c-Jun N-terminal kinases) belong to a subfamily of the MAPKs and were first identified by their ability to phosphorylate the N-terminal transactivation domain of the transcription factor c-Jun [2,3]. Additional substrates include other transcriptional factors such as ATF2 (activating transcription factor 2), Elk1 (E-26-like protein 1), NFAT (nuclear factor of activated T-cells), and p53, as well as non-nuclear proteins, including the Bcl-2 family and EGFR (epidermal growth factor receptor) . The JNK signalling pathway is activated in response to environmental stresses, cytokines and growth factors, establishing the role of JNKs as critical mediators of extracellular stimuli .
There are three JNK genes in mammals (JNK1, JNK2 and JNK3), which produce at least ten different isoforms of JNK through alternative splicing. JNK1 and JNK2 are expressed ubiquitously, whereas JNK3 is expressed predominantly in the heart, brain and testis . It has been reported that the specific functions of each JNK isoform and its substrates vary in mouse disease models with some degree of redundancy, which could result from regulatory cross-talk between the isoforms [7,8]. Since JNKs regulate a number of transcription factors that have been linked to the regulation of immune responses, JNK was pursued initially as a drug target for chronic inflammatory diseases. The strong evidence linking the JNK signalling pathway to neuronal cell death also implicated the therapeutic potential of JNK inhibition in stroke and Parkinson's disease . Recent studies further established the critical role of JNK1 in the development of Type 2 diabetes, possibly by promoting insulin resistance, suppressing insulin biosynthesis and facilitating pancreatic failure . It has been reported that JNK1 activity is elevated in tissues under diabetic conditions, and that the activation of the JNK1 pathway interferes with insulin action and pancreatic β-cell function . In addition, mice lacking the JNK1 gene exhibit a phenotype of lowered adiposity, and improved insulin sensitivity and signalling. Overall, JNK represents an attractive therapeutic target for a variety of disorders and has triggered extensive drug-discovery efforts. Several ATP-competitive JNK inhibitors have advanced into clinical trials .
JIPs (JNK-interacting proteins) were first identified through yeast two-hybrid studies . In mammals, there are four JIP family members, JIP1–JIP4. All JIP proteins are highly expressed in the brain, whereas JIP1 and JIP2 are also abundant in pancreatic β-cells . Studies indicate that JIPs facilitate JNK activation by providing a scaffold for JNK and its upstream activators, including MLK (mixed-lineage kinase) and MKK7 (MAPK kinase 7) . JNK is activated by dual phosphorylation of threonine and tyrosine residues in its activation loop by MKK4 (MAPK kinase 4) and MKK7, which are in turn activated by MLK. Disruption of the Jip1 gene in mice prevents the activation of JNK induced by extracellular stress , suggesting that it has a critical role in the JNK signalling pathway. Since the full activation of JNK requires the assembly of all three components (JNK, MLK and MKK7) on to the same JIP molecule, it is speculated that JNK activation could be inhibited by preventing the association of these components. JIP 11-mer, an 11-amino-acid peptide derived from the JNK-binding domain of JIP1 , inhibited JNK1 activity potently in vitro with a Kd value of 0.42 μM . Furthermore, JIP1–HIV-Tat (transactivator of transcription)–FITC, a cell-permeable JNK-inhibitory peptide designed by fusing the 20-amino-acid JIP1 peptide to a carrier peptide, was highly effective in the treatment of diabetes in an animal model . The intraperitoneal administration of this peptide markedly improved insulin resistance and glucose tolerance in diabetic mice. These data suggest that the binding of compounds to the interface of JIP and JNK could potentially prevent JNK activation, thereby providing an alternative to the traditional ATP-competitor approach.
Although targeting protein–protein interactions with small molecules poses many challenges, some progress has been made in recent years with a variety of approaches . Heo et al.  reported the co-crystal structure of JNK1 in complex with the JIP 11-mer, showing that the JIP peptide binds to a pocket on the C-terminal lobe of JNK1 on the side opposite from the ATP pocket. Computational analysis indicates that the well-defined JIP–JNK interface is likely to bind to drug-like compounds , and a detailed sequence analysis of the JIP 11-mer identified probable hotspots where small molecules could disrupt binding [16,17,21]. Recently, Stebbins et al.  reported the identification of a JNK inhibitor that was proposed to target the JIP–JNK interaction site.
Overall, structural and biochemical evidence generated to date, as well as the in vivo efficacy determined using a cell-permeable JIP peptide, suggests that the JIP–JNK pocket is druggable by small-molecule inhibitors, and that binding at this site could interfere with JNK activation and potentially treat diabetes. Targeting JNK inhibition by preventing the binding of JIP and JNK provides advantages over ATP-competitive inhibition, with the potential for high specificity and efficacy. As a result, an HTS (high-throughput screening) campaign was initiated in search of small-molecule compounds that disrupt the JIP–JNK interaction.
His6–JNK1α1 (full-length protein, 384 amino acids), which contained a His6-tag at the N-terminus and a thrombin site between the His-tag and the JNK1α1 protein, was expressed in 1 litre of Escherichia coli BL21(DE3) cells in Terrific Broth (Invitrogen) and induced with 200 μM isopropyl β-D-thiogalactoside at 20 °C overnight. All purification steps were carried out at 4 °C. Cells were lysed by passing twice through a microfluidizer in 200 ml of lysis buffer consisting of 50 mM Tris/HCl (pH 7.4), 300 mM NaCl, 10 mM imidazole, 10% glycerol and 2 mM TCEP [tris-(2-carboxyethyl)phosphine] containing protease inhibitors (Roche). The lysate was centrifuged at 41000 g for 60 min using an FO650 rotor and an Allegra 64R centrifuge (Beckman Coulter), and the supernatant was allowed to batch bind for 120 min with 5 ml of Ni-NTA (Ni2+-nitrilotriacetate) superflow resin (Qiagen). The resin was collected by centrifugation at 2000 g for 5 min, placed into an XK16/60 column (GE Healthcare) and washed with lysis buffer until a baseline was reached; the baseline was determined using an A280 signal and baseline was reached when this stabilized. Then, the sample was washed with 2 column vol. of 50 mM Tris/HCl (pH 7.4), 300 mM NaCl, 20 mM imidazole and 2 mM TCEP, and eluted with 50 mM Tris/HCl (pH 7.4), 150 mM NaCl, 300 mM imidazole and 2 mM TCEP. The pooled sample was loaded on to an S200 column equilibrated in PBS containing 2 mM TCEP to remove aggregated protein. The aliquoted protein was flash-frozen and stored at −80 °C. The estimated yield of His6–JNK1α1 was 30 mg/l of cells (see Supplementary Figure S1A at http://www.BiochemJ.org/bj/420/bj4200283add.htm). Biotinylated full-length JNK1α1 was prepared by first treating the purified JNK1α1 with thrombin to remove the His6-tag. The protein was then conjugated with LC-biotin (Thermo Fisher Scientific), following the manufacturer's instructions, and the excess biotin was removed by desalting. The stoichiometry of biotin to JNK1α1 was determined using the HABA [2-(4′-hydroxyazobenzene)-2-carboxylic acid] displacement method , following the instructions from the manufacturer (Thermo Fisher Scientific). The resulting JNK1α1 had an average of 4.1 biotin molecules/JNK molecule. Activated JNK1 (His6–JNK1α1_364) was prepared and activated as described in  using purified MKK4 and MKK7β (JNK1/MKK4/MKK7β, 20:1:1). All protein concentrations were determined using the Bradford reagent (Bio-Rad Laboratories) and BSA as a standard. All other reagents were obtained from Sigma–Aldrich.
GST (glutathione transferase)–MKK7β1-3E (full-length protein, 419 amino acids) was a fusion protein containing a thrombin site between the GST-tag and MKK7β1-3E, and was activated mutationally at three positions in the activation loop (3E: S271E, T275E and S277E). The expression and purification were similar to that of JNK1α1 except that the cells were induced at 23 °C overnight, buffer A consisting of 50 mM Tris/HCl (pH 7.4), 10% glycerol, 300 mM NaCl and 5 mM DTT (dithiothreitol) was used as the lysis buffer, and 25 ml of glutathione Sepharose (GE Healthcare) was used for binding. The column was washed with 2 column vol. of buffer A without glycerol, then with 2 column vol. of buffer B [50 mM Tris/HCl (pH 7.4), 150 mM NaCl and 2 mM DTT], followed by elution with buffer B containing 20 mM GSH. Then, the eluate was dialysed against 50 mM Tris/HCl (pH 7.4), 20% glycerol, 150 mM NaCl, 0.1 mM EGTA and 1 mM DTT. The estimated yield of this protein was 65 mg/l of cells.
The expression and purification of mouse GST– MKK4 (amino acids 35–397; 397 residues in the full-length protein), which contained an N-terminal GST-tag and a thrombin site between the GST-tag and MKK4, was similar to that of GST–MKK7β1-3E except that the storage buffer contained 50 mM Tris/HCl (pH 7.5), 150 mM NaCl, 0.1 mM EGTA, 50% glycerol and 0.1% 2-mercaptoethanol. To activate MKK4, 0.1 mg/ml GST–MEKK1 [MAPK/ERK (extracellular-signal-related kinase) kinase kinase 1]–His6 was incubated with 4 μM GST–MKK4 and 0.1 mM ATP in a buffer consisting of 50 mM Tris/HCl (pH 7.4), 0.1 mM EGTA, 0.1% 2-mercaptoethanol and 10 mM magnesium acetate for 30 min at 30 °C. Then, 300 mM NaCl and 15 mM imidazole were added to the mixture, which was passed over a 2 ml Ni-NTA column to capture the activated MEKK1 protein. The flow-through was concentrated to >1 mg/ml, dialysed against 50 mM Tris/HCl (pH 7.5), 150 mM NaCl, 0.1 mM EGTA and 0.1% 2-mercaptoethanol. The dialysate was then diluted with an equal volume of the dialysis buffer containing 0.04% Brij-35 and 540 mM sucrose. The estimated yield of GST–MKK4 was 20 mg/l of cells (Supplementary Figure S1E).
GST–MEKK1–His6 (amino acids 826–1511; 1511 residues in the full-length protein), containing an N-terminal GST-tag, a C-terminal His6-tag and a PreScission protease (GE Healthcare) site between the GST and MEKK1, was expressed and purified as described for GST–MKK4, except that buffer C [50 mM Tris/HCl (pH 7.4), 300 mM NaCl, 10% glycerol, 10 mM imidazole and 2 mM TCEP] was used as the lysis buffer, and 20 ml of Ni-NTA superflow resin was used for binding. The column was washed with 2 column vol. of buffer C containing 20 mM imidazole, then with 2 column vol. of buffer D [50 mM Tris/HCl (pH 7.4), 150 mM NaCl, 20 mM imidazole and 2 mM TCEP], followed by elution with buffer D containing 250 mM imidazole. The protein was dialysed against 50 mM Tris/HCl (pH 7.4), 150 mM NaCl, 0.1% 2-mercaptoethanol and 0.1 mM EGTA. The dialysate was then diluted with dialysis buffer containing 540 mM sucrose and 0.04% Brij-35. The estimated yield of GST–MEKK1–His6 was 2 mg/l of cells.
His6–JNK2α2 (amino acids 1–364; 424 residues in the full-length protein), which contained a His6-tag at the N-terminus and a thrombin site between the His-tag and the JNK2α2 protein, was prepared using the same procedures as JNK1, except that the NaCl concentration was 500 mM in the lysis, wash and elution buffers. The pooled sample was loaded on to an S200 column equilibrated in 50 mM Tris/HCl (pH 7.4), 500 mM NaCl and 2 mM TCEP to remove aggregated protein. Monomeric protein was pooled, then diluted with an equal volume of 540 mM sucrose, 50 mM Tris/HCl (pH 7.4), 100 mM NaCl and 0.2 mM EGTA. The estimated yield of this protein was 30 mg/l of cells (Supplementary Figure S1B). His6–JNK3α2 (amino acids 40–402; 464 residues in the full-length protein), containing an N-terminal His6-tag, was prepared according to the same procedures as JNK2α2, except that the BL21 cells were induced at 18 °C overnight. The estimated yield of His6–JNK3α2 was 160 mg/l of cells (Supplementary Figure S1C).
His6–JIP1 (full-length protein, 711 amino acids), containing a His6-tag at the N-terminus and a thrombin site between the His-tag and the JIP1 protein, was expressed using the Sf21 insect-cell Bac-to-Bac expression system (Invitrogen). A total of 2 litres of cells were harvested 3 days post infection by the BIIC (baculovirus-infected insect cells)/TIPS (titreless infected-cells preservation and scale-up) method  and processed using the same procedures as for unlabelled JNK1α1 except for the following buffer modifications: the concentration of NaCl was 500 mM in the lysis and wash buffers, and 100 mM in the elution buffer; and 0.1 and 0.05% Triton X-100 were added to the lysis and wash buffers respectively. Before elution, an additional wash was performed using 2 column vol. of wash buffer containing 100 mM NaCl and no Triton X-100. The peak was pooled and loaded on to a 1-ml HiTrap Q column (GE Healthcare) equilibrated with 50 mM Tris/HCl (pH 8.0), 100 mM NaCl and 2 mM TCEP. The column was washed with buffer until a baseline was reached then developed with a 40-column-vol. gradient to 50 mM Tris/HCl (pH 8.0), 500 mM NaCl and 2 mM TCEP. The peaks were analysed by SDS/PAGE, and the JIP peaks were pooled and dialysed overnight against 50 mM Tris/HCl (pH 7.4), 150 mM NaCl, 10% glycerol and 2 mM TCEP. The estimated yield of His6–JIP1 was 1.8 mg/l of cells (Supplementary Figure S1D).
All JIP peptides, containing the JNK-docking motif , were synthesized by New England Peptide with a TAMRA (6-carboxytetramethylrhodamine) label at the N-terminus and a free acid at the C-terminus (TAMRA–JIP1, TAMRA–JIP2 and TAMRA–JIP3). The sequence for the JIP 11-mer was RPKRPTTLNLF (amino acids 153–163 of JIP1); the JIP2 peptide was HKHRPTTLRLT (amino acids 134–144 of JIP2) and the JIP3 peptide was RKERPTSLNVF (amino acids 202–212 of JIP3). Peptide KRELVEPLTPSGEAPNQALLR, derived from the EGFR sequence, was ordered from the American Peptide Company.
FP (fluorescence polarization) assay
To measure the binding of the JIP 11-mer with JNK1, an FP assay was developed and optimized for HTS. Under standard assay conditions, 100 μM compounds (ten times the final concentrations) were pre-incubated with 0.8 μM His6–JNK1α1 (non-activated) for 30 min at room temperature (22 °C) in the assay buffer consisting of 10 mM Hepes/KOH (pH 7.4), 150 mM NaCl, 10 mM MgCl2, 0.005% Brij-35, 0.1% 2-mercaptoethanol and 0.05% BSA. Then, TAMRA–JIP1 was added to the reaction mixture at a final concentration of 5 nM in a 384-well plate (30 μl per well). Plates were read on an Analyst AD plate reader (Molecular Devices) with a G-factor of 0.8. When ATP was included in the assay, 1 mM ATP was incubated with JNK1 for 30 min before compounds were added. The FP assay for all JIP and JNK isoforms, which included nine binding assays encompassing all possible JIP- and JNK-isoform combinations (JIP1, JIP2 and JIP3, and JNK1, JNK2 and JNK3), was developed under the same conditions as the JIP1/JNK1 assay (see above).
Full-length JIP1/JNK1-binding assay
AlphaScreen™ technology  was utilized to develop a full-length JIP- and JNK-binding assay. The AlphaScreen™ histidine detection kit was purchased from PerkinElmer. Biotinylated full-length JNKα1 (non-activated) and full-length JIP1 (His6–JIP1) were pre-incubated at a final concentration of 10 nM each in 15 μl of assay buffer containing 10 mM Hepes/KOH (pH 7.4), 150 mM NaCl, 10 mM MgCl2, 0.005% Brij-35, 0.1% 2-mercaptoethanol and 0.05% BSA in a 384-well plate for 10 min at room temperature. Then, 10 μl of streptavidin donor beads and nickel chelate acceptor beads (20 μg/ml) were added to the reaction mixture, bringing the final volume to 25 μl per well. After an additional incubation of 60 min in the dark as instructed by the manufacturer, the resulting fluorescence signal of the acceptor beads was read on an EnVision reader (PerkinElmer).
pJNK (pre-activated JNK) assay
The activated JNK1 was tested using the Kinase-Glo™ Plus assay (Promega), which quantified the amount of ATP remaining in solution following a kinase reaction. Compounds were added to 40 nM activated JNK1 catalytic domain (His6–JNK1α1–364) in a buffer containing 20 mM Hepes/KOH (pH 7.5), 10 mM MgCl2, 0.01% Tween 20 and 1 mM DTT, followed immediately by an addition of 200 μM EGFR peptide and 50 μM ATP in the same buffer to initiate the reaction. The reaction mixture was incubated in a volume of 20 μl per well in a 384-well plate at room temperature for 120 min, then an equal volume of the Kinase-Glo™ reagent was added to the mixture. After 25 min of incubation, the plates were read on an Analyst AD plate reader. Inhibition of JNK1 activity was also tested with the protein substrate ATF2, using a procedure similar to the published method . Full-length His6–JNK1α1 (384 amino acids), activated as described previously , and GST–ATF2 (amino acids 19–96) were obtained from the Division of Signal Transduction Therapy, University of Dundee (Dundee, U.K.). Enzyme kinetic studies using ATF2 as a phospho-acceptor substrate were conducted as detailed previously  with the following modifications. Reactions contained 0.5 nM enzyme, varied ATP (1–32 μM) with fixed ATF2 (1 μM, ∼2×Km) or varied ATF2 (0.2–3.2 μM) with fixed ATP (15 μM, ∼3×Km), 0.2–1.2 μCi of [γ-33P]ATP (PerkinElmer) per reaction, and the appropriate inhibitor or DMSO (2.5%) in 20 mM Hepes/KOH (pH 7.5) containing 10 mM MgCl2 and 1 mM DTT. Phosphorylated protein substrate was captured in 96-well MultiScreenHTS-PH phosphocellulose filter plates (Millipore) and quantified by scintillation counting using 50 μl of MicroScint™ 20 (PerkinElmer). The conversion of the limited substrate into product was <10%. The kinetic analysis was conducted using GraphPad Prism version 5 (GraphPad Software). Data were fitted to competitive, non-competitive and uncompetitive equations  using a non-linear regression (global fit). Inhibition models were compared statistically to choose the best model, which was confirmed further by inspecting Lineweaver–Burk double-reciprocal plots and plots of IC50 against substrate concentration .
Assessing druggability through HTS
In order to identify small molecules that inhibit the JIP–JNK interaction, a binding assay was used to screen an internal library containing more than 2 million compounds. An FP assay that measured the binding of TAMRA–JIP1 to full-length non-activated JNK1 was developed and optimized for screening. The Kd was determined to be 0.8 μM, at which an average shift of 45 mP (millipolarization) units was observed (Figure 1A). Non-specific binding was determined by comparing the binding in the presence or absence of saturating unlabelled JIP 11-mer, and was shown to be negligible. To determine the optimal concentration of TAMRA–JIP1 for screening, a titration was performed using a JNK1 concentration of 0.8 μM (Kd) (results not shown). A concentration of 5 nM TAMRA–JIP1 was selected, which provided the largest assay window. The S/B ratio (signal-to-background ratio) was 40±5 with a CV (coefficient of variation) of 2.0±0.5%. The S/B ratio was calculated as mean mP shift (non-compound-treated state)/mean mP shift (compound-treated state). In a time-course study, the JIP–JNK interaction was shown to be very rapid and the assay signal remained stable for at least 2 h. Once the binding was at equilibrium, an excess amount of unlabelled JIP 11-mer quickly displaced the labelled peptide, confirming the reversibility of this interaction (results not shown). In order to validate the assay, the unlabelled JIP 11-mer was used as a control inhibitor and its IC50 value was determined to be 0.4 μM (Figure 1B). This value was consistent with a value of 0.42 μM (Kd) obtained by ITC (isothermal titration calorimetry) . A final concentration of 0.8 μM JNK1, 5 nM TAMRA–JIP1 and 10 μM test compound was chosen, and the average Z′-factor  remained in the range 0.7–0.9 throughout the entire screen (Figure 1C). To identify false positives, all plates were pre-read to measure the intrinsic fluorescence intensity of the compounds before the reaction was initiated. A hit (i.e. active-target identification) cut-off of 30% inhibition was chosen on the basis of the average percentage inhibition of all compounds screened plus 3×S.D. (Figure 1D). After excluding compounds with intrinsic fluorescence, the primary hit rate of compounds exhibiting >30% inhibition was 0.36%, for a total of 7336 hits. All primary hits were re-tested individually at 10 μM and 1197 of them (19%) were confirmed. A selection of structurally representative compounds was chosen for IC50 determination. Unexpectedly, a majority of the hits were structurally similar to known ATP-site inhibitors, even though the assay focused on the JIP–JNK interaction site. In order to demonstrate that these compounds were binding at the ATP pocket, the FP assay was repeated using the confirmed hits after pre-incubating JNK with a saturating concentration of ATP (1 mM). A small shift of the binding affinity for the JIP 11-mer and JNK1 was observed in the presence of 1 mM ATP (results not shown). It was predicted that compounds binding at the ATP site would lose their activity in the presence of saturating ATP, whereas those binding at the JIP site would display similar activity regardless of the ATP concentration. The ratios of compound potencies in the presence or absence of ATP were measured, and the compounds were classified as probable JIP-site binders or dual inhibitors for further evaluation.
TAMRA–JIP1 and full-length JNK1 FP assay development and HTS of a compound library
Characterization of lead series
The results from HTS suggested that two novel types of JIP/JNK inhibitor had been discovered: (i) potential dual inhibitors, which interfered with the JIP–JNK interaction allosterically by binding to the ATP site; and (ii) potential JIP-site binders, which interacted exclusively with the JIP site. We hypothesized that these two types of inhibitor could be distinguished biochemically from each other and from the classical ATP-site inhibitors that did not disrupt the JIP–JNK interaction, using the assays and outcomes summarized in Table 1. Owing to ATP competition, dual inhibitors were predicted to be active in the FP assay in the absence of ATP, but inactive in the presence of saturating ATP. For the same reason, they should also be active in the pJNK assay, which monitored the phosphorylation of EGFR peptide by active JNK in the presence of ATP. It is also possible that the dual inhibition resulted from compounds binding at the JIP site and altering the binding affinity for the ATP site, although evidence strongly suggested that the dual inhibitors bind to the ATP site. In contrast, JIP-site binders were predicted to be active in the FP assay both in the presence and absence of ATP, and inactive in the pJNK assay, due to the lack of competition with ATP or EGFR peptide. Classical ATP-site inhibitors that did not disrupt JIP–JNK binding were predicted to be inactive in the FP assay in both the presence and the absence of ATP, and active in the pJNK assay. Our data revealed that <1% of the confirmed hits from HTS were potential JIP-site binders, and the reminder belonged to the dual-inhibitor class. It is interesting to note that not all ATP-site inhibitors displaced JIP allosterically in the FP assay. A structurally diverse selection of 56 potent JNK inhibitors that probably bind in the ATP site was tested in the FP assay, and although 73% of them displaced JIP, 27% had no effect (results not shown). Cross-talk between the two sites was noted, with the binding of the JIP peptide apparently causing a conformational change in the ATP site in the crystal structure and reducing the biochemical affinity for ATP 3-fold .
|Dual inhibitors||JIP-site binders||ATP-site inhibitors|
|Pre-activated JNK assay (EGFR as substrate)||Active||Inactive||Active|
|Full-length JIP/JNK assay||Active||Active||Inactive|
The structural and biochemical data for the representative compounds from each of these types of inhibitor are shown in Table 2. Compound 1 is a representative of the dual-inhibitor class on the basis of its dual actions towards both the ATP and JIP sites, but probably binding in the ATP site on the basis of the loss of JIP–JNK inhibition in an excess of ATP and on its predicted binding mode. In a study by Lippa et al. , closely related analogues of compound 1 were docked to the ATP site. Compound 2 belongs to the polychloropyrimidine series and is characterized as a potential JIP-site binder. Compound 3 is a typical ATP-competitive aminopyrimidine-series inhibitor, probably binding at the ATP site and inactive in the JIP–JNK assay.
|Compound||IC50 (μM)…||JIP 11-mer/JNK−ATP||+ATP||Full-length JIP/JNK−ATP||pJNK phosphorylation EGFR||ATF2|
|Compound||IC50 (μM)…||JIP 11-mer/JNK−ATP||+ATP||Full-length JIP/JNK−ATP||pJNK phosphorylation EGFR||ATF2|
Ki (μM) values determined from global fits of data to a competitive inhibition model with varied ATP and fixed GST–ATF2 for compound 1 and compound 3, and varied GST–ATF2 and fixed ATP for compound 2.
To evaluate further whether inhibitors of JIP–JNK interaction were capable of inhibiting other signalling cascades activated by JNK, an additional JNK-activity assay was performed using ATF2 protein instead of EGFR peptide as a substrate. Kinetic parameters for JNK1α1 with GST–ATF2 (amino acids 19–96) were as follows: kcat=6.3±1.5 min−1; Km,ATF2=0.6±0.2 μM, Km,ATP=4.9±0.8 μM (mean values±S.D., n=3). These values were comparable with the published values for full-length ATF2 . Compounds from all three series inhibited ATF2 phosphorylation, although compounds 1 and 2 were an order of magnitude less potent than the JIP 11-mer in both the JIP–JNK-binding assay and the ATF2-phosphorylation assay (Table 2). From statistical comparison of inhibition models using non-linear regressions with global fitting of data (see the Experimental section), compounds 1 and 3 showed competitive inhibition with ATP and pure non-competitive inhibition with ATF2. Conversely, compound 2 showed competitive inhibition with ATF2 and pure non-competitive inhibition with ATP (results not shown). These results were confirmed further by inspecting the dependence of IC50 on the concentration of an appropriate substrate (Figure 2) and Lineweaver–Burk plots (see Supplementary Figure S2 at http://www.BiochemJ.org/bj/420/bj4200283add.htm), which provided a useful visual diagnostic of the inhibition mode. A linear increase of IC50 with increasing substrate concentration, as seen for the appropriate plots in Figure 2, is a hallmark of competitive inhibitors. A zero slope was consistent with pure non-competitive inhibition (i.e. no significant cross-talk between substrate-binding sites). Double-reciprocal plots of 1/v against 1/[S] at various inhibitor concentrations (Supplementary Figure S2) tend to intercept at the y-axis for competitive inhibitors (compounds 1 and 3 with varied ATP; compound 2 with varied ATF2) and at the x-axis for pure non-competitive inhibitors (when the other substrate was varied). It has been proposed that ATF2 shares the docking site for JNK1 with the JIP 11-mer, on the basis of the sequence similarity between the JIP1 peptide and the JNK-binding domain of JNK substrates (ATF2 and c-Jun), as well as the in vitro data indicating the JIP 11-mer as a potent inhibitor of ATF2 phosphorylation by JNK1 . Consequently, JIP-site-binders were expected to prevent the JNK1-mediated phosphorylation of ATF2 in a manner similar to that of the JIP 11-mer. Indeed, the inhibition mode of compound 2 mimicked that of the JIP 11-mer, which was shown to be competitive with ATF2 [28,33] and c-Jun . The Ki values for JIP 11-mer (0.6 μM) interaction with JNK1α1 (Table 2) were within 2-fold of the value reported for JNK1 [17,21] and JNK2 . Interestingly, JNK activity was not inhibited by either the JIP 11-mer or JIP-site-binders in the pJNK assay that used EGFR peptide as a substrate (Table 2). This is probably because EGFR peptide lacks the recognition sequence for the JIP site in JNK1.
IC50 values as a function of substrate concentration for JNK1α1-catalysed GST–ATF2 phosphorylation with different types of inhibitors
In order to demonstrate that the JIP-site-binders and dual inhibitors identified from HTS, which measured the binding of a JIP peptide, were also able to displace full-length JIP from its complex with JNK, an assay was developed using the AlphaScreen™ technology  that monitored the binding between full-length JIP and JNK proteins. Each protein was conjugated to donor or acceptor beads respectively, and the subsequent binding between the two proteins brought the beads into close proximity, thereby enabling the transfer of energy. A time-course assay and protein titration at a 1:1 ratio of each protein were performed in order to determine the binding affinity for full-length JIP and JNK. In the time-course study, the assay signal peaked at 10 min of pre-incubation between the two proteins, followed by a time-dependent decrease in the signal (results not shown). It was possible that the full-length JIP was unstable or aggregated gradually at room temperature, and the degraded or aggregated JIP was then unable to form an active complex with JNK. After 10 min of pre-incubation, the Kd for the complex of full-length proteins was determined to be 20 nM (see Supplementary Figure S3 at http://www.BiochemJ.org/bj/420/bj4200283add.htm), in contrast with a Kd of 800 nM in the FP assay, where the JIP peptide was used. It is unclear why there was a 40-fold shift of JIP–JNK binding affinity in these two assays. Final conditions of 10 nM proteins with a 10-min pre-incubation were used in the assay, achieving an average Z′-factor of 0.7 in a 384-well plate format. The average S/B ratio was 10±2, with a CV of 5±1%. Compounds 1, 2 and 3, and the JIP 11-mer were tested in this assay (Table 2), and had potencies comparable with those obtained in the FP assay. Furthermore, both compound 1 and compound 2 showed full inhibition identical with that using the JIP 11-mer (results not shown). These data indicated that the potential JIP-site binder was able to displace the full-length JIP in a manner similar to the displacement of the JIP peptide from JNK, increasing confidence in its hypothesized binding mechanism. The dual inhibitor also inhibited the full-length protein binding, probably through the allosteric effect of its binding at the ATP site. As expected, compound 3, which is an ATP-site inhibitor and inactive in the FP assay, had no effect on the JIP–JNK protein binding.
SAR (structure–activity relationship) of the JIP-site-binding lead series
The novelty of the biochemical data for compound 2, suggesting that this compound binds in the JIP site, prompted the development of a basic SAR study within the series to understand further the nature of the binding. Changes at both the polychloropyrimidine core and the 2-position amine highlighted the structural elements important for binding.
Table 3 shows the results of compounds in the JIP/JNK FP assay that contained modifications to the 2-aniline moiety of compound 2, keeping the 4,6-bis-trichloromethyl pyrimidine moiety constant. Both aromatic and aliphatic amines were tolerated in the 2-position, provided that there was an appropriately placed tertiary amine. When the initial hit, compound 2, was replaced by a 3-position tertiary amine one methylene unit away from the aromatic ring and a 4-position ethoxy moiety, the IC50 value was 1.3 μM. Compound 4 illustrated that the 3-position tertiary amine located one methylene unit away from the aromatic ring alone with no 4-ethoxy moiety maintained the IC50 value in the single-digit micromolar range. However, if the methylene spacer was removed and the tertiary amine was attached directly to the aromatic ring (compound 5), the inhibitory activity was lost. There was some flexibility as to where the tertiary amine with the methylene spacer was placed: compound 6 has the tertiary amine with a one-methylene spacer in the 4-position, and the IC50 value was 3.2 μM. The unsubstituted aniline derivative, compound 7, which lacks a tertiary amine, had no activity. The aliphatic amine compounds 8–11 followed the same pattern. Compounds 8 and 9 have a two-carbon spacer between the two amines, whereas compound 10 has a four-carbon spacer; these three compounds had IC50 values in the 3–6 μM range, illustrating again that there was some flexibility in the placement of the tertiary amine. Compound 11, a cyclohexyl amine derivative, lacks a tertiary amine and therefore showed no activity in the FP assay. Similarly, the unsubstituted 2-amino derivative, compound 12, was inactive.
Table 4 shows changes made to the 4,6-bis-trichloromethyl pyrimidine core. Results showed that all six chlorine atoms were necessary for activity, but the pyrimidine can be replaced by a triazine and still maintain activity. The substitution of a hydrogen atom for just one of the six chlorine atoms in compound 2 (compound 13) resulted in a >20-fold loss of potency (resulting in IC50 values of 1.3 and 28.8 μM for compounds 2 and 13 respectively). Substitution of a phenyl moiety for one of the trichloromethyl substituents (compound 14) resulted in complete loss of activity. The isosteric 4,6-bis-t-butyl pyrimidine analogue, compound 15, was also inactive in the FP assay, indicating that the electronic properties of the 4,6-bis-trichloromethyl pyrimidine core were important for activity. Direct replacement of the pyrimidine core with a triazine (compound 16) was tolerated, as shown by an IC50 of 4.5 μM. Taken together, the SAR indicated two specific structural motifs that were important for interruption of the JIP–JNK interaction: the 4,6-bis-trichloromethyl and an appropriately placed tertiary amine.
The activities of representative compounds from each inhibitor type as well as the JIP 11-mer were evaluated against 34 diverse kinases at 10 μM using the Z′-Lyte technology (SelectScreen™ Kinase Profiling Service, Invitrogen). Since two of the kinases, p38α and PDK1 (3′-phosphoinositide-dependent kinase-1) were assayed in the cascade format (i.e. activity was measured through the activation of substrate kinases), compounds were also evaluated against each substrate kinase separately. As expected, the JIP 11-mer had little off-target activity, with <50% inhibition among all kinases (Figure 3). Compound 2, a JIP-site binder, had even less off-target activity, with <10% inhibition among all kinases tested. Compound 1, a representative dual inhibitor, was significantly less selective than the JIP-site binder, showing >50% inhibition against 11 kinases. Compound 3 is an ATP-site inhibitor and is very similar structurally to a p38α inhibitor studied previously . It was relatively selective for JNK but also inhibited p38α potently in the p38α/MK2 (MAPK-activated protein kinase 2) pathway. Overall, JIP-site binders demonstrated distinctive selectivity against a panel of representative kinases, including the MAPK family, whereas dual inhibitors were considerably less selective. This observation is reasonable given the uniqueness of the JIP pocket compared with the ATP active site.
The selectivity of compound 1, compound 2, compound 3 and the JIP 11-mer at 10 μM against 34 kinases on the basis of the data from the SelectScreen™ Kinase Profiling Service (Invitrogen)
Sequence analysis of the JIP site on JNKs shows that the JIP site is highly conserved among the three JNK isoforms . In order to evaluate the selectivity of the compounds within the JNK family, binding assays between JIP1 and JNK2 and JNK3 were developed using the same FP assay format as for JIP1–JNK1, as described above. The binding affinities of the JIP 11-mers with JNK1, JNK2 and JNK3 were determined (Figures 4A–4C; summarized in Figure 4D). The IC50 values of compounds 1 and 2 against JNK isoforms are shown in Figure 4(E). As expected, the compound potencies were similar (within 5-fold) among the three JNK isoforms. In order to compare the compound potencies against all three JIP and JNK isoforms (JIP1, JIP2 and JIP3 against JNK1, JNK2 and JNK3) accurately, and examine whether JNKs have preferences over different isoforms of JIP, JIP2 and JIP3, binding assays were developed in parallel. TAMRA-labelled JIP2 and JIP3 peptides, containing an appropriate recognition sequence , were utilized to develop FP binding assays under similar assay conditions, and Figure 4(D) shows the Kd of each combination. The data indicated that all JNK isoforms had higher affinities (approx. 3–10-fold) for JIP1 than for JIP2 and JIP3, suggesting that the primary effect of a JIP–JNK inhibitor would be on JIP1. Compounds 1 and 2 were tested in all nine assays and their IC50 values were determined (only data obtained using JIP1 are shown in Figure 4E). Overall, the rank order of compound potencies in JNK1, JNK2 and JNK3 remained the same, demonstrating the similarity of the interactions.
Cross-interaction between the JIP 11-mer and JNK isoforms
A number of drug-discovery efforts focused on the development of ATP-competitive JNK1 inhibitors have been published . However, there are potential concerns associated with ATP-competitive inhibition. For instance, the IC50 of the published ATP-competitive JNK inhibitor, SP600125, was 25–50-fold higher in the cell-based assay compared with the biochemical assay . One explanation is that the high level of endogenous ATP competes with this compound, reducing its efficacy significantly. As an alternative, compounds targeting a non-ATP site could provide advantages since these inhibitors do not need to compete with the high intracellular levels of ATP. On the basis of the critical role of JIP in JNK activation and the in vivo efficacy demonstrated using a JIP peptide, the JIP–JNK interface emerges as a potential target for JNK inhibition, as well as an ideal model system for the application of alternative approaches to identify novel non-ATP-competitive inhibitors. Since the JIP pocket is unique to the MAPK family, it was predicted that chemical series identified from the JIP–JNK screen would be highly selective across the kinome. This prediction was confirmed for the identified JIP-site-binders through the selectivity data of the representative kinases. However, it was demonstrated that, owing to the high sequence homology within the JNK family in the region of the JIP site, the JIP-site-binders did not show significant selectivity across JNK isoforms. Furthermore, it is known that some MAPKs, such as p38α, JNK1 and ERK2, use topologically similar binding sites to interact with the docking-peptide motifs . Specifically, the JIP site on JNK is homologous with the binding sites between ERK2 with pepHePTP (short docking sequences of haemopoietic protein tyrosine phosphatase), and p38α with pepMEF2A (short docking sequences of myocyte enhancer factor 2A). On the basis of the kinase selectivity data, compound 2 was inactive in the p38α/MK2 cascade assay, in contrast with the potent activity of compound 1 and compound 3 in this assay. These data demonstrated the potential selectivity of JIP-site-binders within the MAPK family, and could be explained by the observation that the hydrophobic docking groove between p38α/MK2 and p38α/pepMEF2A or the JNK/JIP 11-mer is conserved, but MK2 binds to p38α with opposite orientation compared with pepMEF2A or the JIP-binding groove . Zhou et al.  suggest that the specificity for each MAP kinase may be achieved by their unique binding interactions and conformational changes. Further studies to investigate the selectivity of JIP-site binders among the homologous sites in the MAPKs will be beneficial. For instance, similar binding assays could be developed to evaluate whether JIP-site-binders have cross-activities towards the binding interfaces of p38α and ERK2 with their substrates.
The biochemical data for compound 2 support the hypothesis that the compounds in this series bind to the JIP site as their mechanism for the interruption of the JIP–JNK interaction. However, the results from the SAR study, namely the identification of two specific structural motifs that are important for interrupting the JIP–JNK interaction (the 4,6-bis-trichloromethyl and an appropriately placed tertiary amine), do not clearly indicate the binding mode of the compounds. There is no obvious structural overlap of the compound footprint with the hotspots identified in the interaction between the JIP 11-mer and JNK [16,17,21]. However, it is intriguing that such a small molecule is capable of interrupting a large protein–protein interaction, so further understanding of the binding mode of compound 2 would be beneficial to the field. Co-crystallization attempts with compound 2 and JNK1 protein using the conditions described in  did not yield crystals. A novel crystal form that was obtained for JNK1 (S. Greasley and P. Chen, unpublished work) did not yield bound inhibitor, probably owing to partial occlusion of the JIP site by a symmetry molecule. Further efforts will be required to find a crystal form that can accommodate the bound JIP-site inhibitor, and the resulting structure should provide details of the mechanism of this novel ligand–protein interaction.
Recently, a different series of small-molecule JNK inhibitors targeting the JIP–JNK interaction site was reported . Using an HTS method similar to that reported here, the authors identified the inhibitor BI-78D3, which had an IC50 value of 0.5 μM in a JIP-peptide/JNK-binding assay and 0.28 μM in the JNK-activity assay. Additionally, the compound inhibited the phosphorylation of cellular JNK substrates and restored insulin sensitivity in mouse models of Type 2 diabetes. We were unable to assess the cellular activity of compound 2 owing to its poor permeability, as predicted using a parallel artificial membrane permeability assay and high cLogP (calculated logarithm of the octanol/water partition coefficient) at pH 7.8. Additional medicinal chemistry will be necessary to develop a permeable analogue of compound 2 for mechanistic comparison with BI-78D3 in cells. Unfortunately, there is also no crystallographic evidence reported for the binding mode of BI-78D3, although NMR and docking studies suggest JIP-site binding.
The identification of dual inhibitors was an unanticipated result of this HTS campaign. One explanation for the mechanism of action of these compounds is that the binding of a dual inhibitor at the ATP site induces a conformational change of JNK1 and distortion of the JIP site, leading to the displacement of JIP peptide. The structural data displaying the alteration of the ATP site after the binding of JIP peptide, as well as the biochemical data indicating a 3-fold shift of ATP-binding affinity in the presence of saturating JIP peptide , support the theory of cross-talk between the ATP site and the JIP site. Interestingly, the dual inhibitor, compound 1, displaced both the JIP 11-mer and full-length JIP from the complex with JNK1 (as detected by various assay technologies), but showed a pure non-competitive inhibition with ATF2. One possible explanation for this result is that, although JNK is believed to share the docking site for JIP and ATF2, additional interaction sites are probably present that may differentiate interaction with an inhibitor. Indeed, affinities of full-length JIP for JNK (determined in the present study) and ATF2 for JNK3α1  were between one and two orders of magnitude higher than for the respective peptides containing the single high-affinity docking site. Since both in vitro and in vivo data have demonstrated positive effects using the JIP peptide in β-cell preservation, glucose uptake and insulin sensitivity , the dual inhibitors could potentially exhibit synergetic effects compared with the ATP-site-only inhibitors. Additional studies, such as determining co-crystal structures and obtaining functional assay data from insulin receptor substrate 1, would be useful to test this hypothesis. Further study will also be needed to address the structural basis of why only particular ATP-competitive JNK inhibitors have inhibitory effects towards JIP binding.
The binding affinity for JIP and JNK shifted significantly when the full-length JIP protein was used instead of the JIP 11-mer. Since the JIP 11-mer in the FP assay has a TAMRA label, it is not surprising that there was a slight shift of Kd from 0.42 μM in the ITC assay , in which unlabelled JIP peptide was used, to 0.8 μM in the FP assay. The IC50 value of unlabelled JIP 11-mer in the FP assay was 0.4 μM, which was consistent with the ITC data. On the other hand, a much higher binding affinity was observed between the full-length JIP and JNK protein (Kd=20 nM). One explanation could be that only part of the energy of the JIP–JNK complex is derived from the JIP–JNK interaction at its docking site. Potential limitations of assay formats could also contribute to the difference. However, the IC50 values of the JIP-site-binders and the JIP 11-mer were retained in the full-length JIP assay as in the FP assay, providing quantitative evidence that enables the usage of peptide–protein-binding assays during HTS in place of protein–protein-binding assays, which are often limited by cost, availability of reagents or amenable assay formats.
In conclusion, the present study has made advances into the challenging area of developing small-molecule inhibitors to target protein–protein interactions, particularly for the MAPK family. Initial evidence on the basis of biochemical methods indicates that the polychloropyrimidine series identified from HTS was able to disrupt the JIP–JNK protein–protein interaction in vitro. The cross-talk discovered between the ATP site and the JIP site could lead to the development of dual inhibitors as therapeutic agents with potentially enhanced efficacy. In addition, the present study suggests that targeting the JIP–JNK interface may provide a better opportunity of achieving overall kinase selectivity, particularly within the MAPK family, compared with targeting the ATP active site. It also validates the approach of using peptides rather than full-length proteins in the high-throughput binding assays and HTS. Future work will be needed to determine the exact binding site and evaluate the cellular activity of this series. Overall, the knowledge obtained and lessons learned can facilitate similar approaches targeting other kinases or protein–protein interaction systems in the future.
We sincerely thank Mary Ellen Banker and Carol Harley for performing the HTS and some additional follow-up experiments. We also thank Karen Siegel, Kristin Abrams, Zhili Song and Lawrence Drew for developing and performing various assays, and Roman Herrera, Beth Lunney, Andy Hardy, Mike Pollastri and Alan Cheng for engaging in scientific discussions and providing helpful suggestions during the course of the project.
activating transcription factor 2
coefficient of variation
epidermal growth factor receptor
isothermal titration calorimetry
c-Jun N-terminal kinase
mitogen-activated protein kinase
MAPK/ERK kinase kinase 1
MAPK-activated protein kinase 2
short docking sequences of myocyte enhancer factor 2A
- S/B ratio