Crystallisation

50 nL of a solution containing 24 mg/mL ZAK_5–309 and 500 µM vemurafenib (20 mM HEPES pH 7.4, 150 mM NaCl, 0.5 mM TCEP, 5% glycerol) were transferred to a 3 well crystallisation plate (Swissci), mixed with 100 nL precipitant solution (100 mM bis-tris-propane pH 6.5, 200 mM sodium malonate, 20% PEG3350, 10% ethylene glycol) and incubated at 4°C. The slowly growing crystal was detected after 14 days and mounted after 42 days in precipitant solution cryoprotected with additional 20% ethylene glycol. Data were collected at Diamond Light Source, analysed, scaled and merged with Xia2(29). The structure was solved by molecular replacement with Phaser(30) using a MLK1 model as a template (PDB ID 3DTC) and refined with Refmac5(31). The model was validated using MolProbity(32). A summary of data collection and refinement statistics is given in Table S3. The model and the structure factors have been deposited with the PDB ID 5HES.

T_M shift assay

The assay was performed according to a previously established protocol.(33) A solution of 2 µM ZAK_5–309 in assay buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 0.5 mM TCEP, 5% glycerol) was mixed 1:1000 with SYPRO Orange (Sigma). The compounds to be tested were added to a final concentration of 10 µM. 20 µL of each sample were placed in a 96-well plate and heated from 25 to 96°C. Fluorescence was monitored using a Mx3005P real-time PCR instrument (Stratagene) with excitation and emission filters set to 465 and 590 nm, respectively. Data were analysed with the MxPro software.

Kinase activity assay

ZAKtide ( ANHWHTVHLRA) was synthesised by Micheal Berne (Tufts Medical School). 20 µM peptide were phosphorylated by 4 nM ZAK_5–309 in the presence of inhibitors (assay buffer 20 mM HEPES pH 7.4, 150 mM NaCl, 0.5 mM TCEP, 5% glycerol, 100 µM ATP, 5 mM MgCl₂). The reaction mixtures were incubated at 37°C for 1 hour before being quenched by the addition of formic acid to a final concentration of 1% (v/v). Both the substrate and product peptides were monitored by RapidFire™ mass spectrometry. Samples were loaded onto a C4 solid phase extraction (SPE) cartridge. The cartridge was then washed to remove non-volatile buffer salts using deionised water containing 0.1% formic acid (solvent A) at a flow rate of 1.5 mL/min for 4.5 s. The sample was eluted with acetonitrile containing 15% deionised water and 0.1% formic acid (solvent B) at a flow rate of 1.25 ml/min for 4.5 s. The eluent was analysed on an Agilent 6530 Accurate-Mass quadrupole time-of-flight (Q-TOF) mass spectrometer operated in positive ionisation mode. Ion chromatograms were extracted for the +3 charge state of the substrate (MW 1340.69 Da) and product (MW 1420.65 Da) and integrated using the RapidFire™ Integrator software. The ratio of product area to product plus substrate area was used as the response, and the baseline signal from a sample inactivated with formic acid prior to addition of substrate was subtracted. The data were fit to the following equation using the nls function in the R programming language:

Mean pIC50 was obtained by taking the geometric mean of pIC50s determined from duplicate experiments.

Isothermal titration calorimetry

Measurements were performed at 37°C on a MicroCal VP-ITC (GE Healthcare). ZAK_5–309 was dialysed overnight into assay buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 0.5 mM TCEP, 5% glycerol). The syringe was loaded with 120 µM ZAK_5–309, the cell was filled with 15 µM vemurafenib. Every 10 minutes 10 µL of the protein solution were injected into the cell for a total of 28 injections. The heat flow data were analysed with the MicroCal ORIGIN software package employing a single binding site model.

Proteomic sample preparation

The samples were prepared as previously described.(34) In brief, protein samples were reduced and alkylated in solution and subjected to chloroform/methanol precipitation. The precipitate was resuspended in 6 M urea in 100 mM Tris pH 7.4, digested with elastase at 37°C overnight and desalted using C18 Sep Pak column cartridges (Waters). After drying down in vacuo samples were resuspended in buffer A (98% H₂O, 2% acetonitrile, 0.1% formic acid).

Tandem mass spectrometry and data analysis

Samples were analysed using a nano-liquid chromatography tandem mass spectrometry (LC-MS/MS) system consisting of a Dionex Ultimate 3000 UPLC coupled to a hybrid quadrupole orbitrap instrument (Q Exactive, Thermo). Samples were loaded at a flow rate of 20 µL/min onto a PepMAP pre-column (C18, 300 µm×5mm, 5 µm particle size, Thermo) for one minute and separated at a flow rate of 250 nl/min on an nEASY column (C18, 75 µm×500 mm, 2 µm particle size, Thermo) for 60 minutes using a gradient of 2%–35% acetonitrile (v/v) in 5% DMSO (v/v) and 0.1% formic acid (v/v). All scans were performed at a resolution of 70,000 at 200 mass/charge and the 15 most abundant precursors were selected for HCD fragmentation. Raw MS data were de novo sequenced by PEAKS Version 7 (Bioinformatics Solutions) with search criteria at 10 ppm for MS¹ and 0.05 Da for MS². A database search (human SwissProt, 85,809 sequences) with subsequent posttranslational modification searches, where all modifications reported in UNIMOD were considered, was then applied to the de novo identified MS/MS spectra. False discovery rates of 1% threshold were applied. MS/MS spectra with phosphorylation modifications were inspected manually.

ZAK kinase substrate specificity

ZAK is a catalytically active kinase.(8) Testing ZAK for its ability to phosphorylate a combinatorial peptide library(35) revealed a unique substrate specificity (Figure 1C and Table S2). Besides preferring threonine as the phosphoacceptor, ZAK was selective for hydrophobic residues at positions +1 (AVIFY) and +3 (VFYWQ) from the phosphoacceptor. The additional preferences (tryptophan at −2, histidines at −1 and +2) have not been previously reported for a kinase. While the sequence flanking the primary ZAK autophosphorylation site T161 ( RFHNHTTHMS) corresponded well to the determined consensus sequence (Figure 1C), the secondary ZAK autophosphorylation site S165 ( HTTHMSLVGT) was a more distant match to the consensus sequence. Interestingly, even the activation segments of the physiological ZAK substrates MAP2K4 ( GQLVDSIAKT) and MAP2K7 ( GRLVDSKAKT) are predicted to be poor ZAK substrates as isolated peptides. These findings suggest that selective and efficient substrate targeting by ZAK involves complex formation with downstream signalling factors, which has been generally observed for MAP3K interactions with MAP2Ks(39).

Validation of T_M shift hits with a kinase activity assay

The optimal ZAK peptide substrate (ZAKtide, ANHWHTVHLRA) was synthesised (Michael Berne, Tufts Medical School) and served as the substrate in a kinase activity assay. ZAKtide was incubated with ZAK and ATP/Mg²⁺ at 37°C, and the relative levels of unphosphorylated and phosphorylated ZAKtide in the reaction mixture were quantified using mass spectrometry. Under the chosen conditions the apparent K_M value for ATP was determined to be 140 µM which is in the range of K_M values for previously investigated protein kinases(40). We measured IC50 values for compounds that stabilised ZAK by more than 5°C in T_M shift assay (Figures 1D and S1). All of the tested compounds inhibited ZAK with IC50s ranging from 12 nM for rebastinib to 210 nM for apatinib confirming the T_M shift data (Figure 1E). However, probably due to differences in binding modes T_M shift values did not always correlate well with experimental IC50 values.

ZAK crystallisation

Our attempts to co-crystallise ZAK with ATP-γ-S and a subset of the T_M shift hits (ponatinib, motesanib and vemurafenib) were not successful (summarised in Figure 2B). Though ZAK crystallised in complex with ATP-γ-S the diffraction was poor. Therefore, ZAK was subjected to autophoshorylation resulting in ZAK that was monophosphorylated mainly at the activation loop residue T161 (Figures 2C,D and S2) and ZAK diphosphorylated at T161 and S165. These differentially phosphorylated ZAK variants were co-crystallised with the inhibitors mentioned above. Only the combination of monophosphorylated ZAK and vemurafenib resulted in crystals with suitable diffraction properties. The structure was solved by molecular replacement using a MLK1 model (PDB ID 3DTC) as a template. A summary of data collection and refinement statistics is given in Table S3.

Open in a separate window

Figure 2

(A) Crystal structure of ZAK_5–309 in complex with vemurafenib. Kinase structural elements are highlighted in green. Phosphorylated T161 is shown in red. (B) Attempts crystallizing differentially phosphorylated ZAK variants with diverse inhibitors. Listed are the space groups and the diffractions of the obtained crystals. Only the combination of ZAK phosphorylated in T161 and vemurafenib resulted in co-crystals with suitable diffraction properties. (C) Intact protein MS spectrum of the monophosphorylated ZAK_5–309 used for crystallisation. (D) Phosphomapping of monophosphorylated ZAK_5–309. MS/MS spectra have been analysed with PEAKS. The predominant phosphorylation site is T161 within the ZAK activation loop. (E) Vemurafenib binds to ZAK in the DFG-in conformation. D151 and F152 are shown in red. (F) The N terminal part of the ZAK leucine zipper is tightly attached to the αG helix. Leucine residues forming a hydrophobic surface patch, F267 and W283 are shown in red.

Structure of the ZAK-vemurafenib complex

The ZAK kinase domain comprised the canonical kinase catalytic domain structure with an N lobe containing the β1-β5 sheets and the αC helix and the mainly helical C lobe. As expected, the inhibitor vemurafenib interacted with the ATP binding site (Figure 2A). Besides this conserved architecture common to all protein kinases the ZAK structure revealed some unique features.

The αC helix is predicted by PSIPRED to comprise residues 48–59. However, in the crystal structure it was moved away from the ATP binding site and comprised only residues 53–59. This SRC/CDK-like inactive conformation was probably due to the interaction with vemurafenib which pushed the αC helix outward. As a result the conserved salt bridge between K45 and E53 which is regarded a hallmark for the active kinase conformation was not formed (Figure S3).

The ZAK kinase and SAM domains are linked by 60 residues that are predicted to comprise a leucine zipper. The helix in the N terminus of the linker was attached to the kinase C lobe. W283 formed a π stacking interaction with F276, while L291, L294 and L297 formed a hydrophobic surface patch interacting with residues located in the αG helix (Figure 2F). The C terminus of the linker was not resolved in the structure with the electron density dropping sharply after residue 299. Based on the prediction of helical secondary structure for residues 301–322 we assume that the linker is composed of two helices interconnected by a flexible hinge. For Nek2, another protein kinase with a leucine zipper succeeding the kinase domain, the leucine zipper is required for homodimerisation and activation.(41) It remains to be investigated if a similar mechanism is also valid for ZAK. However, our construct ZAK_5–309 did neither dimerise in size exclusion chromatography nor did the leucine zippers interact in the crystal.

The conserved P loop typically constitutes the most flexible part of the kinase N lobe. It bridges the β1 and β2 sheets and contributes to position the ATP β- and γ-phosphates for catalysis. In kinases bound to ATP and in 98% of the publicly available structures of kinase-inhibitor complexes, the P loop adopts an extended conformation.(42) However, in the ZAK-vemurafenib complex, the P loop was folded over the inhibitor thus forming a hydrophobic cage. This kinked conformation was mainly driven by the G23 backbone amide having polar contacts with the vemurafenib chlorine atom and by F27 forming a π stacking interaction with the chlorophenyl moiety.

As mentioned above ZAK tightly bound to a range of type II kinase inhibitors suggesting that the ZAK DFG motif at the base of the activation loop is capable of adopting the DFG-out conformation. However, the bound vemurafenib stabilised the DFG-in conformation with D151 pointing towards the ATP binding pocket and F152 pointing towards the αC β4 loop that anchors the C terminus of the αC helix to the top of the C lobe (Figure 2E). The protein used for crystallisation was phosphorylated in the activation loop residue T161 (Figures 2C,D and S2). In the crystal structure, the phosphate moiety of pT161 was in proximity of E51 N terminal of the αC helix, H158 in the activation loop and C285 of a neighbouring ZAK molecule interacting through crystal contacts. This interaction seems to be indispensable for crystallisation in space group P 4₁ since we did not obtain any such crystals of ZAK in a different phosphorylation state. The location of pT161 in the crystal structure is surprising as this phosphorylation site is located at the expected position in the sequence for activating phosphorylation sites (approximately 11 residues N terminal of the APE helix) and is expected to anchor the activation segment to the HRD arginine. It remains therefore to be investigated if the observed conformation is a consequence of inhibitor binding or may also be observed in the apoprotein. The region C terminal of pT161 was not resolved in the experimental electron density creating a 6 residue gap and indicating a disordered activation loop in solution despite its phosphorylated state. We speculate that T161 phosphorylation alone may not be sufficient to induce the attachment of the activation loop to the C lobe indicating additional S165 phosphorylation to be required to achieve full ZAK activity.

Thermodynamics of vemurafenib binding to ZAK

In order to determine the thermodynamic parameters for the binding of vemurafenib to ZAK in solution we performed isothermal titration calorimetry (ITC). Due to the poor vemurafenib solubility the experimental setup had to be optimised. Finally we carried out the binding experiment as a reverse titration in which a solution of 120 µM unphosphorylated ZAK was titrated into 15 µM vemurafenib at 37°C (Figure 3A). The binding was enthalpically driven (ΔH= −9.6 kcal·mol⁻¹) with a minor entropic contribution to binding (−TΔS= −1.1 kcal·mol⁻¹) resulting in a dissociation constant (K_D) of 29 ± 4 nM. The slow equilibration following the injections likely reflected structural rearrangements within ZAK. Takeninto account the tight binding of type II inhibitors we suggest that in solution both the ZAK DFG-in and DFG-out conformations prevail in equilibrium, and that the interconversion rates are low. The measured K_D was consistent with the measured IC50 value of 23 ± 4 nM (Figure 1D) and with KINOMEscan data of the close vemurafenib analogue PLX4720 to ZAK with a K_D of 41 nM.(38) The insertion of the phenyl ring into PLX4720 was reported to increase the inhibitor’s affinity for ZAK by factor 2.(21)

Open in a separate window

Figure 3

(A) Thermodynamic properties of vemurafenib binding to ZAK assessed by ITC. (B) Vemurafenib bound to ZAK interacts with a range of kinase structural motifs. The 2mFo-DFc electron density map is contoured at 2.5σ. (C) IC50s of PLX4720 and vemurafenib for ZAK and B-RAF as determined by Vin et al.²¹ (D) Vemurafenib binding to ZAK induced a kinked P loop conformation. F27 formed a π stacking interaction with the vemurafenib chlorophenyl ring. (E) Vemurafenib binding to B-RAF stabilized an extended P loop conformation. F468 did not interact with vemurafenib (PDB ID 3OG7).

Vemurafenib binding mode

Our structural model suggests that vemurafenib does not inhibit ZAK by simply blocking the ATP binding site but by significantly distorting the kinase fold. Vemurafenib tightly interacted with five ZAK structural motifs (Figure 3B). The central 7-azaindole firmly anchored vemurafenib in the position generally occupied by the ATP purine forming two ATP mimetic hydrogen bonds with the backbone of E83 and A85 in the ZAK hinge region. The interaction was further strengthened by a π stacking interaction of Y84 with the condensed ring system.

The bifluorinated phenyl ring substituted with a propylsulfonamide group interacted with the DFG motif and penetrated with the propyl moiety into a pocket formed by the displaced αC helix and the DFG motif. Three hydrogen bonds were formed with all backbone nitrogens of the DFG motif thus trapping it in the DFG-in conformation. A water bridge interconnected the vemurafenib ketone linker and the D151 side chain contributing additional interactions with ZAK.

The second substituent of the 7-azaindole in vemurafenib is a chlorophenyl ring in 5 position. Even though the ring protruded from the ATP binding site it was entwined by the displaced ZAK P loop thus shielded from the surrounding solvent. The main driving force of the kinked P loop conformation seemed to be a π stacking interaction of F27 with the chlorophenyl ring.

Comparison of B-RAF and ZAK bound to vemurafenib

The vemurafenib binding modes in B-RAF and ZAK share the same hinge interaction, but they differ considerably in the P loop conformation. While in B-RAF the P loop adopts an extended conformation not making contacts to vemurafenib(43), the ZAK P loop extensively interacts with vemurafenib (Figure 3D,E). The inhibitor PLX4720 differs from vemurafenib by the substitution of the chlorophenyl ring for a chlorine atom (Figure 3C).(43) The affinities of PLX4720 and vemurafenib for ZAK and B-RAF have been determined by a KINOMEscan competitive binding assay.(21) PLX4720 is a more potent inhibitor for B-RAF when compared with vemurafenib (IC50s 32 nM vs 65 nM), for ZAK this property is reversed (IC50s 9.5 nM vs 4.0 nM).(21) Our structural data suggest that this is due to the P loop interacting with the chlorophenyl ring. Folded and distorted P loop conformations have been associated with favourable inhibitor selectivity.(42) A structural consequence of a folded P loop, in particular when it is associated with an αC out movement, is the generation of a cavity between the P loop and αC. Such a cavity has for instance successfully targeted by the ERK inhibitor SCH772984.(44) The structure of the vemurafenib ZAK complex revealed a similar binding pocket extending from the carbonyl moiety of the inhibitor suggesting that this structural feature could be explored for the generation of selective ZAK inhibitors. Distortion of the P loop has also been linked to the specificity of imatinib for the tyrosine kinase ABL over c-SRC. The K_D for imatinib binding ABL is by factor 2300 lower than for c-SRC(45) despite the high conservation of both binding sites. Besides the more favoured DFG-out conformation in ABL it is the P loop conformation (kinked in ABL(46) and extended in c-SRC(45)) that accounts for the difference.(47),(48),(49)

SM is grateful for support by Boehringer Ingelheim. The SGC is a registered charity (number 1097737) that receives funds from AbbVie, Bayer, Boehringer Ingelheim, the Canada Foundation for Innovation, the Canadian Institutes for Health Research, Genome Canada, GlaxoSmithKline, Janssen, Lilly Canada, the Novartis Research Foundation, the Ontario Ministry of Economic Development and Innovation, Pfizer, Takeda, and the Wellcome Trust [092809/Z/10/Z]. BMK is supported by the John Fell Fund 133/075 and the Wellcome Trust [097813/Z/11/Z].