The ASK1 catalytic domain structure displays a typical protein kinase fold comprising the five β sheets and helix αC constituting the small lobe (residues 670–757) and a larger mainly alpha helical C-terminal lobe (residues 761–940) (Figure 1). The hinge region connecting the two domains lines the catalytic ATP binding site, which is occupied by the ATP competitive inhibitor staurosporine. The asymmetric unit harbors two ASK1 molecules, which can be superimposed with an rmsd of 0.6 Å and can therefore be considered structurally identical. Residues located in the loop regions at the beginning of αC (Asp715-Ser719) and the activation segment residues Gly831-Glu837 were not visible in the electron density and were assumed to be disordered. The N-terminus forms a short β sheet (β₀), which is connected to the sheet β₁ by an unusually large loop comprising residues 675–684. This large loop insertion may play a role in regulation of ATP binding since it determines the position of the phosphate binding loop, whereas the N-terminal sheet β₀ extends the antiparallel sheet network in the smaller lobe, possibly adding stability to this domain.

ASK1 is only distantly related in sequence to kinases of known structure. Superimposition with the catalytic domain of its next structural neighbor PAK1 (PDB code: 1YHW; 29% sequence identity) revealed that the two kinases superimpose with an rmsd of about 2 Å (Figure 1C). However, structure determination by molecular replacement was challenging and required an improved search model built on the basis of SAD data collected from another crystal form (Table 1). The most striking difference between these two structures is that helix αC in ASK1 is moved further away from the active site. This has also been described for the MAP3K family member and the closest relative on the phylogenetic tree, TAK1 (PDB code: 2EVA). As a consequence of the outward shifted αC helix, the hydrogen bond between the conserved active site lysine (Lys709) and αC glutamate (Glu725) typically found in active kinases is not present in the ASK1 structure described here, and the domain therefore assumes a catalytically nonproductive conformation.

Staurosporine binds to the ATP binding site of ASK1 in a mode that has been described for a number of kinase-staurosporine complexes (Underwood et al., 2003; Bertrand et al., 2003) (Figure 2). Two hydrogen bonds anchor the lactam moiety to the hinge region and mimic the hydrogen-bonding pattern of the adenine base. The stauroporine ligand forms a third hydrogen bond with the backbone oxygen of the catalytic loop residue Asn808, an interaction that has been described previously (Lamers et al., 1999; Lawrie et al., 1997; Prade et al., 1997; Zhao et al., 2002; Zhu et al., 1999). The heterocyclic condensed five-ring system packs against the hydrophobic side chains of the ASK1 residues Leu686, Val694, and Ala707 on the N-terminal lobe side as well as Val738 and Leu810 on the C-terminal lobe side of the binding pocket. In addition to staurosporine, we identified a number of diverse inhibitors in the melting temperature screen performed on a library of 158 known kinase inhibitors (Table S1 and Figure S1 in the Supplemental Data available with this article online). The inhibitors have been used in cocrystallization experiments, but well-diffracting crystals were only obtained in the presence of staurosporine.

Although not phosphorylated, the activation segment was well defined by electron density except for the unstructured tip of the activation loop (residues 832–836). Interestingly, the conserved αC glutamate (Glu725) forms a hydrogen bond with the main-chain nitrogen of the DFG motif residue Phe823, whereas the conserved active site lysine (Lys709), which usually forms a salt bridge with the αC glutamate in active kinases, also hydrogen bonds to the main chain of Phe823 (Figure 3A). The DFG aspartate (Asp822) forms a hydrogen bond with the activation segment residue Ser826. The observed noncatalytic conformation might be influenced by the presence of the bulky ATP competitive ligand staurosporine, which forms hydrophobic contacts with the DFG Phe823. In contrast, C-terminal to the DFG motif the activation segment is stabilized by interactions typically found in activated kinases. Comparison with the structure of phosphorylated, active PAK4 (Eswaran et al., 2007) revealed that the ASK1 activation segment residue Glu837 mimics the hydrogen-bond pattern formed by an activating phosphate moiety. In active kinases, phosphorylation at this position typically results in formation of a hydrogen bond network linking the catalytic loop arginine of the HRD motif (Arg802 in ASK1) with both termini of the activation segment (Lawrie et al., 1997). These interactions stabilize the activation segment in a conformation competent for substrate binding (Figures 3B and 3C). Activation segment mutants that lead to activation of kinases such as the S474E mutant in PAK1 (Lei et al., 2005), or activation segments of constitutively active wild-type kinases, mimic the hydrogen bond patterns formed by the activating phosphate moiety. Thus, the position and interactions formed by Glu837 in ASK1 suggest an activating and stabilizing function of this residue in unphosphorylated ASK1. Indeed, we observed that unphosphorylated ASK1 is active and autophosphorylates rapidly at three sites in vitro.

Full-length ASK1 has been described as a dimeric molecule stabilized in vivo by coiled-coil domain located C-terminal to the catalytic domain. Consequently, dimerization of the kinase domain was not an anticipated finding. Analysis of crystal packing revealed that the two ASK1 molecules in the asymmetric unit dimerize with their own symmetry equivalent. Both molecules bury a total surface area of about 1,997 Ų consisting of 473 Ų of hydrophobic and 1524 Ų of hydrophilic surface. In order to verify the dimeric state of the ASK1 kinase domain in solution, we carried out analytical ultracentrifugation experiments (Figure 4). Sedimentation velocity measurements led to the determination of an apparent sedimentation coefficient corrected for water at 20°C, s20,w0, of 4.518 S, as well as the determination of a protein molecular weight of 66 kDa, which is in excellent agreement with the expected mass of an ASK1 catalytic domain dimer. Equilibrium experiments performed at three concentrations (c0, c0/2, and c0/4, where c0 is the concentration used in the velocity experiment) confirmed a stable dimeric state of the kinase domain in solution (Figure 4B). Nonlinear least square fit to a self association fitting model resulted in determination of a dissociation constant (K_D) of 0.22 ± 0.2 μM.

The interaction between ASK1 monomers is based primarily on shape complementarity over a large surface area spanning almost the entire length of the protein. To achieve such a tight interaction, the molecules associate in a head-to-tail fashion with the N-terminal domain of one molecule interacting with the C-terminal domain of the other molecule and vice versa. The two catalytic sites are on the same side of the dimer near the equatorial plane and around 20 Å apart. Interaction between the molecules involves a number of direct hydrogen bonds (Leu700-Asn776', Asn702-Tyr783', Gln703-Thr779', and Arg705-Thr813' [and vice versa]), as well as water-mediated hydrogen bonds (Gln756-HOH-Tyr814') (the apostrophe denotes the symmetry equivalent molecule). In addition, multilayer π-stacking and hydrophobic interactions Arg705-Tyr814'-Pro758/Pro758'-Tyr814-Arg705' may also contribute to dimerization (Figures 4C and 4D and Table S3).

One of the identified phosphorylation sites (Thr813) is located in the dimer interface, and phosphorylation at this residue is likely to result in formation of hydrogen bonds to Arg705 located in the interacting protomer. We were therefore interested to explore if phosphorylation at Thr813 effects dimerization of ASK1. However, AUC data indicated that dephosphorylated ASK1 (mixture between de- and monophosphorylated ASK1) homogeneously triphosphorylated ASK1 as well as a T813A mutant formed all stable dimers in solution, and sedimentation equilibrium experiments revealed similar association behavior (Figure 4).

In order to determine the sequence specificity of active site-mediated phosphorylation, we screened a peptide library with recombinant ASK1 (Figure 5A). The library was constructed to evaluate the contribution of all amino acid residues at each of nine positions (−5 to +4) surrounding a fixed central phosphoacceptor site (Hutti et al., 2004). The phosphorylation data revealed that ASK1 has a substantial preference for threonine over serine as a phospho-acceptor. ASK1 appears to be most selective at position +1 relative to the phosphorylation site, where it strongly prefers both aromatic and aliphatic hydrophobic residues in peptide substrates. We also observed strong phosphorylation of peptides bearing threonine residues in either the −2 or the +2 position. Because peptides in the library with fixed threonine residues can theoretically be phosphorylated at more than one site, these signals could reflect authentic selection for threonine at only one of the two positions. ASK1 also has secondary preferences for glutamine at the −2 position and for serine, arginine, and tyrosine at the +2 position. The kinase is not strongly selective at any of the other positions represented in the peptide library. The determined consensus motif is shown in Figure 5B.

The mass spectrum of the intact recombinant protein treated with ATP/Mg²⁺ clearly indicated the presence of at least three phosphorylation sites (Figure S2). The protein used for crystallization also contained small amounts of the monophosphorylated species arising during expression of the protein. The electron density map showed that Thr838 is partially phosphorylated, but due to the low occupancy of the phosphate moiety at this site, the phosphothreonine could not be convincingly modeled. An autophosphorylation site was identified on Thr813, which is located in the loop region connecting the sheets β8 and β9 pointing toward the C terminus of helix αD. The site is located at the protein surface very distant from the cofactor or substrate binding sites, but phosphorylation at Thr813 could impact lobe dynamics and thereby indirectly influence catalysis. Thr813 is also located in the dimer interface in close proximity to Arg705 of the interacting catalytic domain located in the loop between β₂ and β₃, but significant effects on dimerization were ruled out by AUC experiments carried out on an alanine mutant. In dimeric ASK1, Thr813 is not accessible, and the dissociation constant determined for the dimer association suggests that phosphorylation at this site is a slow process. However, structural changes in the ASK1 signalosome, which are likely to occur upon activation, might render this site more accessible.

A second site was identified at Thr838, which is located in the activation segment. In our structure, the side chain of Thr838 is oriented toward the solvent, and phosphorylation at this site is not expected to stabilize the activation segment in the absence of structural rearrangements. Furthermore, as described above, the neighboring residue Glu837 forms interactions typically formed by a phosphate moiety located at the corresponding position within the activation segment. However, Thr838 has been shown to play an essential role in activation of ASK1 by oxidative stress (Tobiume et al., 2002). A third phosphorylation site was identified at Thr842 located in close proximity (2.6 Å) to the catalytic aspertate (Asp803), and phosphorylation at Thr842 may lead to formation of hydrogen bonds with the catalytic loop lysine Lys805. All three sites, in particular Thr838, are in good accordance with our determined consensus peptide phosphorylation motif for ASK1 (Figure 5B) in contrast to known phosphorylation sites of downstream ASK1 substrates (Figure 5C). The location of the three autophosphorylation sites in our structural model is shown in Figure 5D.

To study the role of each of the identified autophosphorylation sites, we cloned site-directed alanine mutants into an expression vector and studied the effects of the mutations in transient transfection assays, monitoring JNK/p38 reporter gene activity. As expected from earlier studies, mutation of Thr838 drastically reduced reporter gene activity when compared to unstimulated control levels. Interestingly, mutation of the other two sites also provided a significant reduction in ASK1 function (Figure 6A), suggesting that autophosphorylation at the residues Thr842 and Thr813 regulates ASK1 signaling.

Furthermore, we studied the effect of the site-directed alanine mutants on kinase catalytic activity in vitro. Recombinant ASK1 mutants autophosphorylated rapidly and had specific activities that were comparable to wild-type ASK1 (data not shown). To test if the generated ASK1 mutants are also active on specific ASK1 substrates, we reconstituted a minimal ASK1 signaling cascade (ASK1-MKK6-p38) in vitro with recombinant MKK6 and His₆-p38α by detecting p38 phosphorylation with an antibody specific for phosphorylated p38. Surprisingly, all three mutants showed catalytic activity in this in vitro phosphorylation experiment comparable to wild-type ASK1, suggesting that phosphorylation at the three catalytic domain autophosphorylation sites may be required for the recruitment of signaling partners and ASK1 substrates to this large signalosome rather than for regulation of catalytic activity.

Human ASK1 (residues 659–951; gi|5174547) was subcloned into the T7 expression vector SGC-pLIC. The protein was expressed as an N-terminally His6-tagged fusion protein with a tobacco etch virus (TEV) protease cleavage site by using Escherichia coli BL21(DE3).

A 10 ml overnight culture was used to inoculate 1 l of Terrific Broth media containing 100 μg/ml kanamycin. Cultures were grown at 37°C until the OD₆₀₀ reached ∼2.0. After that, the temperature was adjusted to 25°C, and expression was induced for 18 hr with 1 mM IPTG. Cells were centrifuged and pellets were resuspended in binding buffer (50 mM HEPES [pH 7.5], 300 mM NaCl, 20 mM imidazole including a protease inhibitor cocktail [Complete, Roche]) and lysed with a high-pressure homogenizer. The lysate was cleared by centrifugation, and the protein was purified by Ni-affinity chromatography with a wash buffer (50 mM HEPES [pH 7.5], 1 M NaCl, 20 mM Imidazole) and an elution buffer of the same composition but containing 250 mM imidazole. The eluted protein was treated with lambda and alkaline phosphatase together with TEV protease for 12 hr at 4°C to remove phosphorylation and the His6-tag, respectively. The eluted fraction was further purified by gel filtration chromatography with Superdex S75 (60 × 1 cm) column equilibrated in 10 mM HEPES (pH 7.5), 500 mM NaCl, 5% (w/v) glycerol. DTT was added to the protein sample to a final concentration of 10 mM, and the protein was concentrated to 13 mg/ml. Purity and integrity of ASK1 were confirmed by SDS/PAGE and electrospray ionization time-of-flight mass spectrometry (Aglient LC-ESI TOF).

Sedimentation velocity experiments were carried out on a Beckman XL-I Analytical Ultracentrifuge equipped with a Ti-50 rotor and cells with double-sector centerpieces. Protein samples were studied at a concentration of 13 μM in 10 mM HEPES (pH 7.4) (at 25°C), 120 mM NaCl, 0.25% glycerol at 4°C, employing a rotor speed of 40,000 rpm. Radial absorbance scans were collected at a wavelength of 280 nm in 1 min intervals. Data were analyzed with SEDFIT (Schuck, 2000) calculating c(s) distributions. The software package SEDNTERP was used in order to normalize the obtained sedimentation coefficient values to the corresponding values in water at 20°C, s20,w0, taking into account the solvent density (ρ_o = 1.00503 g/ml), viscosity (η_o = 1.5862 × 10⁻²), and the protein partial specific volume (0.7309 ml/g). Sedimentation equilibrium experiments were performed with a six-sector centerpiece and protein concentrations of 6.5, 11.9, and 25.2 μM. Equilibrium was obtained after 24 hr at each of the two centrifugation speeds (10,000 rpm and 15,000 rpm). Baseline offsets were determined by using a meniscus depletion run. Sedimentation equilibrium data were evaluated by using Ultraspin.

Phosphorylation motifs for ASK1 kinases were determined by using a positional scanning peptide library approach essentially as described before (Hutti et al., 2004). Reactions were carried out in multiwell plates in 50 mM HEPES (pH 7.4), 10 mM MgCl₂, 1 mM DTT, 0.1% Tween 20, 100 μM ATP (including 0.3 μCi/μl γ-[³³P]-ATP), 50 μM peptide substrate, and 50 μg/ml unphosphorylated ASK1 for 2 hr at 30°C. Peptide substrates had the general sequence YAXXXXX-S/T-XXXXAGKK(biotin), where S/T represents an even mixture of serine and threonine, K(biotin) is ɛ-(biotinamidocaproyl)lysine, and X is a roughly equimolar mixture of the 17 amino acids excluding cysteine, serine, and threonine. Each well contained a distinct peptide in which one of the X positions was replaced with 1 of 22 residues (one of the unmodified proteogenic amino acids exlcuding pSer and pThr). In addition, three additional wells were included that contained either no peptide, the peptide YAXXXXX-S-XXXXAGKK(biotin), or the peptide YAXXXXX-T-XXXXAGKK(biotin), to test phosphoacceptor residue preference. At the end of the incubation time, aliquots of each reaction were spotted onto streptavidin membrane, which was processed as described by Hutti and coworkers (Hutti et al., 2004).

The protein was denatured by boiling, cysteine residues were reductively alkylated, and then a tryptic digest was performed overnight (100:1 protein/enzyme ratio). The peptides were separated on a Dionex 3000 nano-LC system with a C18 Pepmap column by using a water/acetonitrile gradient with 0.1% formic acid and analyzed with a Bruker HCT Ultra ion trap in MS/MS mode. Alternating fragmentation cycles were performed in data-dependent MS/MS by using collision-induced dissociation (CID) and electron transfer dissociation (ETD), the data were submitted to Mascot searches (http://www.matrixscience.com/) separately, and the information from the two alternative approaches was combined.

ASK1-WT and ASK1-T838 (human; T845 for murine) were described previously (Zhang et al., 2005). ASK1 mutants (T813A and T842A) were constructed by site-directed mutagenesis with a Quickchange site-directed mutagenesis kit (Stratagene) according to the protocol of the manufacturer. Transfection of 293T cells were performed by Lipofectamine (Invitrogen). Luciferase activity followed by renilla activity was measured twice in duplicate with a Berthold luminometer. All data were normalized as relative luciferase light units/renilla unit.

Cell lysates were subjected to SDS-PAGE followed by immunoblot (IB) with a specific antibody (e.g., anti-HA) followed by HRP-conjugated anti-mouse secondary antibody. Protein was detected by chemiluminescence with an ECL kit according to the instructions of the manufacturer (Amersham Life Science).

Initial screens were set up with ASK1 kinase domain and several high-affinity small-molecular inhibitors. Crystals appeared in several conditions when 1 mM staurosporine was added to the protein. Screen conditions that yielded crystals all contained 20%–25% PEG 4000 or PEG 6000, 0.2 M salt, and had a pH around 6.0 (buffered or unbuffered). After optimisation, diffraction quality needle-like crystals were obtained from 25% PEG3350, 0.17 M (NH₄)₂SO₄, and 15% glycerol with average dimensions 0.2 × 0.05 × 0.05 mm³ and were mounted with a loop and flash frozen by plunging into liquid nitrogen. Datasets were collected at the PXII beamline at the Swiss Light Source with a MAR225 detector at 0.972 Å.

Selenomethionine-labeled protein did not yield crystals using the same crystallization condition, and initial screening was repeated. Eventually, needle-like crystals were obtained from 18% PEG6000 and 0.1 M H₃Cit/Na₃Cit (pH 4.9). A SAD dataset was collected at the PXII beamline at the SLS by using the selenium peak wavelength (0.9789 Å, determined from a fluoresence scan). Diffraction images were processed with XDS (Kabsch, 2001) different scans were scaled together with XSCALE (Kabsch, 2001). Data collection statistics are shown in Table 1.

Although molecular replacement solutions could be obtained by employing either an ensemble of most similar structures (29%–31% sequence identity, including only the C-terminal domain) or a homology model as search model, refinement was unstable and diverged. Unfortunately, the anomalous signal in the selenomethione-labeled protein was too weak to result in structure solution on its own. Although selenium positions could be located from the anomalous data with SHELXD (Sheldrick et al., 2001), no interpretable map was obtained.

Initially, the homology model that was used in molecular replacement of the unlabeled protein was used in Phaser (McCoy et al., 2005), and a molecular replacement solution for the selenomethionine crystal was obtained. This model was refined in refmac5 (Murshudov et al., 1997), and phase restraints (phase probability distribution was obtained by SHARP [La Fortelle and Bricogne, 1997] from Se positions determined by SHELXD and expressed as Hendrickson-Lattman coefficients) were vital for keeping refinement stable. After several rounds of manual rebuilding, the model improved significantly and could be used for molecular replacement of the native dataset. Phaser located two molecules in the asymmetric unit, revealing that solutions obtained previously were incomplete and missing the second molecule. Multicrystal averaging, enabled by the fact that crystals from the unlabelled and selenomethionine-labeled protein were nonisomorphous and had a different cell and space group, was performed and resulted in a clear electron density map that was manually traced. Refinement was completed in refmac5 and converged to an R/R_free of 0.198/0.251. The resulting model was deposited with the Protein Data Bank under accession ID 2CLQ.