Structural Insight into the 14-3-3 Protein-dependent Inhibition of Protein Kinase ASK1 (Apoptosis Signal-regulating kinase 1)*

Apoptosis signal-regulating kinase 1 (ASK1, also known as MAP3K5), a member of the mitogen-activated protein kinase kinase kinase (MAP3K) family, regulates diverse physiological processes. The activity of ASK1 is triggered by various stress stimuli and is involved in the pathogenesis of cancer, neurodegeneration, inflammation, and diabetes. ASK1 forms a high molecular mass complex whose activity is, under non-stress conditions, suppressed through interaction with thioredoxin and the scaffolding protein 14-3-3. The 14-3-3 protein binds to the phosphorylated Ser-966 motif downstream of the ASK1 kinase domain. The role of 14-3-3 in the inhibition of ASK1 has yet to be elucidated. In this study we performed structural analysis of the complex between the ASK1 kinase domain phosphorylated at Ser-966 (pASK1-CD) and the 14-3-3ζ protein. Small angle x-ray scattering (SAXS) measurements and chemical cross-linking revealed that the pASK1-CD·14-3-3ζ complex is dynamic and conformationally heterogeneous. In addition, structural analysis coupled with the results of phosphorus NMR and time-resolved tryptophan fluorescence measurements suggest that 14-3-3ζ interacts with the kinase domain of ASK1 in close proximity to its active site, thus indicating this interaction might block its accessibility and/or affect its conformation.

processes, such as apoptosis, cytokine secretion, or cell differentiation by activating the c-Jun N-terminal kinase and p38 signaling pathways (1). The activity of ASK1, which is triggered by various stress stimuli, is involved in the pathogenesis of cancer, neurodegeneration, inflammation, and diabetes (2). The human ASK1 consists of 1374 amino acid residues with the serine/threonine kinase domain being centrally located and flanked by two coiled-coil (CC) domains. ASK1 forms a high molecular mass complex called ASK1 signalosome under nonstress conditions in which ASK1 homo-oligomerizes through the C-terminal CC domain and interacts with several other proteins including ASK2, thioredoxin, and the 14-3-3 protein (Fig.  1A) (3,4). Thioredoxin presumably prevents homo-oligomerization of ASK1, whereas 14-3-3 binds to the motif located at the C terminus of the kinase domain and suppresses the catalytic activity by unknown mechanism (4 -7). In response to oxidative stress, both thioredoxin and 14-3-3 dissociate from ASK1 followed by the homo-oligomerization of ASK1 and the recruitment of tumor necrosis factor receptor-associated factors 2 and 6. This, in turn, accelerates the autophosphorylation of Thr-838 within the activation segment resulting in ASK1 activation (5,8).
14-3-3 proteins, a family of dimeric proteins ubiquitously expressed in all eukaryotic cells, bind to other proteins by recognizing motifs containing either a phosphorylated serine (Ser(P)) or a phosphorylated threonine (Thr(P)) residue (9). Through these binding interactions, 14-3-3 proteins regulate many biological processes such as cell cycle progression, initiation of apoptosis, control of gene transcription, and neuroendocrine transduction (10 -12). 14-3-3 proteins often function as molecular scaffolds that modulate the conformation of their binding partners; if the binding partner is an enzyme, this can affect its catalytic activity. For example, the 14-3-3-dependent activation of serotonin N-acetyltransferase (AANAT) and yeast neutral trehalase (Nth1) has been shown to be based on a direct structural change of the catalytic site (13,14). On the other hand, tyrosine hydroxylase, Raf-1 kinase, and plant plasma membrane H ϩ -ATPase are regulated through the 14-3-3-dependent modulation of intradomain interactions and/or the quaternary arrangement (15)(16)(17).
A functional role of 14-3-3 in the regulation of ASK1 was first suggested by the observation that under non-stress conditions 14-3-3 recognizes a motif containing phosphoserine Ser(P)-966 located at the C terminus of ASK1 kinase domain (Fig. 1A) and that the replacement of this residue to alanine dramatically accelerates ASK1-induced cell death (6). It was subsequently shown that the oxidative stress-induced dephosphorylation of Ser(P)-966 is accompanied by an increased autokinase activity of ASK1 and that the 14-3-3 binding-defective mutant ASK1 S966A shows increased both autokinase and transkinase activities (7). However, the molecular mechanism of the 14-3-3-dependent regulation of ASK1 remains elusive.
In this work, we performed structural analysis of the complex between the kinase domain of ASK1 (sequence 659 -973) phosphorylated at Ser-966 and the 14-3-3 protein using small angle x-ray scattering (SAXS) and chemical cross-linking. In addition, we also investigated whether the 14-3-3 binding affects conformation of the active site of ASK1.

Results
Biophysical Characterization of the pASK1-CD⅐14-  Complex-The interaction between pASK1-CD and 14-3-3 was first studied using sedimentation velocity (SV) analytical ultracentrifugation (AUC) by analyzing their mixtures at various molar ratios (Fig. 1D). The direct modeling of obtained SV AUC data using the Lamm equation revealed the best fit apparent equilibrium dissociation constant (K D ) of 36 Ϯ 7 M using a Langmuir binding model with one dimer of pASK1-CD interacting with one dimer of 14-3-3WT. To increase the stability of the complex, we prepared 14-3-3 deleted of its C-terminal 15 residues (14-3-3⌬C), which binds ligands with significantly higher affinity (17,18). Indeed, 14-3-3⌬C showed significantly increased binding affinity for pASK1-CD with the best-fit K D value of 4 Ϯ 2 M (Fig. 1E). The observed value of s w(20,w) of 6.2S for the complex (based on c(s) distribution of mixture with highest molar ratio of 14-3-3⌬C to pASK1-CD shown in purple in Fig. 1E) corresponds to a M r of ϳ121, thus suggesting the 2:2 molar stoichiometry of the complex (a dimer of pASK1-CD bound to a dimer of 14-3-3⌬C, theoretical M r 128.9). Considering the significantly improved binding affinity of pASK1-CD for 14-3-3⌬C compared with WT, we decided to use 14-3-3⌬C rather than 14-3-3WT throughout this work.
14-3-3 Binding Suppresses the Kinase Activity of pASK1-CD-The pASK1-CD used in this study is enzymatically active, allowing us to perform activity measurements to compare the properties of complexed and free pASK1-CD. Human mitogenactivated protein kinase kinase 4 (MKK4), a physiological target of ASK1, was used as a specific substrate. As noticed, the addition of 14-3-3⌬C suppressed the activity of pASK1-CD relative to the uncomplexed protein by ϳ40%, whereas no significant decrease in kinase activity was observed for the 14-3-3 binding-defective mutant pASK1-CD S966A (Fig. 1F, the specific activities are listed in supplemental Table S1). The 14-3-3-dependent inhibition of pASK1-CD enzyme activity is consistent with data published by Fu and co-workers (6, 7) who showed that ASK1 binding to 14-3-3 caused a significant suppression of ASK1-induced apoptosis through inhibition of its kinase activity.
Characterization of the pASK1-CD⅐14-3-3 Complex by SAXS-The scattering data for 14-3-3⌬C alone and ASK1-CD alone as well as the pASK1-CD⅐14-3-3⌬C complex (mixed with 2:2 molar stoichiometry) were collected at various concentrations ( Table 1). The complex was prepared at concentrations of ϳ60, 95, and 137 M, thus sufficiently above the K D (ϳ4 M). The Guinier plots and the concentration dependence of the forward scattering intensity I(0), the Porod volume V p , and the radius of gyration R g were used to assess the data quality. Biophysical characterization of interaction between pASK1-CD and 14-3-3. A, schematic representation of the domain structure of ASK1 with bound thioredoxin (TRX) and 14-3-3. NCC, N-terminal CC; CCC, C-terminal CC; CD, catalytic domain; pS, phosphoserine. The 14-3-3 binding motif is shown in green. B, the normalized continuous sedimentation coefficient distributions, c(s), for 14-3-3WT and 14-3-3⌬C. C, the normalized c(s) distributions for notphosphorylated ASK1-CD and phosphorylated pASK1-CD. D, series of area normalized c(s) distributions for mixtures of pASK1-CD and for 14-3-3WT at various molar ratios. The protein concentration was 18 M for pASK1-CD and 1.8 -180 M for 14-3-3WT. E, series of area-normalized c(s) distributions for mixtures of pASK1-CD and for 14-3-3⌬C at various molar ratios. The protein concentration was 18 M for pASK1-CD and 1.8 -180 M for 14-3-3⌬C. F, the catalytic activity of pASK1-CD and the 14-3-3 binding-defective S966A mutant both in the absence and the presence of 14-3-3⌬C. Human MKK4 (kinase-dead mutant K131A) was used as a substrate. The activities in the presence of 14-3-3⌬C are normalized to activities in its absence. The activity of kinase-dead ASK1-CD K709M mutant is normalized to the pASK1-CD activity. Results are the means Ϯ S.D., n ϭ 4. Specific activities are listed in supplemental Table S1.
The Guinier plots were sufficiently linear for all samples except a small deviation at the highest complex concentration (Fig. 2). Because the concentration dependence of R g and I(0) suggested the presence of repulsive interparticle interactions in samples of 14-3-3⌬C and attractive interparticle interactions in samples of ASK1-CD and the complex at highest protein concentrations (Table 1), data for 14-3-3⌬C, ASK1-CD, and the complex at concentrations of 7.7, 6.3, and 12.2 mg⅐ml Ϫ1 , respectively, were selected for ab initio shape reconstruction and structural modeling. The apparent M r of 68.2 and 51 for ASK1-CD and 14-3-3⌬C, respectively, estimated from I(0) correspond well with the expected M r of their dimers ( Table 1). The estimated M r of 117.4 for the complex as well as its V p of 236 nm 3 correspond well to a 2:2 molar stoichiometry (theoretical M r 128.9), thus corroborating results from SV AUC (for globular proteins, V p in nm 3 should numerically be ϳ1.7 times the M r ; Ref. 19).
The shape of the calculated distance distribution function, P(r), suggests that the complex is more extended compared with dimers of ASK1-CD and 14-3-3⌬C as its P(r) function shows longer intraparticle distances and a significantly larger maximum particle distance (D max ) ( Fig. 3A and Table 1). The normalized Kratky plots for both 14-3-3⌬C and ASK1-CD show bell-shaped curves with a maximum at sR g ϭ 1.7 consistent with folded molecules (Fig. 3B) (20). However, the normalized Kratky plot of the complex indicates an increase in the degree of the flexibility exhibited by the complex as indicated by the more gradual decrease of the curve toward zero at higher sR g and a maximum at sR g ϭ 2.2.
The superposition of ab initio molecular envelope of 14-3-3⌬C with the crystal structure of 14-3-3 (21) revealed correct reproduction of its molecular shape (Fig. 4A). The theoretical scattering curve calculated from the crystal structure fits the experimental data reasonably well (Fig. 4B, ϭ 0.97, R g ϭ 29.2 Å). The ab initio shape reconstruction statistics are listed in Table 2. The superposition of ab initio molecular envelope of ASK1-CD with the AllosMod model of ASK1-CD (659 -973) also showed correct reproduction of the shape and fit well the experimental SAXS data (Fig. 4, C and D, ϭ 1.00, R g ϭ 33.3 Å). The molecular envelope of the pASK1-CD⅐14-3-3⌬C complex is, as expected, significantly more extended compared with ASK1-CD and 14-3-3⌬C alone, and its shape, with one side being narrower than the other, suggests that protein components are arranged asymmetrically (Fig. 4E).
Chemical Cross-linking of the pASK1-CD⅐14-3-3⌬C Complex-The chemical cross-linking experiments were performed to obtain distance restraints for structural modeling of the complex. Identified intermolecular cross-links connecting four different regions of pASK1-CD with four different regions of 14-3-3⌬C are listed in Table 3. In several cases we were unable to identify the exact cross-linked residue, as the corresponding peptides contained multiple, usually closely located, lysine residues. Cross-link #1 connects the N-terminal amino group of Met-658 of pASK1-CD and the N-terminal part of helix H6 of 14-3-3⌬C containing two lysines, Lys-122 and Lys-138 (supplemental Fig. S2). Cross-link #2 was identified between the C terminus of helix ␣C from the N-lobe of pASK1-CD containing Lys-730 (or Lys-733) and the C terminus of helix H6 of 14-3-3⌬C containing Lys-158. Cross-link #3 connects Lys-925 from the loop between helices ␣H and ␣I of the C-lobe of pASK1-CD and Lys-85 from the helix H4 of 14-3-3⌬C. Cross-link #4 connects the C-terminal part of pASK1-CD containing lysines Lys-944, Lys-945, and Lys-946 and the C terminus of helix H3 of 14-3-3⌬C containing Lys-68. The last cross-link was identified between lysines Lys-944 -Lys-946 of pASK1-CD and residue Lys-120 from the helix H5 of 14-3-3⌬C. However, this cross-link was also observed for the mixture of non-phosphorylated ASK1-CD and thus was not used in structural modeling of the complex as it likely results from unspecific interaction between ASK1-CD and 14-3-3⌬C.
All-Atom Modeling of the pASK1-CD⅐14-3-3⌬C Complex-The all-atom modeling of the pASK1-CD⅐14-3-3⌬C complex was first performed using the AllosMod-FoXS method (22,23). The best-scoring single-state AllosMod model fits the SAXS data with ϭ 1.66 and shows the C-lobe of one protomer of ASK1-CD interacting with ␣-helices H4 and H6 of 14-3-3 outside its central channel through a relatively small binding interface (Fig. 5, A and B). Although this model satisfies three identified cross-links within a distance threshold of 30 Å and shows a reasonable agreement with the ab initio molecular envelope (Fig. 5C), calculated value together with theoretical values of R g and D max (43.1 and 156.7 Å, respectively), which are significantly smaller compared with values obtained from the SAXS data (Table 1), suggested this model does not reproduce the experimental SAXS data correctly.
Because normalized Kratky plot indicated an increased degree of flexibility exhibited by the complex (Fig. 3B), a multistate modeling of the complex for up to five states using the MultiFoXS method that enables modeling of conformationally heterogeneous systems was performed next (24,25). The resulting best-scoring model is based on a weighted combination of three states and provides a significantly improved fit to the SAXS data with of 0.97 as well as agreement with crosslinking data within a distance threshold of 30 Å (Fig. 6). Combinations of two, four, or five states did not yield any solution that would better describe the SAXS data compared with the three-state model. The model consists of two compact states with population weights of 43% (R g ϭ 41.6 Å, D max ϭ 153.5 Å)  (40). c Calculated using the program GNOM (41). d The excluded volume of the hydrated particle (the Porod volume). e Molecular weight was estimated by comparison of the forward scattering intensity I(0) with that from the reference solution of bovine serum albumin. Theoretical molecular weights of the 14-3-3⌬C dimer, ASK1-CD dimer, and the 14-3-3⌬C⅐ASK1-CD complex (with 2:2 stoichiometry) are 56.6, 72.2, and 128.9 kDa, respectively. and 18% (R g ϭ 37.9 Å, D max ϭ 139.0 Å) where the C-lobe of one protomer of ASK1-CD faces either the ␣-helix H6 or loops between helices H2-H3 and H4-H5 of 14-3-3 and one extended state with the population weight of 39% (R g ϭ 56.6 Å, D max ϭ 214.3 Å), where 14-3-3 and ASK1-CD are separated but still connected through the C-terminal segment of ASK1-CD. A significantly better agreement of this three-state model with the SAXS data compared with the single-state AllosMod model suggested that the pASK1-CD⅐14-3-3⌬C complex is conformationally heterogeneous in solution with both proteins sampling various mutual orientations. The 14-3-3 Binding Affects Local Environment and Dynamics of Thr(P)-838 and Thr(P)-842 in the Activation Segment of pASK1-CD-Structural modeling suggested that 14-3-3 interacts with the C-lobe of pASK1-CD in close proximity to both the active site and activation segment (Fig. 7). This brings up the possibility that the 14-3-3 binding might affect conformation of this region either directly or indirectly. Because the acti-  . Shape reconstructions calculated from the experimental SAXS data. A, ab initio shape reconstruction of 14-3-3⌬C calculated from SAXS data (represented as a gray envelope) with a superimposed crystal structure of 14-3-3 (21). B, comparison of the calculated scattering curve of the crystal structure of 14-3-3 (red line) with the experimental scattering data (black line). C, ab initio shape reconstruction of ASK1-CD calculated from SAXS data with superimposed AllosMod model of ASK1-CD (659 -973). D, comparison of the calculated scattering curve of ASK1-CD (659 -973) model (red line) with the experimental scattering data (black line). E, ab initio reconstruction of the pASK1-CD⅐14-3-3⌬C complex calculated from SAXS data. The ab initio shape reconstructions were performed using the program DAMMIF (43). Calculated molecular envelopes were aligned to structural models using the program SUPCOMB (45). Theoretical scattering curves were calculated from structural models and fitted to experimental data using CRYSOL (46). SEPTEMBER 23, 2016 • VOLUME 291 • NUMBER 39 vation segment contains two autophosphorylation sites that are phosphorylated in our recombinant protein (Thr(P)-838 and Thr(P)-842, supplemental Fig. S1), 31 P NMR experiments were performed to study structural changes of this region upon the 14-3-3 binding. The 31 P NMR spectrum of pASK1-CD alone showed a strong and narrow resonance at 4.6 ppm and 2 overlapping resonances at 3-4.25 ppm (Fig. 8). All these resonances lie within the range ϳ3-5 ppm expected for chemical shifts of Ser(P) and Thr(P) at neutral pH (26). The spectrum of the 14-3-3 binding-defective mutant pASK1-CD S966A alone contains only resonances at 3-4.25 ppm, thus strongly suggesting that resonance at 4.6 ppm corresponds to 31  Phe, and a single tryptophan was introduced at positions 823 and 839 by mutating the corresponding Phe to Trp. However, only the mutant containing single Trp at position 823 possessed sufficient solubility and stability for further experiments. We also generated mutants containing single Trp-770 or Trp-865 (by mutating the other Trp to Phe); nonetheless, only the mutant containing Trp-770 was soluble and stable. Residue Trp-770 is located at the C terminus of helix ␣D (Fig. 7); thus, this mutant was used to monitor structural changes in this part of ASK1-CD molecule. The stability of prepared mutants was checked by measuring the thermally induced protein denaturation using differential scanning fluorimetry. No significant differences in the temperature of the unfolding transition (T m ) were observed for prepared ASK1-CD mutants (supplemental Table S2). The fluorescently silent human 14-3-3noW missing all Trp residues was used in all tryptophan fluorescence measurements (27). Fluorescence decay measurements revealed that the complex formation significantly decreased the meanexcited state lifetime mean of Trp-823 from 5.8 to 5.4 ns ( Table  4). The mean shortening is clearly visible in the raw data presented in Fig. 9 where the difference between fluorescence decays in the presence and absence of 14-3-3noW is visibly above the data noise level. As can be noticed from the lifetime distribution analysis in Fig. 10, the decrease in mean is caused mainly by the binding-induced shortening of the two longest lifetime peaks representing together almost 70% of the total fluorescence intensity. These two peaks dominate the mean excited state lifetime, as mean ϭ ⌺ i (f i i ), where f i represents intensity fraction of the ith lifetime component i (28). Statistical significance of the observed difference is supported by the bootstrap confidence-interval analysis (29) presented in supplemental Fig. S3. The binding effect of 14-3-3noW on mean of Trp-770 was found to be considerably weaker ( Table 4). The suppression of mean indicates a significant 14-3-3 bindinginduced increase in polarity in the close vicinity of Trp-823 and/or change in quenching interactions upon the complex formation. This result corroborates the hypothesis that the 14-3-3 binding affects conformation of the activation segment of pASK1-CD.

Structural Analysis of the 14-3-3⅐ASK1 Complex
Analysis of the time-resolved emission anisotropy decays of pASK1-CD alone revealed three correlation times, one short ( 1 ϭ 2.5 ns), the second close to 18 ns ( 3 ) and the third very long unresolved ( 4 Ͼ 200 ns). Although the 1 reflects fast segmental motions, the 3 closely corresponds to the rotation of globular protein with M r ϳ70 (28) and can, therefore, be assigned to the overall rotational motion of pASK1-CD. The long correlation time 4 cannot be resolved with short Trp fluorescence lifetime and likely reflects a presence of some protein aggregates. For both mutants, the complex formation increased the correlation time to values corresponding to the overall rotational motion of protein with M r of 110 -150 (28), thus reflecting the higher molecular mass of the complex. Compared with pASK1-CD alone, complex formation resulted in an appearance of a new correlation time component ( 2 ), which is too long to be attributed to segmental motions. It rather reflects domain motions of pASK1-CD within the complex and/or the complex asymmetricity and structural heterogeneity as has been suggested by SAXS (Figs. 3B and 6).

Discussion
The location of the 14-3-3 binding motif close to the kinase domain of ASK1 (25 amino acid residues downstream of its C terminus) raised the possibility that the 14-3-3 protein binding might modulate the structure of this domain and/or the accessibility of its active site, as has been suggested for other enzymes regulated in the 14-3-3-dependent manner (13,14). The aim of this study was to investigate this hypothesis by performing structural analysis of the complex between 14-3-3 and pASK1-CD. The ab initio shape reconstruction of the complex from the SAXS data suggested that dimers of pASK1-CD and 14-3-3⌬C are arranged asymmetrically and not, as one would expect, symmetrically, as both proteins form dimers with 2-fold symmetry. The SAXS data analysis also suggested that the molecule of the pASK1-CD⅐14-3-3⌬C complex consists of several rigid domains that are tethered by linkers with rather compact conformation (Fig. 3B), thus resembling, for example, the NADPH oxidase activator p47 phox (20,30). This was further corroborated by the all-atom structural modeling, which revealed that only a model based on a weighted combination of multiple structural states can correctly reproduce both the SAXS and cross-linking data (Fig. 6). This suggested that the complex between 14-3-3 and pASK1-CD is dynamic and conformationally heterogeneous with both proteins sampling various mutual orientations. The absence of the large and well defined contact interface between 14-3-3 and pASK1-CD, besides interactions between the 14-3-3 ligand binding groove and the phosphorylated 14-3-3 binding motif of pASK1-CD, corresponds well with the weak binding affinity determined using SV AUC (Fig. 1, D and E) (31).

Cross-linked peptides Cross-linked residues
Observed mass Error ASK1-CD The suggested structural model also explains the observed stoichiometry of the complex in which one protomer of 14-3-3 dimer interacts with one protomer of pASK1-CD dimer, thus leaving one ligand binding groove of the 14-3-3 dimer and one 14-3-3 binding motif of the pASK1-CD dimer-free (Figs. 5 and 6). However, the formation of complexes with higher stoichiometries has not been observed even in the presence of large excess of one component (Fig. 1, D and E). The sampling of various conformational states likely causes steric hindrances, thus blocking the formation of complexes with higher binding stoichiometries. This, on the other hand, does not rule out the possibility that within the multiprotein ASK1 signalosome the 14-3-3 dimer interacts simultaneously with ASK1 and yet another protein, e.g. ASK2 (4). In addition, the suggested structural model of the pASK1-CD⅐14-3-3 complex also explains the observed partial inhibition of pASK1-CD activity in the presence of 14-3-3 (Fig. 1F) as only one protomer of pASK1-CD is bound to, and thus, inhibited by 14-3-3.

14-3-3⌬C ASK1-CD-14-3-3⌬C
Structural analysis also indicated that 14-3-3 interacts not only with the phosphorylated motif containing Ser(P)-966 but also with several regions from the C-lobe of pASK1-CD in the vicinity of the active site (Fig. 7). Although these contacts seem transient, it is reasonable to speculate they could affect, either directly or indirectly, the structure of the active site and/or its accessibility as has been observed for other kinases regulated through protein-protein interactions. For example, the cyclindependent kinase 2 is activated through the cyclin A-induced conformational change that realigns active site residues and, together with the conformational change of the activation segment, relieves the blockade of the active site (32). On the other hand, the inhibition of PKA is based on direct interaction of the activation segment with the regulatory subunit (33). Yet FIGURE 6. Multistate model of the pASK1-CD⅐14-3-3 complex. A, the best scoring three-state model of the 14-3-3⅐ASK1-CD complex contains two compact states with population weights of 43% (R g ϭ 41.6 Å, D max ϭ 153.5 Å) and 18% (R g ϭ 37.9 Å, D max ϭ 139.0 Å) and one extended state with the population weight of 39% (R g ϭ 56.6 Å, D max ϭ 214.3 Å). The 14-3-3 is shown in brown; ASK1-CD is shown in cyan (each protomer is shown in a different shade). The activation segment of ASK1-CD is shown in dark blue. The weighted combination of these three states (red line) agree well with experimental SAXS data (black line, ϭ 0.97). Multistate model analysis against the SAXS data was performed using the MultiFoXS server (24,25). B, models corresponding to states 2 and 3 from the best scoring three-state model of the 14-3-3⅐ASK1-CD complex satisfy the cross-linking data within a distance threshold of 30 Å. Intermolecular cross-links are labeled according to Table 3 (cross-linked residues are shown in red), and their distances are shown in parentheses.
another example is the inhibition of p38␣ kinase by MAPKactivated protein kinase 2, which blocks access to the p38␣ active site upon the complex formation (34). Results of 31 P NMR and time-resolved tryptophan fluorescence measurements (Figs. 8 -10 and Table 4) suggested that the 14-3-3 binding affects conformation of the activation segment of pASK1-CD, thus corroborating the hypothesis that the 14-3-3 protein interacts with pASK1-CD in close proximity to its active site. This suggests that the 14-3-3 protein might inhibit pASK1-CD through several mechanisms including structural modulation of its active site, the steric blocking of Thr-838 phosphorylation, and/or the blocking of interaction between ASK1 and its substrates.

Experimental Procedures
Preparation of the Catalytic Domain of ASK1-DNA encoding the catalytic domain and the 14-3-3 binding motif of human ASK1 (residues 659 -973) with the His 6 tag and TEV cleavage site at the C terminus was ligated into pST39 vector using XbaI and BamHI restriction sites. The 3-phosphoinositide-dependent protein kinase 1 (PDK1) is the only protein kinase shown to phosphorylate Ser-966 of ASK1 so far (35). The activity of commercially available PDK1, however, did not allow for preparation of milligram quantities of ASK1 (659 -973) stoichiometrically phosphorylated at Ser-966 needed for structural studies. Therefore residue Ser-964 was mutated to Arg, thereby making Ser-966 a substrate for the cAMP-dependent protein kinase (PKA). The mutant S964R (denoted as ASK1-CD throughout this work) as well as all other mutants of ASK1-CD were generated using the QuikChange site-directed mutagenesis kit (Stratagene), and mutations were confirmed by sequencing. ASK1-CD was expressed by leakage expression at 25°C for 18 h and purified from Escherichia coli Rosetta TM (DE3) cells using Chelating Sepharose Fast Flow (GE Healthcare) according to the standard protocol. The His 6 tag was cleaved by incubation with TEV protease (1 mg of TEV/30 mg of fusion protein) at 4°C for 12 h during dialysis against buffer containing 50 mM Tris-HCl (pH 7.5), 200 mM NaCl, 5 mM EDTA, and 2 mM ␤-mercaptoethanol. The final purification step was size-exclusion chromatography on a Superdex 200 HiLoad 26/60 column (GE Healthcare) in buffer containing 50 mM Tris-HCl (pH 7.5), 200 mM NaCl, 1 mM EDTA, 5 mM DTT, and 10% (w/v) glycerol.
Phosphorylation of ASK1 (659 -973)-ASK1-CD (1 mg⅐ml Ϫ1 ) was phosphorylated by incubation at 30°C for 2 h and then overnight at 4°C with 100 units of PKA (Promega) per mg of protein in the presence of 0.75 mM ATP and 12 mM MgCl 2 . After the phosphorylation, ATP was removed using size-exclusion chromatography. The result of phosphorylation reaction was verified using HPLC-MS. The HPLC-MS analyses confirmed almost stoichiometric phosphorylation of Ser-966 (supplemental Fig. S1D) and a significant autophosphorylation of Thr-838 and Thr-842 for the majority of samples (supplemental Fig. S1, B and C). If incomplete phosphorylation of Ser-966 was observed, then this procedure was repeated.
HPLC-MS Analysis-HPLC-MS was performed on a HPLC 1200 (Agilent Technologies, Waldbronn, Germany) connected to an electrospray ionization Fourier transform ion cyclotron MS (15T SolariX XR). Proteins were digested online on a pepsin column (66-l bed volume, flow rate 100 l⅐min Ϫ1 ), and the peptides were trapped on a Peptide MicroTrap (Michrom Bioresources, Auburn, CA). After complete desalting they were separated on a C18 reversed phase column (1 ϫ 100 mm, Jupiter, Phenomenex) using a linear gradient of 10 -35% solvent B in solvent A in 20 min, where solvent A was 2% acetonitrile and 0.4% formic acid in water, and solvent B was 95% acetonitrile, 5% water, and 0.4% formic acid. Peptides were identified by MASCOT against a database containing a sequence of human ASK1 (the only allowed partial modification was Ser/Thr phosphorylation). Data were analyzed using the program Data-Analysis version 4.2 (Bruker Daltonics).
Differential Scanning Fluorimetry-The thermofluor assay was performed using a real-time PCR LightCycler 480 II (Roche Applied Science) as described previously (14). The proteins at a concentration of 0.3 mg⅐ml Ϫ1 were tested in the presence of 8ϫ concentrated Sypro Orange (Sigma) in a total reaction volume of 25 l. AUC-SV Experiments Were Performed Using a Proteom-Lab TM XL-I Analytical Ultracentrifuge (Beckman Coulter) as described previously (37). Briefly, samples were dialyzed against buffer containing 50 mM Tris-HCl (pH 7.5), 200 mM NaCl, and 2 mM 2-mercaptoethanol before analysis. All sedimentation profiles were recorded with absorption optics at 42 000 rpm. The diffusion-deconvoluted sedimentation coefficient distributions c(s) were calculated from raw data using the package SEDFIT (38). SV AUC analysis of mixtures of pASK1-CD and 14-3-3 (WT or ⌬C mutant) at various molar ratios was performed with 18 M pASK1-CD and 1.8 -180 M 14-3-3. Obtained data were fitted, and apparent K D was estimated using the global Lamm equation modeling with a 1:1 Langmuir binding model implemented in the SEDPHAT software (39).
SAXS-SAXS data were collected on the European Molecular Biology Laboratory (EMBL) P12 beamline on the storage ring PETRA III (Deutsches Elektronen Synchrotron (DESY), Hamburg, Germany). The SAXS measurements were conducted in buffer containing 50 mM Tris-HCl (pH 7.5), 200 mM  NaCl, 1 mM EDTA, and 2 mM DTT. The ASK1-CD concentrations were 4.1-9.1 mg⅐ml Ϫ1 , the 14-3-3⌬C concentrations were 7.7-17.4 mg⅐ml Ϫ1 , and the pASK1-CD⅐14-3-3⌬C com-plex (2:2 molar stoichiometry) concentrations were 7.7-17.6 mg⅐ml Ϫ1 . The forward scattering I(0) and the radius of gyration R g were calculated using the Guinier approximation for the s (s ϭ 4sin()/, where 2 is the scattering angle, and is the wavelength) range, which satisfies the sR g Ͻ 1.3 condition (40). The distance distribution functions P(r) and the maximum particle dimensions D max were computed using the program GNOM (41). The solute apparent molecular mass (MM exp ) was estimated by comparing the forward scattering with that from reference solutions of bovine serum albumin. The excluded volume of the hydrated particle (the Porod volume, V p ) was computed using the program PRIMUS (42). The program DAM-MIF (43) was used to calculate ab initio molecular envelopes. Multiple iterations of DAMMIF were averaged using the program DAMAVER (44). Calculated molecular envelopes were aligned to structural models using the program SUPCOMB (45). Theoretical scattering curves were calculated from structural models and fitted to experimental scattering data using the programs CRYSOL (46) and FoXS (47).
Structural Modeling of the pASK1-CD⅐14-3-3 Complex Using SAXS and Cross-linking Data-The AllosMod-FoXS web server was used to prepare models of ASK1-CD and the ASK1-CD⅐14-3-3 complex as well as to generate their alternate conformations consistent with the experimental SAXS data (22,23). The model of ASK1-CD (659 -973) was prepared using the crystal structure of the kinase domain of ASK1 (670 -940) (48). The starting model of the complex was prepared using the model of ASK1-CD and the crystal structure of 14-3-3 (21). The C-terminal 14-3-3 binding motif of ASK1 containing Ser-966 (sequence 963 RSISLP 968 ) was restrained in the ligand binding groove of 14-3-3 in the orientation observed in structures of 14-3-3 with bound ligands (13,17,21). Simulations consisting of 10 -30 runs (generating 101 conformations for each run) were allowed to sample the most probable conformations consistent with the starting structures and used distances derived from intermolecular cross-links (Table 3, cross-links #1-4) restrained to 20 Ϯ 8 Å from C␣ to C␣. Multistate modeling of the pASK1-CD⅐14-3-3 complex was performed using the Mul-tiFoXS web server with the same starting structure (24,47). The ASK1-CD regions 659 -669 and 941-973 (with the exception of restrained 14-3-3 binding motif of one ASK1-CD protomer, region 963-968) were defined as flexible regions. The simulation generated 10,000 conformations that were used to interpret the SAXS data.
Chemical Cross-linking-Equimolar mixtures of both pASK1-CD and ASK1-CD with 1433⌬C were cross-linked using cross-linking agent bis(sulfosuccinimidyl)glutarate (BS 2 G) to identify intermolecular cross-links at the binding interface. Cross-linking reactions were performed with 80 M protein solutions in buffer containing 10 mM HEPES (pH 7.5), 200 mM NaCl, 0.1 mM TCEP (tris(2-carboxyethyl)phosphine). Reactions were started by adding freshly prepared stock solution of BS 2 G (10 mg⅐ml Ϫ1 ) in a 50ϫ molar excess to protein solutions, and reactions were incubated for 2 h at room temperature. After cross-linking, proteins were separated on a reducing NuPAGE 4 -12% Bis-Tris gel using MES running buffer, and the bands corresponding to both a monomeric and a dimeric cross-linked protein were excised and analyzed for intermolec- Raw data clearly show significant shortening of the decay in the presence of 14-3-3noW, as the difference between the two decays is visibly above the data noise level. ϩ denotes instrument response function. Excited state lifetime distribution of pASK1-CD mutant containing single tryptophan Trp-823 in the absence (unfilled distribution) and the presence of 14-3-3noW (gray-filled distribution). As can be noticed, the two longest lifetime peaks dominating the mean excited state lifetime become shortened upon the 14-3-3noW binding. As a consequence, mean of Trp-823 is reduced. Errors represent the S.D. ular cross-links using high resolution MS as described previously (49). All identified cross-links were manually checked in the raw data to remove false positives.
NMR Measurements-The 31 P NMR spectra with proton decoupling were measured on a Bruker AVANCE III instrument ( 1 H at 500.0 MHz and 31 P at 202.4 MHz) equipped with a 5-mm BBO CryoProbe. Phosphoric acid (85%) was used as external standard of 31 P chemical shifts. Typical experimental conditions for the 31 P spectra were 10,000 scans, with a spectral width of 400 ppm and an acquisition time of 0.4 s, yielding 64,000 data points. The repetition delay of 2 s led to the total measurement time of ϳ6 h per spectrum. The Lorenzian broadening by 20 Hz was applied in processing. Experiments were recorded at 23°C on samples containing 100 -200 M pASK1-CD (or the 14-3-3 binding-defective mutant pASK1-CD S966A) and 500 -900 M 14-3-3⌬C dissolved in 50 mM Tris-HCl (pH 7.5), 200 mM NaCl, 1 mM EDTA, 5 mM DTT, 10% (w/v) glycerol, and 10% (v/v) D 2 O. The resulting spectra had signal-to-noise ratio in the range of 20 -80.
Time-resolved Fluorescence Measurements-Time-resolved fluorescence intensity and anisotropy decay measurements as well as data analysis were performed as has been described previously (37). Tryptophan emission was excited at 298 nm by a tripled output of the Ti:Sapphire laser. Tryptophan fluorescence was isolated at 355 nm by a combination of monochromator and a stack of UG1 and BG40 glass filters (Thorlabs) placed in front of the input slit. Samples were placed in a thermostatic holder, and all experiments were performed at 23°C in buffer containing 50 mM Tris-HCl (pH 7.5), 200 mM NaCl, 1 mM EDTA, 5 mM DTT, 10% (w/v) glycerol, and 0.05% (w/v) Nonidet P-40. The pASK1-CD and 14-3-3noW concentrations were 15 and 110 M, respectively.