Structural Basis for the Regulation of the Mitogen-activated Protein (MAP) Kinase p38α by the Dual Specificity Phosphatase 16 MAP Kinase Binding Domain in Solution*♦

Background: Dual specificity phosphatases play a crucial role in MAP kinase regulation. Results: DUSP16 (MKP7) and p38α interact in a unique manner that is different from other DUSPs. Conclusion: DUSP16 binds p38α via an extended binding surface that includes helix α4. Significance: This study shows that DUSPs interact differently with p38α and lays the structural basis for their differential regulation of MAPKs. Mitogen-activated protein kinases (MAPKs) fulfill essential biological functions and are key pharmaceutical targets. Regulation of MAPKs is achieved via a plethora of regulatory proteins including activating MAPKKs and an abundance of deactivating phosphatases. Although all regulatory proteins use an identical interaction site on MAPKs, the common docking and hydrophobic pocket, they use distinct kinase interaction motif (KIM or D-motif) sequences that are present in linear, peptide-like, or well folded protein domains. It has been recently shown that a KIM-containing MAPK-specific dual specificity phosphatase DUSP10 uses a unique binding mode to interact with p38α. Here we describe the interaction of the MAPK binding domain of DUSP16 with p38α and show that despite belonging to the same dual specificity phosphatase (DUSP) family, its interaction mode differs from that of DUSP10. Indeed, the DUSP16 MAPK binding domain uses an additional helix, α-helix 4, to further engage p38α. This leads to an additional interaction surface on p38α. Together, these structural and energetic differences in p38α engagement highlight the fine-tuning necessary to achieve MAPK specificity and regulation among multiple regulatory proteins.

stress and inflammatory cytokines. It is their respective roles in these processes that have made MAPKs major drug targets over the last 15 years (2)(3)(4)(5)(6).
Dual phosphorylation of a Thr-X-Tyr sequence in the MAPK activation loop leads to a more than 1000-fold increase of their catalytic activity. This activation is orchestrated by a linear enzyme cascade in which MAPK kinases kinases activate MAPK kinases, which, in turn, activate the MAPKs. In contrast, the timely and spatial deactivation of MAPKs is performed by a large variety of phosphatases, including the tyrosinespecific phosphatases (kinase interaction motif-proteintyrosine phosphatases, KIM-PTP) 3 and dual specificity phosphatases (DUSPs) (7)(8)(9). All of these activating and deactivating regulatory proteins bind MAPKs via a KIM (also known as a D-motif) (10,11). The KIM binds to a binding site that is formed by a hydrophobic groove and an acidic patch named the common docking (CD) site; together, they are known as the KIM binding or D-motif recruitment site (12,13). KIM sequences contain basic residues that engage the CD site and a hydrophobic motif (⌽ A -X-⌽ B ) that engages the hydrophobic pocket. For many MAPK regulatory proteins, the KIM is an ϳ15-amino acid peptide sequence (e.g. for KIM-PTPs and MAPKKs) usually found in an unstructured N-terminal extension of the protein. The interaction of KIMs with MAPKs has been studied via multiple techniques, including x-ray crystallography as well as biomolecular NMR spectroscopy in solution (10,(13)(14)(15)(16)(17)(18)(19).
In contrast, the KIMs in DUSPs are part of well folded protein domains, the MAPK binding domains (MKBDs, ϳ15 kDa). DUSPs vary in size but typically contain an N-terminal MKBD and a C-terminal catalytic phosphatase domain. Of the 25 human DUSPs, 10 have a KIM-containing MKBD that mediates their direct interaction with MAPKs (8,9). The engagement of the DUSP MKBD with a MAPK functions both to localize the DUSP catalytic phosphatase domain to the phosphorylated MAPK activation loop residues, as well as, in some cases, to enhance the activity of the DUSP catalytic domain. Multiple structures of DUSP catalytic domains have been reported (20). In contrast, far fewer MKBDs have been structurally investigated. Moreover, despite the small sample size, the three-dimensional structures of the MKBDs from DUSP6 (MKP-3) (21), DUSP10 (MKP-5) (22), and DUSP16 (MKP-7) (23) are quite different. This raises the possibility that the differences in their structures may contribute to their differential selectivity and activity toward different MAPKs. Moreover, only a single structure of a MAPK⅐DUSP-MKBD (viz. p38␣⅐DUSP10/ MKP-5 MKBD) has been determined (23). Indeed, this complex showed that although the location of MKBD binding on p38␣ via its structured KIM is conserved, the mode of interaction is significantly different from that observed for the linear KIM peptides of most other regulators, e.g. KIM-PTPs.
The limited structural similarity between the DUSP MKBDs is due, in part, to their limited sequence conservation. For example, the sequence similarity of the MKBDs from DUSP10 and DUSP16 is only 32%. These sequence differences, in addition to the differences in their structures, also suggest that their mode of binding to MAPKs may not be strictly conserved. Furthermore, as observed previously, solution state studies, in addition to crystallographic studies, often reveal new insights into the structure and function of key signaling complexes (17)(18)(19)24). Thus, additional studies that investigate how, at a molecular level, other DUSPs interact with MAPKs are critical for elucidating the structural basis of specificity of these key regulatory proteins. Here we integrate biochemical, isothermal titration calorimetry (ITC), biomolecular NMR, and small angle x-ray scattering (SAXS) studies to determine how the MKBD of DUSP16 binds p38␣ in solution. Our study shows that the interaction between the MKBD of DUSP16 and p38␣ is stronger than those reported for KIM-PTPs peptides as well as the MKBD from DUSP10. In addition, our NMR results show that DUSP16 MKBD binding to p38␣ does not influence the chemical shift environment of the p38␣ hinge or activation loop. Furthermore, although the overall interaction modes, via helices ␣2 and ␣3 and the ␣2-␣3 loop, are similar between the MKBDs of both DUSP16 and DUSP10, the DUSP16 MKBD interacts more extensively and includes residues in helix ␣4. Taken together, although this is only the second study describing the interaction of a DUSP MKBD with a MAPK, this work has identified important structural differences in how these related MKBDs bind p38␣ that likely reflect the subtle structural and dynamic fine-tuning needed to achieve the tight regulation of MAPK activity in the cell.

EXPERIMENTAL PROCEDURES
Protein Cloning, Expression, and Purification-The coding sequences of DUSP16 MAP MKBD (corresponding to residues 5-138) were amplified using PCR, digested with NdeI/XhoI, and subcloned into a pET30a vector (Novagen) with a noncleavable C-terminal His 6 purification tag. Escherichia coli BL21 (DE3) RIL cells (Agilent) transformed with the expression vector for DUSP16 were grown at 37°C in LB broth containing selective antibiotics. The proteins were overexpressed by the addition of 1 mM isopropylthio-␤-D-galactoside when the optical density (A 600 ) reached 0.8 and the cultures were grown for an additional 18 -20 h at 16°C. Cells were harvested by centrifugation (6000 ϫ g, 12 min, 4°C) and stored at Ϫ80°C until purification.
HADDOCK Calculations-HADDOCK (30,31) was used to dock p38␣ and the DUSP16 MKBD using ambiguous NMRderived restraints. p38␣ (Protein Data Bank (PDB) ID 1P38 (32)) and the DUSP16 MKBD (PDB ID 2VSW) were used as inputs. Active residues were defined as those that experience a chemical shift perturbation (CSP) (greater than the mean plus 1) or are broadened beyond detectability and have high solvent accessibility (side chain or total accessible surface area of greater than 50%) in the unbound protein structure as calculated using the program MOLMOL (33 Isothermal Titration Calorimetry-ITC experiments were performed at 25°C using a VP-ITC microcalorimeter (Micro-Cal Inc.). Titrant (10 l per injection) was injected into the sample cell over a period of 20 s with a 250-s interval between titrations to allow for complete equilibration and base-line recovery. 28 injections were delivered during each experiment, and the solution in the sample cell was stirred at 307 rpm to ensure rapid mixing. To determine the thermodynamic parameters (⌬H, ⌬S, ⌬G) and binding constants (K), the DUSP16 MKBD was titrated into p38␣, and the data were analyzed with a one-site binding model assuming a binding stoichiometry of 1:1 using Origin 7.0 software. A nonlinear least squares algorithm and the titrant and sample cell concentrations were used to fit the heat flow per injection to an equilibrium binding equation, providing values of the stoichiometry (n), change in enthalpy (⌬H), and binding constant (K). All data were repeated in duplicate.
Small Angle X-ray Scattering-The p38␣⅐DUSP16 MKBD complex was generated as described above. Data were collected at 0.8, 2.2, and 4.5 mg/ml. All samples were prepared within 48 h of data acquisition and stored on ice at 4°C. All samples were filtered (0.02-M filter, Whatman) immediately prior to data collection. All data were recorded at beamline X9 at the National Synchrotron Light Source (NSLS) using a Dectris Pilatus 300k (3.40 m from the sample for SAXS) and a Photonic Science CCD (0.47 m from the sample for wide angle X-ray scattering) detector. 20 l of sample was continuously flowed through a 1-mm diameter capillary and exposed to an x-ray beam for 60 s. Normalization for beam intensity, buffer subtraction, and merging of the data from both detectors were carried out using PRIMUS (34). A Guinier approximation, I(q) ϭ I(0)exp(Ϫq 2 R g 2 /3), where a plot of ln(I(q)) and q 2 is linear for q Ͻ 1.3/R g , was performed on at least four independent scattering trials and averaged to determine the radius of gyration. The linearity of the Guinier region and the intensity at zero scattering angle, I(0), were used to validate that all samples were monodisperse in solution. I(0)/c, where c is concentration, was consistent for all measurements for a single complex. GNOM (35) was used to determine the pair distribution function, P(r), for each complex. 24 envelopes were generated for each complex using GASBOR (36) and were aligned and averaged using the DAMAVER program suite (37).

RESULTS
The MKBD of DUSP16-As expected (23), the DUSP16 MKBD (residues 5-138, 15.9 kDa; Fig. 1A) showed high levels of soluble overexpression in E. coli and behaves as a monomer in solution as verified by SEC. NMR analysis of the DUSP16 MKBD showed that the sequence-specific backbone assignment of 123 of the 132 expected backbone amide NH pairs (93.2%, 2 prolines; excluding the C-terminal His 6 tag and a 3-residue cloning artifact) can be readily achieved (Fig. 1B). Interestingly, no assignment was possible for residues Cys-50 -Met-54, which are part of the KIM motif, as well as for Met-5, Ile-6, Ser-100, and Phe-103. The weak intensity of peaks N-terminal to Cys-50 and C-terminal to Met-54 indicate that these peaks are likely broadened beyond detectability by conformational exchange, indicating that these residues sample multiple conformations in the s-ms time regime in the free form of the DUSP16 MKBD. Chemical shift index calculations derived from C ␣ and C ␤ chemical shifts showed that the secondary structure elements of the DUSP16 MKBD in solution are highly similar to that observed in the DUSP16 MKBD crystal structure (23). Furthermore, 15 N, 1 H NOE (heteronuclear NOE) analysis showed that the DUSP16 MKBD has very limited fast time scale (ps/ns) dynamics in its unbound form (Fig. 1C).
Direct comparison of the two-dimensional 1 H, 15 2D). The perturbed residues, although largely clustered within the CD site and the hydrophobic binding groove, are decisively different from those that interact with KIM-PTPs (13-15, 18, 19). Particularly, the typical interaction with p38␣ residues in helix ␣C, strand ␤4, and the ␣C-␤4 loop, residues that flank the

p38␣⅐DUSP16 MKBD Complex
⌽ B binding pocket, is missing. These data show that neither the MKBD of DUSP10 (23) nor the MKBD of DUSP16 (this work) requires the ⌽ B pocket for binding p38␣.
Interestingly, no CSPs are detected for residues in the p38␣ hinge or the p38␣ activation loop. This shows that these regions of p38␣ do not experience a change in their chemical shift environment upon DUSP16 MKBD binding. This suggests that the lack of electron density for the p38␣ activation loop in the p38␣⅐DUSP10 MKBD crystal structure is likely due to a change in the dynamics of the activation loop upon MKBD binding, potentially via an allosteric effect.
Identification of the DUSP16 MKBD Residues That Mediate p38␣ Binding-We also defined the binding interface of p38␣ on the DUSP16 MKBD. In this experiment, [ 2 H, 15 MKBD was incubated with unlabeled p38␣, the p38␣⅐DUSP16 MKBD complex was purified by SEC, and a two-dimensional 1 H, 15 N TROSY spectrum was recorded (Fig. 3A). Direct comparison of the unbound and p38␣-bound DUSP16 MKBD twodimensional 1 H, 15 N TROSY spectra revealed large differences between the spectra, making chemical shift tracking analysis difficult. Thus, we completed the sequence-specific backbone assignment of the DUSP16 MKBD when bound to p38␣. For these studies, [ 2 H, 15 N, 13 C]DUSP16 MKBD was incubated with unlabeled p38␣, purified by SEC, and concentrated to ϳ500 M, and a two-dimensional 1 H, 15 N TROSY spectrum and a three-dimensional TROSY-HNCA spectrum were recorded. Analysis of the NMR spectra allowed for the sequence-specific backbone assignment of the p38␣-bound DUSP16 MKBD and subsequent CSP analysis. As observed for unbound DUSP16 MKBD, residues Cys-50 -Met-54 were also not detected when bound to p38␣.
Comparison of the two-dimensional 1 H, 15 Fig. 3, B and C). The perturbed   (23)), but also includes additional residues outside this site on both p38␣ and DUSP16 MKBD, both of which are consistent with a strong K D for this complex as measured by ITC (Fig. 2B).
The p38␣⅐DUSP16 MKBD Complex-In the three-dimensional crystal structure of the p38␣⅐DUSP10 MKBD complex (23), the DUSP10 MKBD adopted a conformation nearly identical as that in its unbound form, suggesting that the conformations of MKBDs change little, if at all, upon MAPK binding. Furthermore, 15 N, 1 H NOE NMR data for the DUSP16 MKBD confirmed that in its unbound form, this protein is rigid. Lastly, the 13 C␣ chemical shifts of the unbound and p38␣-bound DUSP16 MKBD are essentially identical, showing that the secondary structure of the DUSP16 MKBD does not change upon p38␣ binding. Therefore, the DUSP16 MKBD is an ideal protein for performing NMR data-driven docking using the program HADDOCK (30,31). We used our NMR CSP data, which identified the residues that compose the interaction interface of the p38␣⅐DUSP16 MKBD complex, together with residue surface accessibility, to restrain the docking of the DUSP16 MKBD (PDB ID 2VSW) with p38␣ (PDB ID 1P38 (32)) using HADDOCK. The lowest energy cluster (HADDOCK score: Ϫ134.5 Ϯ 6.2) contained 107 structures with a root mean square deviation of 0.5 Ϯ 0.3 Å. The lowest energy structure from this cluster is shown in Fig. 4A. In this model, the DUSP16 MKBD helices ␣2, ␣3, and ␣4 bind to p38␣ in its KIM binding site. The interface between the two proteins buries 2035 Å 2 of surface area, 185 Å 2 (ϳ10%) larger than the p38␣⅐DUSP10 MKBD complex, which buries 1850 Å 2 of surface area and does not interact with p38␣ via helix ␣4.
To experimentally test the overall validity of the p38␣⅐DUSP16 MKBD model, we recorded SAXS data for the p38␣⅐DUSP16

p38␣⅐DUSP16 MKBD Complex
MKBD complex (Fig. 4, B and C). We compared the calculated (determined using Hydropro (38)) R g of the lowest energy p38␣⅐DUSP16 MKBD complex model (R g ϭ 28.4 Å) with the R g determined experimentally ( Table 2). The strong consistency between the R g values is further supported by the excellent fit of the lowest energy p38␣⅐DUSP16 MKBD HADDOCK model into the molecular envelope determined for the p38␣⅐DUSP16 MKBD complex (Fig. 4D, Table 2).
Consistent with the NMR chemical shift data, the HADDOCK model of the p38␣⅐DUSP16 MKBD complex showed that the protein-protein interface of p38␣ with the DUSP16 MKBD is largely similar to that of p38␣ with the DUSP10 MKBD (Fig. 5). Specifically, Phe-34 and Val-35 anchor the DUSP16 MKBD helix ␣2 to the p38␣ ⌽ A interaction site (Fig. 5A). As observed in the p38␣⅐DUSP10 MKBD complex, no interaction of the DUSP16 MKBD with the p38␣ ⌽ B pocket is identified in solution. Furthermore, many interactions by the DUSP16 MKBD residues 47-55, which are N-terminal to and within helix ␣3, with p38␣ residues in the p38␣ acidic patch are clearly important for the observed nM K D interaction between these two proteins. Finally, the p38␣ ⌽ H interaction is conserved (Fig. 5B). However, and different from the p38␣⅐DUSP10 MKBD complex, helix ␣4 from the DUSP16 MKBD also interacts directly with p38␣ (Fig.  5C). Indeed, even in the unbound DUSP MKBD structures (23), helix ␣4 of the DUSP16 MKBD is approximately one turn longer than that of the DUSP10 MKBD. Seven residues of the DUSP16 MKBD helix ␣4 show CSPs, and one residue (Ala-73) is broadened beyond detectability upon p38␣ binding. A key sequence difference between DUSP16 and DUSP10 is His-71 (Cys-218 in DUSP10). In the HADDOCK model, His-71 is predicted to make a polar interaction with p38␣ Gln-128 (helix ␣ e ), which results in helix ␣4 being in much closer proximity to p38␣. Taken together, DUSP16 MKBD helix ␣4 interacts with p38␣ residues on helices ␣ e and ␣ i . To confirm this interaction biochemically, we produced the DUSP16 MKBD H71A/S72A mutant and measured its affinity for p38␣ using ITC. The onedimensional 1 H NMR spectra of the DUSP16 MKBD H71A/  S72A mutant is nearly identical to WT DUSP16 MKBD, showing that the protein is well folded in solution. As predicted, the measured K D decreased about 2-fold (Fig. 5D, Table 1).

DISCUSSION
The activity of MAPKs is finely tuned in a cell type-specific and temporal manner by the concerted effort of multiple regulatory proteins. MAPKs are activated by a linear cascade of kinases, namely MAPKKKs and MAPKKs. In contrast, deactivation is highly diverse and achieved by a variety of phosphatases, including protein-Tyr-specific phosphatases (KIM-PTPs), dual specificity phosphatases (DUSPs, MKPs), and Ser/ Thr phosphatases (e.g. protein phosphatase 2A (PP2A)). In particular, KIM-PTPs and DUSPs share a common MAPK interaction/anchoring site on MAPKs (13-15, 17, 19, 23, 39, 40), a site also used by MAPKKs (15,41) and substrates (42)(43)(44). A comprehensive understanding of how the activity of MAPKs is finely controlled by their MAPK regulatory proteins can only be obtained by understanding how these proteins interact at a molecular level as it is the small structural and energetic differences between these proteins that ensure pathway fidelity.
Our work provides new insights into these subtle, yet critical differences. For example, a direct comparison of the K D values between KIM peptides derived from KIM-PTPs (17,19,39) and MAPKs shows that the interaction between p38␣ and the DUSP16 MKBD, which contains a structured KIM, is ϳ5-10 times stronger. Furthermore, we show that the interaction of a MAPK with KIMs from proteins within a single family, namely the DUSP MKBDs, is also structurally distinct. Specifically, although the KIMs from KIM-PTPs or MAPKKs are flexible and linear, which allow them to interact with MAPKs in a continuous and extended manner, the binding possibilities of the KIMs from DUSP10 and DUSP16 are much more restricted as they are part of a well folded domain. Thus, the DUSP16 MKBD, like the DUSP10 MKBD, binds p38␣ via helices separated by loops that do not themselves continuously interact with p38␣. The chain directionality is also different. Although most linear KIM peptides bind to the CD and hydrophobic pockets of MAPKs in an N-to-C-terminal direction, the DUSP FIGURE 5. Interaction of DUSP10 and DUSP16 with p38␣ is different. A, overlay of the p38␣⅐DUSP16 MKBD complex (p38␣ is shown as gray surface; DUSP16 MKBD is shown as a wheat ribbon) with the p38␣⅐DUSP10 MKBD complex (DUSP10 MKBD is shown as a light blue ribbon), illustrating the interaction of helix ␣2 with the p38␣ ⌽ A pocket. Key residues from DUSP10 and DUSP16 are shown as sticks and labeled. B, same as A but illustrating the interaction of helix ␣3 with the p38␣ ⌽ H pocket. Key residues from DUSP10 and DUSP16 are shown as sticks and labeled. C, same as A but illustrating the interaction of DUSP16 MKBD helix ␣4 with p38␣. His-71 from DUSP16 is shown as sticks and labeled. D, raw isothermal titration calorimetry data (upper panels) and derived binding isotherm plotted versus the molar ratio of titrant fit using a one-site model (lower panels) for p38␣ with the DUSP16 MKBD H71A/S72A mutant.

p38␣⅐DUSP16 MKBD Complex
MKBDs bind differently, with the N-terminal helix ␣2 binding in the hydrophobic pocket and more C-terminal helix ␣3 binding at the CD site. This reversed directionality has also recently been identified in the linear KIM peptide from PEA-15 (44), and thus, it has become evident that the hydrophobic interaction site on MAPKs is bidirectional binding-competent. Interestingly, we have also recently shown that the family of KIM-containing PTPs bind very differently to p38␣, showing that structural differences can account for the different activities of the regulatory enzymes (45).
Clearly, there are significant energetic and structural differences in the modes of interaction between MAPKs and their different families of regulatory proteins. Even more interesting, and what we show here, is that there are also key differences in these interactions between proteins from a single family. Specifically, although both DUSP10 and DUSP16 belong to the family of KIM-containing DUSPs that are selective for JNK and p38, there are clear differences in how they bind p38␣. Although the DUSP10 MKBD primarily uses residues in helices ␣2 and ␣3 as well as part of the ␣2-␣3 loop for its interaction with p38␣, the DUSP16 MKBD uses not only residues in these helices but also residues in helix ␣4. Taken together, this additional binding surface leads to an enhanced interaction between these two proteins. This shows that although the ␣2 and ␣3 interaction is likely conserved between the family of DUSP MKBDs, the additional interactions of C-terminal (DUSP16) and potentially N-terminal residues in other DUSPs are likely important for different MAPK interaction strengths and, in turn, MAPK regulation and signal fidelity.