|Home | About | Journals | Submit | Contact Us | Français|
MAP kinases regulate essential cellular events, including cell growth, differentiation and inflammation. The solution structure of a complete MAPK–MAPK-regulatory protein complex, p38α–HePTP, was determined, enabling a comprehensive investigation of the molecular basis of specificity and fidelity in MAPK regulation. Structure determination was achieved by combining NMR spectroscopy and small-angle X-ray scattering data with a new ensemble calculation–refinement procedure. We identified 25 residues outside of the HePTP kinase interaction motif necessary for p38α recognition. The complex adopts an extended conformation in solution and rarely samples the conformation necessary for kinase deactivation. Complex formation also does not affect the N-terminal lobe, the activation loop of p38α or the catalytic domain of HePTP. Together, these results show how the downstream tyrosine phosphatase HePTP regulates p38α and provide for fundamentally new insights into MAPK regulation and specificity.
Mitogen-activated protein kinases (MAPKs; ERK, p38, JNK) are serine/threonine kinases that mediate the cellular response to a variety of extracellular stimuli ranging from growth factors to inflammatory cytokines. The activity of MAPKs is finely tuned in a cell type–specific and temporal manner by the concerted effort of multiple regulatory proteins, including activating kinases (such as MEK1, MKK3 and MKK6), deactivating phosphatases (such as KIM-PTPs and MKPs) and scaffolding proteins (such as KSRs)1. Critically, disruptions in the tight regulation of the MAPK signaling pathways are directly correlated with numerous diseases, including Alzheimer’s disease, rheumatoid arthritis and cancer2.
The MAPK pathways consist of three components: a MAP kinase kinase kinase (MAP3K), a MAP kinase kinase (MAP2K) and a MAPK3. Stimulation of the pathway ultimately results in MAPK activation by the dual phosphorylation of a threonine and a tyrosine residue (Thr-Xaa-Tyr) located in the MAPK activation loop. Though the kinase cascades that direct MAPK activation seem stereotypical, the guidance and fine-tuning that direct cell- and situation-specific responses are governed by MAPK-specific phosphatases. In particular, MAPKs are dephosphorylated by the kinase interaction motif protein tyrosine phosphatases (KIM-PTPs), 11 ‘typical’ MAP kinase phosphatases (MKPs) and several ‘atypical’ dual-specific phosphatases as well as serine/threonine phosphatases4. The reason for this abundance of phosphatases is most likely related to the many roles that MAPKs play in the cell, which is profoundly affected by the duration of MAPK activation and the subcellular localization of active MAPKs. A comprehensive understanding of how the activity of MAPKs is finely controlled by MAPK regulatory proteins can only be obtained by understanding how these proteins interact at a molecular level.
MAPKs have a two-lobed architecture with a five-stranded β-sheet in the N-terminal lobe and six α helices in the C-terminal lobe5. The activation loop, which contains the Thr-Xaa-Tyr motif, is at the top of the C-terminal domain immediately adjacent to the N-terminal domain. In 1994, it was discovered that MAPKs contain a docking site in the C-terminal lobe6. This docking site is located on the face of the C-terminal lobe opposite to the activation loop and binds a conserved 13- to 16-amino-acid sequence known as the kinase interaction motif (KIM) or the D motif. This motif has since been identified in nearly every MAPK regulatory and substrate protein. The KIM is characterized by a group of basic residues (two to three lysines or arginines) and two hydrophobic residues (ΦA-Xaa-ΦB) that mediate the interaction with the KIM binding site of MAPKs. Structures of MAPKs bound to KIM peptides from activating kinases, deactivating phosphatases and substrates have revealed new insights into the molecular determinants that direct these interactions7–13. However, though the KIM is required for binding14,15, it has recently become clear that residues outside the KIM are necessary to confer specificity16–18.
The KIM-PTPs are a small family of phosphatases that include hematopoietic tyrosine phosphatase (HePTP; immune system–specific)19, striatum-enriched phosphatase (STEP; brain-specific)20 and STEP-like PTP (PTPRR or PTP-SL; brain-specific)21. Each phosphatase possesses a C-terminal catalytic domain (the PTP domain) and an N-terminal unstructured extension that contains the 15-amino-acid KIM. KIM-PTPs regulate the activity of p38 and ERK by both protein localization (by retaining the MAPKs in the cytosol) and dephosphorylation of their active sites22. Despite a wealth of crystallographic data for p38 (refs. 5,7,8,23,24), little is known about the atomic-level detail of its interaction with regulatory proteins outside the KIM binding pocket and the role that these residues have in distinguishing between MAPK regulatory proteins to ensure pathway specificity. Though current structures provide clues to the conformational variability in p38 in the free and KIM peptide–bound forms, additional structures are necessary to obtain comprehensive structural models that describe the molecular interactions between MAPKs and the full-length regulatory proteins that define MAPK pathway specificity and fidelity.
Here, we combine biomolecular NMR spectroscopy, small-angle X-ray scattering (SAXS) and isothermal titration calorimetry (ITC) to determine how p38α, the p38 isoform that is highly and ubiquitously expressed in most cell types, is recognized by its KIM-PTP regulator HePTP under functional conditions. We show that, although the KIM peptide is essential for binding, p38α residues outside the KIM binding pocket play a central role in HePTP recognition. We also show that the interaction between p38α and HePTP is largely confined to the unstructured HePTP N terminus and the KIM-PTP catalytic-domain extension. Finally, using SAXS and a new ensemble-calculation routine that incorporates chemical-shift information derived from NMR spectroscopy, we show that the complex adopts an extended conformation in solution and provide an atomic model of the p38α–HePTP complex. Collectively, these results provide new insights into the molecular interactions that regulate the strength and duration of MAPK signaling and, in turn, provide unique avenues for therapeutic interventions of MAPK-related diseases.
NMR spectroscopy is a powerful technique for defining the interactions of proteins at atomic resolution. However, a precise definition of these interactions requires a highly complete sequence-specific backbone assignment of the protein (or proteins) under investigation. Because of their large size (~40 kDa) and complex dynamic characteristics, serine/threonine kinases are notoriously difficult targets for NMR spectroscopy, and only a few of them, such as the serine/threonine kinase PKA25,26 and the tyrosine kinase Abl27, have been investigated in detail using this technique. This is especially true for the MAPKs, for which only two sequence-specific backbone assignments have been reported28,29. However, unlike the assignments for PKA (80% complete)30 or Abl (96% complete)31, the assignment of p38α is only partially complete (64% complete)29. Moreover, many functionally important residues are unassigned.
To further our understanding of these previously unassigned regions of p38α, we improved the sequence-specific backbone assignment of p38α to 82% (Fig. 1a–d). Of the 334 expected peaks for nonproline residues, 297 peaks were observed in the two-dimensional 1H,15N transverse relaxation-optimized spectroscopy (TROSY) spectrum of p38α. The 37 peaks not observed could be missing because of fast solvent (H2O)-amide exchange on the NMR timescale; lack of D2O back exchange after protein growth in D2O-based expression medium, especially for amino acids in the hydrophobic core; or intermediate conformational exchange, which broadens the peaks beyond detection. To minimize solvent-amide exchange, all experiments were performed at pH 6.8 using water suppression by gradient-tailored excitation (WATERGATE)–based NMR experiments. In addition, analysis of unassigned residues suggested that only a small fraction of residues are sufficiently buried in the hydrophobic core to be missing as a result of slow D2O back exchange. Thus, the majority of missing peaks are probably in a microsecond-millisecond conformational exchange regime that broadens the peaks beyond detection.
Most newly assigned residues are located in functionally critical regions of p38α (Fig. 1b). The current assignment now includes 88% of the N-terminal lobe, including all of the GxGxxG ATP binding motif; 83% of the activation loop, which contains the Thr-Gly-Tyr phosphorylation site; 100% of the Thr-Gly-Tyr motif; 74% of the MAPK insert (residues 243–273), which includes helices α1L14 and α2L14 and forms a C-terminal domain binding pocket that is currently being exploited for the development of new p38-specific inhibitors32; 100% of the αEF-αF loop, which forms a platform on which the activation loop folds; 100% of helix αH; and 93% of the KIM-peptide binding pocket. Taken together, this is the most complete sequence-specific backbone assignment of any MAPK reported to date.
The assignment of residues surrounding the ATP binding pocket were confirmed by titration of both 2′,3′-(2,2,5,5-tetramethyl-3-pyrroline-1-oxyl-3-carboxylic acid ester) adenosine triphosphate (SL-ATP)33,34, a spin-labeled ATP analog, and the prototypical p38α inhibitor SB203580 (ref. 35), a competitive inhibitor of ATP that binds at the ATP binding site, into [2H,15N]p38α (two- to four-fold excess of titrant) using two-dimensional 1H,15N-TROSY spectroscopy. As expected, only residues surrounding the ATP binding pocket showed either reduced intensities (SL-ATP; Supplementary Results, Supplementary Fig. 1) or large chemical shift changes (SB203580; Fig. 1c). In addition, titration with SB203580 did not change the number of observed peaks (296 peaks), demonstrating that the microsecond-millisecond intermediate timescale dynamics of p38α are unchanged when it is bound to SB203580.
Notably, the largest contiguous patch of unassigned p38α residues connects the N- and the C-terminal lobes of p38α (Supplementary Fig. 2a,b). This connection is mediated by the DFG loop and stretches into the MAPK insert in the C-terminal lobe. Titration with BIRB796 (Fig. 1d), a prototypical DFG-out inhibitor that binds and stabilizes the DFG loop in the inactive ‘out’ conformation36, resulted not only in chemical shift perturbations of residues near the BIRB796 binding site but also in altered microsecond-millisecond intermediate timescale p38α dynamics, as a total of 15 additional peaks were observed in the spectrum. Thus, BIRB796 changes the conformational equilibrium of all DFG loop residues, including Phe169 (ref. 36), as well as additional residues in this region (Supplementary Fig. 2c). This is notable as this finding correlates well with a recently introduced model for the catalytic reactions of kinases that is based on hydrophobic spines that coordinate internal kinase dynamics37. The hydrophobic spine, the R-spine (Phe169, unassigned; Leu75, unassigned; Leu86, assigned; His148, assigned), is centered on the DFG loop. Our data show that stabilization of Phe169 (for example, by an inhibitor such as BIRB796) and, in turn, other spine residues such as Leu86 influence kinase dynamics.
We used NMR spectroscopy to define the complete p38α–HePTP resting-state complex interface at atomic resolution (Fig. 2a–e). In this experiment, [2H,15N]p38α was incubated with unlabeled HePTP (constructs used in this study are illustrated in Fig. 2a), the 78-kDa p38α–HePTP complex was purified by size-exclusion chromatography and a two-dimensional 1H,15N-TROSY spectrum of the complex was recorded. The complex interacts tightly (Kd = 2.5 ± 0.6 μM; Table 1 and Supplementary Fig. 3) and is stable for days at 23 °C. The spectrum is of excellent quality for a complex of this size (78 kDa), allowing the chemical shift perturbations in p38α caused by HePTP binding to be readily detected (Fig. 2b).
Direct comparison of the two-dimensional 1H,15N-TROSY spectra of free and HePTP-bound p38α revealed chemical shift perturbations in 45 peaks, 34 of which were more than 4 s.d. (σ) above the mean (Fig. 3a,b). As expected, a subset of the chemical shift perturbations were localized to the KIM binding pocket, including helix αD, the αD-αE loop, the β7-β8 reverse turn, the ED site and the CD site (the latter two sites are often collectively referred to as the acidic binding pocket). However, perturbation of residues that are distal to the KIM binding pocket were also detected, including multiple residues from the αC-β4 loop, the αE-β7 loop, helix αF and the αF-αG loop. To our knowledge, this demonstrates for the first time that residues outside the p38α KIM binding pocket contribute to HePTP interaction and thus are important for binding.
The structure of the HePTPKIM peptide (HePTP residues 16–31) in complex with ERK2 (ref. 13), as well as three structures of KIM peptides (one with MEF2a and two with MKK3b) bound to p38α (ref. 7,8), have been reported. In the p38α–KIM peptide complexes, no interaction of the peptides with the CD domain (Asp313 and Asp316) was observed. In contrast, the HePTPKIM peptide makes multiple contacts with the CD domain of ERK2. Our chemical shift perturbation data with the p38α–HePTP complex unequivocally confirm the interaction of HePTP with the p38α CD and ED domains in solution, which involves CD domain residues Gln310–Asp313, Asp315–Glu317 (Pro314 is undetectable in the two-dimensional 1H,15N-TROSY spectrum) and ED residues Glu160 and Asp161. Thus, HePTP clearly interacts with both the ED and the CD sites in the p38α–HePTP complex and, based on the observed chemical shift changes, most likely adopts a conformation similar to that observed in the ERK2–HePTPKIM peptide complex. This demonstrates a critical point for MAPK selectivity: the sequence of the KIM, and not the identity of the MAPK, defines how KIM peptides from different MAPK effector proteins bind and interact with MAPKs.
To identify the residues of HePTP interacting with p38α, we isolated the individual domains of HePTP that most likely contribute to p38α binding, namely the KIM peptide (HePTPKIM31; Fig. 2a) and the HePTP catalytic domain (HePTPCAT; Fig. 2a). We then used NMR chemical shift perturbation mapping experiments to determine which of these domains participate in binding. First, we probed the interaction of the HePTPKIM31 peptide with p38α at four different protein-peptide ratios (Supplementary Fig. 4). Upon addition of the HePTPKIM31 peptide to [2H,15N]p38α, 35 large chemical shift changes corresponding to residues in the p38α C lobe and Leu108 in the hinge region were observed. Consistent with the p38α–HePTP complex chemical shift perturbation experiment, no substantial chemical shift differences were observed for peaks corresponding to residues in the p38α N-lobe (Figs. 2c and 3a,b). Remarkably, the total number of perturbed peaks was lower in the titration of p38α with HePTPKIM31 than in the p38α–HePTP complex. Whereas complex formation between p38α and HePTP resulted in large chemical shift changes in multiple residues from helix αD, the αD-αE loop, the β7-β8 reverse turn, the ED site and the CD site, titration with HePTPKIM31 resulted in fewer shifts in these regions (Fig. 3a,b). Moreover, titration with HePTP resulted in multiple shifts outside the KIM that were not observed with HePTPKIM31 alone. This shows that, though HePTPKIM31 binds and interacts with p38α, it does not recapitulate the shifts observed with HePTP, demonstrating that surfaces outside the KIM binding pocket are essential for the formation of the p38α–HePTP complex.
We also investigated the role of the HePTP residue Val31. Sequence alignments with other MAPK KIMs predicted that the HePTP ΦA-X-ΦB residues correspond to Leu27-Xaa-Leu29. However, the crystal structure of the ERK2–HePTPKIMpeptide complex13 revealed a register shift in the ΦA-Xaa-ΦB sequence in which Leu27 binds in a third hydrophobic pocket, ΦA-2, and Leu29 actually binds in the ΦA pocket. Similar ΦA-2 and ΦA interactions have been seen for the MEF2a peptide with p38α (ref. 8). Thus, Val31 putatively binds in the ΦB pocket. However, the absence of electron density observed for Val31 indicates flexibility and thus potential dispensability for binding. This is consistent with the observation that ERK2-bound KIM peptide conformations are heterogeneous, with only weak overlap between corresponding ΦA and ΦB residues. In contrast, the KIM peptides bound to p38α adopt highly similar conformations, and the corresponding ΦA and ΦB residues overlap nearly perfectly (Fig. 4a,b and Supplementary Fig. 5).
Because the ΦB interactions between KIM peptides and p38α are so highly conserved, we hypothesized that Val31 is not dispensable but instead has an important role in binding to p38α, which we examined again using NMR chemical shift mapping with the peptide HePTPKIM30, which lacks Val31, at four different protein-peptide ratios (Supplementary Fig. 4). The addition of Val31 resulted in a considerable increase in the number of residues that experienced chemical shift perturbations: from 23 (HePTPKIM30) to 35 (HePTPKIM31) (Fig. 3a,b). The localization of the majority of additional shifts to the ΦB binding pocket indicates that this is the binding site for Val31 (Fig. 4b). Remarkably, the addition of Val31 also resulted in a chemical shift perturbation of Leu108 in the hinge region, which connects the N and C lobes of p38α. This suggests that the domain closure mediated by the hinge in MAPKs upon KIM-peptide binding most likely requires occupation of the ΦB binding pocket. Unexpectedly, increases in the number of chemical shift perturbations for residues that contribute to the ΦA-2 and ΦA binding pockets, as well as the acidic binding pocket, were also observed for HePTPKIM31. Taken together, these data suggest that HePTPKIM31 interacts more tightly with p38α than HePTPKIM30. This was confirmed using ITC, which showed that the HePTPKIM31 binds p38α with a Kd of 5.2 ± 1.3 μM (Table 1 and Supplementary Fig. 3), whereas the Kd for the HePTPKIM30-p38α interaction was sufficiently weak that it could not be accurately determined. These results demonstrate that HePTPKIM31 occupies all three hydrophobic binding pockets (ΦA-2, ΦA and ΦB) in p38α and provide additional evidence that the mechanism of MAPK binding is determined by the KIM sequence and not the MAPK.
Crystallographic studies of MAPK–KIM peptide complexes have also shown that docking interactions alter the structure of the activation loop, resulting in a more ‘active-like’ conformation of the MAPK7–13,18. To determine whether the same changes occur in solution, we examined the chemical shifts of the p38α activation loop residues (172–183) in p38α alone and bound to either the HePTPKIM31 peptide or the HePTP protein. Remarkably, no chemical shift perturbations were observed for any of the peaks corresponding to activation-loop residues (10 of the 12 residues are assigned, including the Thr-Gly-Tyr motif). This demonstrates that, in solution, binding of HePTPKIM31 or HePTP to p38α does not change the local environment of activation-loop residues.
It is well known that the HePTPKIM is necessary for the formation of a stable p38α–HePTP complex15. However, it is unknown whether weak NMR-detectable interactions (mM) occur between p38α and HePTPCAT. Thus, we probed the interaction of p38α with HePTPCAT, which lacks residues 1–43 but includes HePTP α0, a secondary structural element that is unique to the KIM-PTP family among the classical PTPs38. Surprisingly, we detected absolutely no interactions between p38α and HePTPCAT (Figs. 2d and and3b).3b). This was unexpected because activated p38α is dephosphorylated by HePTP15, and thus, at some point during catalysis, the phosphorylated tyrosine of the p38α activation loop must bind the HePTP active site. Thus, all chemical shift perturbations observed for the p38α–HePTP complex are mediated by residues N-terminal to HePTPCAT, namely, the HePTPKIM31, and the residues immediately C-terminal to the HePTPKIM31 known as the kinase specificity sequence (KIS), namely, the HePTPKIS (ref. 16).
To confirm this hypothesis, we performed reverse NMR titration experiments with [2H,15N]HePTP and unlabeled p38α. The two-dimensional 1H,15N-TROSY spectrum of HePTP overlaps nearly perfectly with that of HePTPCAT, demonstrating that there is no interaction between the residues N-terminal to the catalytic domain (residues 15–43) and HePTPCAT (Supplementary Fig. 6a). This allowed us to transfer the previously published assignment of HePTPCAT (ref. 39) to that of HePTP. Additionally, the new peaks in the spectrum, owing to the additional 29 residues at the N terminus, were clustered at ~8.0 p.p.m., a chemical shift range typical for unstructured residues. We also detected chemical shift perturbations for residues 44–56 corresponding to helix α0, which indicate that the flexibility of residues 15–43 alters the chemical environment of residues 44–56.
To test the effect of p38α binding, the complex between [2H,15N]HePTP and unlabeled p38α was formed, and a two-dimensional 1H,15N-TROSY spectrum of the 78-kDa complex was collected (Supplementary Fig. 6b). All crosspeaks that clustered at ~8.0 p.p.m. in the free HePTP spectrum were perturbed in the spectrum of the p38α–HePTP complex, demonstrating that the flexible residues in HePTP, 15–56, bind strongly to p38α and adopt an ordered conformation. Thus, HePTP residues that mediate strong and specific p38α binding include not only the HePTPKIM31 peptide but also the HePTPKIS, which includes helix α0 (residues 31–56) and thus HePTP residues 15–56 (Supplementary Fig. 6c).
To validate these results, we generated the HePTPKIMKIS peptide, which corresponds to HePTP residues 15–56. The HePTPKIMKIS peptide was directly titrated into [2H,15N]p38α (it also copurifies with p38α using size-exclusion chromatography). As predicted by our earlier studies, the chemical shift perturbation mapping of the HePTPKIMKIS peptide bound to p38α almost completely recapitulated that observed in the p38α–HePTP complex, demonstrating that all of the interactions between p38α and HePTP are mediated by the HePTPKIMKIS peptide (Figs. 2e and 3a,b). We also performed a titration with 15N-labeled HePTPKIMKIS peptide and unlabeled p38α as an additional experiment to show that all residues of the HePTPKIMKIS peptide bind to p38α (Supplementary Fig. 7). Notably, the HePTPKIMKIS peptide interaction with p38α showed a few additional chemical shift perturbations. Correspondingly, it had an approximately three-fold increase in binding affinity over HePTP, as measured by ITC (Table 1 and Supplementary Fig. 3). The additional interaction is most likely due to nonpolar contacts between HePTPKIS hydrophobic residues (Ile40, Val43, Pro46 and Val49) and p38α residues (Leu217, Thr218, His228, Asn269, Asn272, Phe274, Ile275 and Ala277). These p38α residues showed larger chemical shift perturbations when bound to the HePTPKIMKIS peptide than when in complex with HePTP. Thus, both NMR and ITC data show that the presence of HePTPCAT slightly weakens the interactions observed for p38α residues that preferentially interact with the HePTPKIMKIS peptide.
SAXS provides information on the global size and shape of a protein in solution. The experimental scattering curve represents the average scattering of all conformations weighted by the number of molecules in each conformational state. We therefore used SAXS to determine the size and shape of the individual p38α and HePTP proteins and the p38α–HePTP complex in solution. SAXS profiles (I(q) versus q, with intensity I and scattering vector q) and pair- distribution functions are shown in Figure 5a,b (p38α–HePTP complex) and Supplementary Figure 8 (p38α and HePTP alone). The I0-to-concentration ratio was consistent for measurements of at least two concentrations, demonstrating no detectable aggregation and thus confirming the high quality of the samples. The experimental radius of gyration (Rg) value of 33.2 ± 0.5 Å for the p38α–HePTP complex was determined from Guinier analysis (n = 4; Supplementary Table 1). As expected, this was considerably larger than that for either protein alone (Rg = 23.7 ± 0.4 Å for p38α and 24.2 ± 0.2 Å for HePTP). Ab initio determination of the molecular envelope also converged on an Rg of 34.1 Å for the complex. These values for Rg are, respectively, 7.2 Å and 8.1 Å larger than what is predicted for a globular protein with the same number of amino acids40. Collectively, these results demonstrate that in the resting state the p38α–HePTP complex is highly extended, with p38α and HePTP associating in a head-to-head, head-to-tail or tail-to-tail manner.
To further our understanding of the p38α–HePTP complex, we developed a new computational procedure, ensemble refinement of SAXS (EROS)-NMR, which allowed us to determine the relative orientation of p38α and HePTP that provides the best fit to the experimental SAXS data. EROS-NMR includes the key features of EROS41: a coarse-grained model for protein binding is used to simulate the molecular assembly, which generates an initial ensemble of starting protein conformations. This ensemble is subsequently refined to improve the fit to the SAXS data. In addition, EROS-NMR accounts for NMR chemical shift data by incorporating a bias in the initial p38α–HePTP simulation. Namely, a nonspecific attractive potential was applied between the HePTPKIS linker (HePTP residues 32–56) and the p38α residues (Asn114, Asn115, Lys118, Lys121, Ser154, Gly219, Arg220) that fulfill three conditions: they show large chemical shift changes, they are surface exposed and they are not in the vicinity of the p38 KIM binding pocket. The bias potential was weak so that the HePTPKIS linker could bind and unbind repeatedly from p38α during the course of the simulation. The conformational space occupied by these rigid bodies was thoroughly sampled by simulation to identify conformations that best fit the SAXS data.
The best fit to the experimental data (χ2 = 1.3) was obtained for an ensemble of four conformations, which includes the single-best-fit model (SBFM) with the highest relative weight (47%), as well as three additional models (relative weights 23%, 16%, 14%; Fig. 5c,d and Supplementary Fig. 9). This fit is considerably better than that observed for the SBFM alone (χ2 = 1.5), for an ensemble that includes only two models (χ2 = 1.4), or for ensembles with five or more models, as the inclusion of more models only marginally improves χ2. All models in the four-model ensemble are globally similar, with p38α and HePTP associating in an end-to-end manner such that the HePTPCAT is localized to the bottom of the p38α C-terminal domain. They only differ in the position of the HePTPCAT domain. In fact, overlaying the models using p38α suggests that in solution the HePTPCAT fluctuates in a fan-like motion below p38α about the central SBFM conformation, with the weights of the additional models steadily decreasing as the distance from the SBFM increases (Supplementary Fig. 9). These changes in the position of HePTPCAT are accommodated by changes in the flexible HePTPKIS peptide, which adopts one of two classes of conformations: it either extends back toward the ATP binding pocket before bending down toward the bottom of the p38α C-terminal domain (the dominant SBFM conformation belongs to this class), or it bends down immediately after the last HePTPKIM residue. Remarkably, in none of the four structures was the active site of HePTP (Cys270) localized near the p38α activation loop, an interaction that must occur when HePTP dephosphorylates activated p38α. To quantify this important point, we also performed a full maximum-entropy refinement of 1,000 simulation structures41. The region of conformational space spanned by the resulting ensemble is consistent with that spanned by the smaller four-model ensemble, with, at most, 1% of structures in which the Cα atoms Tyr182 (p38α) and Cys270 (HePTP) are within 18 Å of each other (Supplementary Fig. 10). Notably, the corresponding structures extend the fan-like motion previously seen for the four-conformer ensemble toward the p38α activation loop (Supplementary Fig. 9). Thus, under resting-state conditions, HePTP and p38α adopt an elongated end-to-end complex that rarely (<~1%) samples the orientation required for dephosphorylation.
It is commonly accepted that KIM peptides are only partially responsible for recognition of effector proteins by MAPKs, as multiple studies have shown that additional interactions between MAPK effectors and substrates are required to maintain the exquisite specificity observed in MAPK signaling pathways16–18. Here, we have combined the molecular detail provided by NMR spectroscopy with the overall molecular organization afforded by SAXS to determine which residues outside the HePTPKIM31 direct and stabilize the resting-state conformation of the p38α–HePTP complex. One of the central findings of this study is that the HePTPKIM31 peptide alone is not able to recapitulate the interactions observed in the p38α–HePTP complex. Instead, multiple NMR titration experiments of p38α with HePTP constructs of varying lengths and domain architectures, coupled with ITC, reveal that both the HePTPKIM31 and HePTPKIS sequences are necessary and sufficient to mediate all binding interactions with p38α. To our knowledge, this is the first time that residues outside a KIM of a MAPK effector protein have been determined at the molecular level.
Using SAXS, we have also shown that in solution the interactions mediated by the HePTPKIMKIS lead to the formation of a p38α–HePTP complex in which p38α and HePTP interact in an elongated, end-to-end manner. This was unexpected because phosphorylated p38α is dephosphorylated by HePTP, which requires the formation of a more globular complex. However, it is clear from our SAXS studies that under resting-state conditions the p38α–HePTP complex is not globular but instead is highly elongated. This result differs considerably from that of a recently published crosslinking study42 in which a resting-state PTP-SL–ERK2 complex was crosslinked and the structure of the complex was modeled using the crosslinking distances as constraints. Jointly, these results demonstrate that the overall orientation of MAPK–KIM-PTP resting-state complexes may differ depending on the MAPK, the KIM-PTP or both.
To generate a structural model for the p38α–HePTP complex in solution, we developed a new SAXS refinement procedure, EROS-NMR, which incorporates NMR chemical shift data into a recently developed SAXS refinement procedure. Incorporation of the NMR chemical shift data ensures that the models are consistent with not only the SAXS data, which define overall shape, but also the NMR chemical shift data, which define molecular interactions that direct the organization of the complex. As expected on the basis of multiple failed attempts to crystallize the resting-state complex, the SAXS refinements show that the complex is dynamic, as four models provide a better fit to the SAXS data than a single model. The models in the four-model ensemble have similar overall conformations, which differ primarily in the relative orientation of the HePTPCAT below the p38α C-terminal domain. The HePTPKIS conformation is variable, but adopts highly similar conformations as it projects downward from the p38α KIM binding pocket. The largest differences are observed in the C-terminal portion of the HePTPKIS, which adopts an extended conformation and interacts differentially with p38α in the four ensemble models of the p38α–HePTP complex. This shows that the N-terminal HePTPKIS residues interact directly with p38α, whereas the C-terminal HePTPKIS residues transiently bind p38α, allowing additional conformations to be sampled about the central dominant conformation (the single-best-fit conformation). This model is also consistent with a slight decrease in binding affinity between HePTPKIMKIS and HePTP with p38α, as measured by ITC, showing that this flexibility slightly lowers the overall strength of binding.
Taken together, these results provide fundamental new insights into the molecular basis of MAPK regulation and specificity. As additional similar studies are completed, the subset of residues outside the KIM binding pocket that are specific for different MAPK–MAPK regulatory protein complexes will be determined and, in turn, will allow for the development of new drug therapies that target these specific complexes.
The HePTPKIM31 peptide (VRLQERRGSNVALMLDV) was synthesized and HPLC purified (> 98% purity; Biosynthesis); the HePTPKIM30 peptide (RLQERRGSNVALMLD) was synthesized and HPLC purified (>98% purity; gift from J. Offer, University of Cambridge). Expression and purification of HePTP and HePTPCAT were carried out as previously described14,38 except that the final size-exclusion chromatography (SEC) step was carried out using p38α protein buffer (50 mM HEPES, pH 6.8, 150 mM NaCl, 5 mM dithiothreitol). p38α and the HePTPKIMKISC42S peptide were subcloned into a derivative of the pET28 vector that includes an N-terminal His6 tag and a TEV protease–recognition sequence43. Unlabeled proteins were expressed in BL21-(DE3) RIL Escherichia coli cells (Stratagene) in LB medium. [2H,15N]p38α and HePTP were expressed in M9 medium supplemented with 15NH4Cl (1 g l−1) in 99% D2O, and [2H,1315N]p38α was expressed in M9 medium supplemented with 15NH4Cl (1 g l−1) and [13C]d-d7-glucose (4 g l−1) in 99% D2O. p38α and the HePTPKIMKIS were purified using standard protocols (Supplementary Methods). For p38α–HePTP complex formation, p38α was incubated with HePTP (1:1 molar ratio) on ice for 45 min and purified using SEC equilibrated in p38α protein buffer. The p38α–HePTP complex elutes at the position expected for the p38α–HePTP heterodimer. The [2H,15N]p38α–HePTP complex and the p38α–[2H,15N]HePTP complex were concentrated to 0.2 mM and stored on ice (<8 h) until NMR data acquisition.
All NMR experiments were performed at 298 K on a Bruker Advance II 800 MHz spectrometer equipped with a TCI HCN z-gradient cryoprobe. NMR samples were prepared in protein buffer in 90%/10% (v/v) H2O/D2O. The NMR spectra were processed with Topspin 2.1 (Bruker) and analyzed using the CARA software package (http://www.nmr.ch/doku.php). Purified [2H,13C,15N]p38α was concentrated to 0.58 mM and stored on ice (<8 h) until data acquisition. The sequence-specific backbone assignment was obtained from TROSY versions of the following experiments: two-dimensional 1H,15N-TROSY, three-dimensional HNCA, three-dimensional HNCACB, three-dimensional HN(CO)CA, three-dimensional HN(CO)CACB and a three-dimensional 15N-edited 1H,1H-NOESY spectrum. The assignment of residues around the ATP binding pocket were confirmed by titration of 2′,3′-(2,2,5,5-tetramethyl-3-pyrroline-1-oxyl-3-carboxylic acid ester) adenosine triphosphate (SL-ATP, a generous gift from P. Vogel, Southern Methodist University)34 into [2H,13C,15N]p38α; final ratios were 0.16 mM p38α/0.32 mM SL-ATP (1:2) and 0.16 mM p38α/0.64 mM SL-ATP (1:4). Two-dimensional 1H,15N-TROSY spectra were recorded. All quantitative spin-label experiments were processed with NMRPipe44, and peak integrals were determined using CARA (http://www.nmr.ch/doku.php).
Both HePTPKIM31 and HePTPKIM30 peptides were solubilized in protein buffer. Experiments were performed at ratios of 1:1, 1:3, 1:5 and 1:8 excess peptide, which were added to [2H,15N]p38α, and a two-dimensional 1H,15N-TROSY spectrum was recorded (100 μM p38α; 1:1 experiment was also recorded at 250 μM p38α). A two-fold excess of unlabeled HePTPCAT was added to [2H,15N]p38α (0.2 mM p38α/0.4 mM HePTPCAT), and a two-dimensional 1H,15N-TROSY spectrum was recorded. An equal molar ratio of unlabeled HePTPKIMKIS was added to [2H,15N]p38α (94 μM p38α/94 μM HePTPKIMKIS), and a two-dimensional 1H,15N-TROSY spectrum was recorded. The [2H,15N] p38α–unlabeled HePTP complex was purified and concentrated to 0.2 mM, and a two-dimensional 1H,15N-TROSY spectrum was recorded. p38α–HePTPKIM30 titrations in ratios of 1:1, 1:3, 1:5 and 1:8 were used to follow chemical shift changes in p38α upon binding, as nearly all peaks showed fast exchange and thus could be easily followed. This data was used for the analysis of the 1:1-, 1:3-, 1:5- and 1:8-ratio p38α–HePTPKIM31 titrations and all other interactions (HePTPKIMKIS and HePTP), allowing for high confidence of peak assignments of p38α in the bound state. All chemical shift differences (Δδ) were calculated using the following equation:
We adapted the EROS method for the refinement of SAXS data41 to incorporate NMR data. EROS was previously used to determine the structure of protein kinase C in solution45, and a related method has been used to refine SAXS data of Hck kinase46. In the simulations of the p38α–HePTP complex, a nonspecific, attractive potential was applied between the KIS linker (HePTP residues 32–56) and the p38α residues (Asn114, Asn115, Lys118, Lys121, Ser154, Gly219, Arg220) that fulfill three conditions: they show large chemical shift changes, they are surface exposed and they are not in the vicinity of the p38 KIM binding pocket. The bias potential increased the well depth of the amino acid interactions by only 1 kBT, such that the KIS linker could easily bind and unbind from p38α during the course of the simulation. The scattering intensities Ik(q) were calculated for all simulated structures k, and the ensemble intensity was defined as
where wk is the relative weight assigned to structure k in an ensemble of n structures. The difference between computed and measured intensities was quantified by
where N is the number of data points and a and b are fitting parameters obtained from conditions χ2/a = χ2/b = 0. The parameter a sets the intensity scale and the parameter b << Iexp(q=0) is a small background correction. In the SAXS profile calculations, the hydration-shell electron-density contrast ρ was also varied, and best fits were obtained for ρ = 0.005e/Å3. To fit the experimental data with a small number of simulation structures, we defined a function G = χ2 + μn, which was minimized numerically by varying the number of structures n with nonzero weights and their weights wk under the constraint Σkwk = 1. Here μ > 0 is a control parameter that captures our preference for having a small number n of structures in the ensemble, analogous to a chemical potential. As μ is decreased, n increases and the value of χ2 decreases upon minimization of G. Beyond n = 4, the quality of the fit does not improve substantially (with χ2 = 1.3, 1.4 and 1.5 for n = 4, 2 and 1, respectively). For reference, we also performed a full maximum-entropy refinement over the entire ensemble41, which produced an ensemble consistent with the region of conformation space spanned by the set of n = 4 structures.
Methodological details of additional p38α–HePTP interaction studies, isothermal titration calorimetry and SAXS are described in Supplementary Methods.
The authors thank L. Yang and M. Allaire (National Synchrotron Light Source, NSLS) for their support at NSLS beamline X9. The authors thank P. Vogel (Southern Methodist University) for providing SL-ATP for the titration studies. This research was supported by grant RSG-08-067-01-LIB from the American Cancer Society to R.P. B.R. was supported by a Marie Curie International Outgoing Fellowship within the Seventh European Community Framework Program. B.R. and G.H. were supported by the Intramural Research Program of the National Institute of Diabetes and Digestive and Kidney Diseases, US National Institutes of Health (NIH). Use of the NSLS at Brookhaven National Laboratory was supported by the US Department of Energy, Office of Science, Office of Basic Energy Sciences under contract no. DE-AC02-98CH10886. 800 MHz NMR data were recorded at Brandeis University and were purchased with support from NIH S10-RR017269. Molecular simulations were performed on the Biowulf computing cluster at NIH.
Author contributions D.M.F. and D.K. cloned and purified all proteins. D.M.F. and W.P. performed and analyzed all NMR measurements. D.M.F., W.P. and R.P. performed and analyzed the SAXS measurements. D.K. performed all ITC measurements. R.P. and W.P. helped design the biochemical and NMR experiments. B.R. and G.H. performed all EROS and EROS-NMR calculations. R.P. and W.P. wrote the paper and all authors discussed the results and commented on the manuscript.
Competing financial interests The authors declare no competing financial interests.
Additional information Supplementary information is available online at http://www.nature.com/naturechemicalbiology/.
Accession codes. Protein Data Bank: the previously published structures of p38 alone, in the presence of SB203580 and in complex with BIRB796 are deposited under accession codes 1P38, 1A9U and 1KV2, respectively. Biological Magnetic Resonance Bank: chemical shifts recorded in this study are deposited under accession number 17471.