Local destabilization, rigid body, and fuzzy docking facilitate the phosphorylation of the transcription factor Ets-1 by the mitogen-activated protein kinase ERK2

Significance

This study provides experimental support for the mechanism of proximity-mediated catalysis by a mitogen-activated protein (MAP) kinase. It describes how the transcription factor Ets-1 uses the cumulative effects of two suboptimal docking interactions, rather than a single canonical docking interaction, to enable a unique bipartite mechanism of recognition of the MAP kinase ERK2. This mode of interaction between Ets-1 and ERK2 facilitates the formation of a highly productive complex that not only induces the proximity of Ets-1 phospho-acceptor (T38) to the ERK2 active site, but does so in optimal fashion, thereby promoting efficient phospho-transfer.

Abstract

Mitogen-activated protein (MAP) kinase substrates are believed to require consensus docking motifs (D-site, F-site) to engage and facilitate efficient site-specific phosphorylation at specific serine/threonine-proline sequences by their cognate kinases. In contrast to other MAP kinase substrates, the transcription factor Ets-1 has no canonical docking motifs, yet it is efficiently phosphorylated by the MAP kinase ERK2 at a consensus threonine site (T38). Using NMR methodology, we demonstrate that this phosphorylation is enabled by a unique bipartite mode of ERK2 engagement by Ets-1 and involves two suboptimal noncanonical docking interactions instead of a single canonical docking motif. The N terminus of Ets-1 interacts with a part of the ERK2 D-recruitment site that normally accommodates the hydrophobic sidechains of a canonical D-site, retaining a significant degree of disorder in its ERK2-bound state. In contrast, the C-terminal region of Ets-1, including its Pointed (PNT) domain, engages in a largely rigid body interaction with a section of the ERK2 F-recruitment site through a binding mode that deviates significantly from that of a canonical F-site. This latter interaction is notable for the destabilization of a flexible helix that bridges the phospho-acceptor site to the rigid PNT domain. These two spatially distinct, individually weak docking interactions facilitate the highly specific recognition of ERK2 by Ets-1, and enable the optimal localization of its dynamic phospho-acceptor T38 at the kinase active site to enable efficient phosphorylation.

The mitogen activated protein (MAP) kinase ERK2 (extracellular signal-regulated kinase 2) lies at the terminus of a three-tiered phosphorylation-based response to a wide range of extracellular cues that include cytokines, hormones, and growth factors (1–4). ERK2 phosphorylates numerous substrates in both the nucleus and the cytoplasm, including a variety of transcription factors, regulatory kinases, phosphatases, proteins of the nuclear pore complex, and cytoskeleton proteins, among others (3, 4). ERK2-mediated phosphorylation occurs on specific serine/threonine residues that immediately precede a proline; however, the (S/T)P motif alone does not provide sufficient substrate affinity or selectivity to distinguish it from other proline-directed kinases, such as CDK2 (5). To attain high specificity toward native substrates, ERK2 and other MAP kinases use one of two so-called “docking sites,” the D-recruitment site (DRS) or the F-recruitment site (FRS) (SI Appendix, Fig. S1), which are spatially distinct from the catalytic machinery (6–10). How these docking interactions support specific phosphorylation events within MAP kinase substrates is largely unknown.

The first of these docking regions, the DRS, is located behind the ATP-binding pocket and recognizes substrates (e.g., the transcription factor Elk-1) that contain a D-site consensus sequence (R/K)_2–3-(X)_2–6-Φ_A-X-Φ_B (where Φ_A/B are hydrophobic residues). The Φ_A/B moieties, which are sometimes flanked by a third hydrophobic residue (Φ_L), insert into one of three hydrophobic grooves—ϕ_A, ϕ_B, or ϕ_L (11)—which we collectively term ϕ_hyd (12). The positively charged segment of the D-site interacts with a complementary acidic patch, termed ϕ_chg (also known as the common docking motif), on the ERK2 surface. The second docking region, the FRS, which forms fully only on kinase activation (8), is located just below the activation loop and binds substrates bearing a consensus F-site sequence (F-X-F-P); e.g., the transcription factors Lin-1, Sap-1, and Elk-1 (13). Recent biochemical and structural studies have suggested that the terminal proline of the F-site is dispensable for FRS/F-site interactions (14, 15).

It is becoming increasingly evident that MAP kinases also can use noncanonical docking interactions to recognize their binding partners. An example of such an association was revealed in nuclear magnetic resonance (NMR) studies (12) of the interactions between ERK2 and an inhibitor, PEA-15 (protein enriched in astrocytes) (16). That PEA-15 indeed engages both the ERK2 DRS and FRS despite lacking canonical D-site and F-site sequences was later confirmed by a crystal structure of the ERK2•PEA-15 complex (17). Indeed, noncanonical interactions have been predicted to drive the highly efficient phosphorylation of the transcription factor Ets-1(18–20), a bona fide ERK2 substrate (21, 22). How such noncanonical interactions drive substrate phosphorylation in terms of structure, dynamics, and mechanism has not yet been established.

Ets-1 (SI Appendix, Fig. S2), which belongs to the ETS (from E twenty-six avian erythroblastosis virus) domain-containing family of transcription factors (23), is a ∼440-residue protein that contains a Pointed (PNT) domain, a transactivation domain, and a DNA-binding ETS domain, as well as an N-terminal region that carries a canonical TP (³⁸TP³⁹) sequence targeted by ERK2. Phosphorylation on T38 drives the transactivation of Ets-1 (24) through the enhanced recruitment of the coactivator CBP/p300 (25). It has been demonstrated that the first 138 residues of Ets-1 (EtsΔ138) are sufficient to bind ERK2 and enable phosphorylation on T38 (22). Extensive biochemical studies and preliminary NMR analyses have established that, like PEA-15, Ets-1 also engages both the DRS and the FRS of ERK2 despite the absence of canonical D-site or F-site sequences (12, 19, 26). However, the precise details of these interactions and how they contribute toward the highly efficient phosphorylation of Ets-1 by ERK2 (27) are not known. Here, using a variety of solution NMR methods and supporting biochemical studies, we provide a structural basis for the recognition of Ets-1 by ERK2 and the role of noncanonical docking interactions in facilitating efficient Ets-1 phosphorylation.

Results

EtsΔ138 Engages Both Inactive and Active ERK2 in a Similar Fashion.

Isothermal titration calorimetry (ITC) measurements demonstrate that under NMR conditions, EtsΔ138 binds inactive ERK2 (referred to as ERK2) and active ERK2 (ppERK2) with similar affinity (K_D = 30.0 ± 7.4 μM for ERK2 and K_D = 28.8 ± 4.3 μM for ppERK2; SI Appendix, Fig. S3A). These interactions enable the efficient phosphorylation of EtsΔ138 on T38 by ppERK2 (Table 1 and SI Appendix, Fig. S4). ERK2 and ppERK2 induce a large number of chemical shift perturbations (CSPs) at several regions of EtsΔ138 (Fig. 1). These include the N terminus, helices H0 and H1, and the latter half of the PNT domain (formed by helices H2–H5). Fits of the chemical shift titration data to a one-site binding model yield K_D values consistent with the ITC results (SI Appendix, Fig. S3B). The CSPs measured for EtsΔ138 in the inactive (EtsΔ138•ERK2) and the active (EtsΔ138•ppERK2) complexes are correlated (Pearson coefficient = 0.86; SI Appendix, Fig. S3C), suggesting that the overall binding modes are very similar in the two cases. Therefore, here we focus on the inactive complex when describing the general structural features of the EtsΔ138/ERK2 interactions.

Table 1.

Kinetic parameters for the phosphorylation of EtsΔ138 positioning mutants by ppERK2

Sequence	Threonine position	kcat, s−1	KM, µM	kcat/KM, µM−1s−1
32ADVPLLTPSSKEMMSQ47	-38TP39-	14 ± 1.6	23 ± 5	0.6
32ADVPLLTASSKEMMSQ47	-38TA39-	3.5 ± 0.24	63 ± 5.8	0.06
32ADVPLTPASSKEMMSQ47	-37TP38-	8 ± 0.1	7 ± 0.4	1.14
32ADVPTPAASSKEMMSQ47	-36TP37-	8.4 ± 0.3	21 ± 3	0.4
32ADVPLLATPSKEMMSQ47	-39TP40-	4 ± 0.25	34 ± 7	0.12
32ADVPLLAATPKEMMSQ47	-40TP41-	5.5 ± 0.25	29 ± 4.5	0.19
32ADVPLLAASTPEMMSQ47	-41TP42-	3 ± 0.2	37 ± 8	0.08

Parameters are the best fits to Eq. S12 in SI Appendix. The first row represents wild-type EtsΔ138. The second row (italicized) represents a mutant in which the Pro at the P+1 position has been replaced by Ala. The TP motif is positioned further away from the PNT domain compared with wild-type for rows 3 and 4 and closer to the PNT domain for rows 5–7.

Fig. 1.

Interactions of EtsΔ138 with active and inactive ERK2. Chemical shift perturbations are shown for EtsΔ138 in the presence of ppERK2 (Left) and inactive ERK2 (Right). The bound-state population of EtsΔ138 is 71% for ppERK2 and 76% for ERK2. The CSPs are shown in orange bars (except for T38, which is shown as a blue bar and labeled with a larger font). Residues for which the resonances are broadened to below the threshold of detection are depicted by red bars. Key structural regions are indicated.

The N-Terminal Region of EtsΔ138 Remains Disordered on Interaction with ERK2.

The N terminus of EtsΔ138 (K2-S40; we refer to EtsΔ138 residues using single-letter codes, and use three-letter codes for ERK2 residues) is highly disordered in the free state, with steady-state {¹H}-¹⁵N nuclear Overhauser effect (NOE) values of −0.28 ± 0.23 (−0.33 ± 0.26 for the K2–F24 region) (SI Appendix, Fig. S5A). A significant degree of this disorder remains on interaction with ERK2 (65% of EtsΔ138 in the EtsΔ138•ERK2 complex for the data in SI Appendix, Fig. S5B), with {¹H}-¹⁵N NOE values of −0.02 ± 0.18 (−0.02 ± 0.21 for the K2–F24 region). Using these values and the measured affinity of EtsΔ138 toward ERK2, a value of 0.07 ± 0.22 (0.09 ± 0.19 for the K2–F24 region) can be estimated for this region when fully bound to ERK2. In contrast, the C-terminal region (CTR; S54–V137), which includes H1 and the PNT domain, is well ordered, with {¹H}-¹⁵N NOE values of 0.76 ± 0.15 and 0.71 ± 0.28 in the free and the extrapolated fully bound states, respectively.

To further characterize the degree of ordering of the N terminus of EtsΔ138 on formation of the EtsΔ138•ERK2 complex, we measured the difference between the values of the spectral density function near the ¹H frequency, J(0.87ω_H), for the apo and extrapolated fully bound states (SI Appendix, Materials and Methods). J(0.87ω_H) is highly sensitive to dynamics on the fast, sub-nanosecond timescale and is independent of the effects of chemical exchange. It is also largely insensitive to the overall correlation time, allowing direct comparison of the fast dynamics between the free and fully bound states of EtsΔ138. Overall, there was a greater degree of ordering [ΔJ(0.87ω_H) > 0] at the N terminus of EtsΔ138 than for the CTR, with the fast dynamics remaining largely unchanged for the latter region on engaging ERK2 (Fig. 2). Whereas ΔJ(0.87ω_H) values indicate that the N terminus of EtsΔ138 shows increased ordering on engaging ERK2, the magnitudes of the {¹H}-¹⁵N NOEs (close to 0 on average, as mentioned above) confirm that it remains largely disordered in the ERK2-bound state.

Fig. 2.

Change in dynamics of EtsΔ138 on the fast time scale on binding ERK2. The changes in spectral density function values of EtsΔ138 near the ¹H frequency, ΔJ(0.87ω_H), on binding ERK2 are plotted against its sequence. Key residues that become significantly more disordered [ΔJ(0.87ω_H) <0] on engaging ERK2 are labeled.

Helix H0 of EtsΔ138 Becomes More Dynamic on Interaction with ERK2.

In contrast to most of the EtsΔ138 N terminus, several residues in H0 become more disordered on binding ERK2 (Fig. 2). These include E43 at the N terminus of H0 [ΔJ(0.87ω_H) = −10.4 ± 1.3 ps/rad], Q47 (−22.8 ± 3.3), and A48 (−20.5 ± 1.4), which lie toward the middle. S54, the first residue of H1, also exhibits increased flexibility on complex formation (−32.4 ± 2.5). Whereas H0 is also dynamic in the free state (SI Appendix, Fig. S5A; the E43–L49 segment displays {¹H}-¹⁵N NOEs <0.5), specific regions become further disordered on engaging ERK2. H0 is located immediately C-terminal to the phospho-acceptor T38 and N-terminal to the relatively rigid CTR. As described below, we suspect that this increased disorder in H0 plays a role in optimally positioning the phospho-acceptor (T38) at the ERK2 active site, thereby increasing the efficiency of phosphorylation.

The N Terminus of EtsΔ138 Engages the ERK2 DRS.

We previously demonstrated that the phosphorylation of EtsΔ138 by ppERK2 can be inhibited by peptides containing either a canonical D-site (Lig-D, which engages the DRS) or a canonical F-site (Lig-F, which binds the FRS) sequence (26). However, Lig-F, but not Lig-D, is able to inhibit the phosphorylation of a construct of EtsΔ138 that was missing 23 N-terminal residues (Δ24EtsΔ138). This observation, together with the CSPs mentioned above (and discussed in greater detail below), suggests that the N terminus of EtsΔ138 (residues 1–23) at least partially engages the ERK2 DRS.

Widespread CSPs are observed at the N terminus of EtsΔ138 on interaction with ERK2, with the largest of these localized in the fragment encompassing residues K8 and I14 (Fig. 1B). In addition, a number of intermolecular NOEs (Fig. 3A) can be detected between the methyl groups of I13 and I14 on EtsΔ138, Leu110, and Leu155 on ERK2. Leu110 and Leu155 belong to the portion (ϕ_hyd) of the ERK2 DRS involved in recognizing the hydrophobic part of the D-site sequence (Φ_A-X-Φ_B; SI Appendix, Fig. S1). This finding is consistent with our previous NMR analysis that showed a large number of CSPs localized on the ϕ_hyd portion of the ERK2 DRS in the presence of EtsΔ138 (28) (SI Appendix, Fig. S6). Furthermore, in the presence of ERK2, the ¹³Cα resonances for EtsΔ138 residues T10 and L11 shift upfield by 0.28 and 0.39 ppm (for the 76%-bound state), respectively, and the corresponding ¹³Cβ resonances shift downfield by 0.34 and 0.85 ppm, respectively (SI Appendix, Fig. S7). The secondary shifts (Δ¹³Cα−Δ¹³Cβ) of −0.62 ppm and −1.24 ppm suggest the formation of a more extended structure in this region of EtsΔ138, as opposed to a random coil, on binding of ERK2. The ¹³Cα and ¹³Cβ resonances of T12 and I13 are broadened beyond the threshold of detection (SI Appendix, Fig. S7).

Fig. 3.

(A) Interactions between select methyl groups of EtsΔ138 and ERK2. Examples of intermolecular NOE cross-peaks (labeled in red) observed in a 3D ¹³C, ¹H NOESY-HMQC experiment on an equimolar mixture of inactive LV, U-[¹⁵N, ²H]-labeled ERK2 and IT, U-[¹⁵N, ²H]-labeled EtsΔ138. The leucine methyl groups have been stereospecifically assigned. (B) Chemical shift perturbations for Ile (δ1), Leu, Val, and Thr methyl groups of EtsΔ138 in the presence of inactive ERK2. Significant perturbations (>10% trimmed mean plus three times the corresponding SD = 0.26 ppm) are denoted by the dashed line. CSPs are indicated by the pink bars, and residues for which the resonances are broadened to below the threshold of detection are depicted by blue bars.

Analysis of the Ile, Leu, Val, and Thr methyl resonances (δ1 only for Ile) revealed that the largest CSP is observed for I13 (0.44 ppm), followed by L11 (0.18 ppm; only one δ position is sufficiently resolved to be reliably analyzed), T10, and I14 (both 0.13 ppm) (Fig. 3B). In contrast, relatively modest CSPs (<0.07 ppm) are seen for the T12 and T16 methyl groups. Taken together, these findings suggest that EtsΔ138 interacts with the ERK2 DRS through the sidechains of the T10–I14 segment. P9 also plays a role in this interaction, as described below. The importance of L11, I13, and I14 are in agreement with the findings of Lee et al. (26), who reported that mutations in L11 (L11D), I13, and I14 (I13A/I14A, I13D/114D) lead to modest (twofold to threefold) increases in K_M values for EtsΔ138 phosphorylation by ppERK2. Thus, the P9–I14 region of EtsΔ138 likely mimics, in part, the hydrophobic segment of a conventional D-site sequence (as will be evident from the description below); we refer to this segment as a D_hyd-site hereinafter.

Although the interaction between the N terminus of EtsΔ138 and the ERK2 DRS occurs through the P9–I14 region of the former, it should be noted that some significant CSPs (>0.1 ppm) can be seen for both amide and methyl resonances (Figs. 1 and 3B) in other N-terminal regions. These include the A4-D6 segment, K8 (see below), K18 and F24 (amide), and L23 (methyl). This raises the possibility that these residues also interact, albeit transiently, with the ERK2 surface, perhaps as a result of the persistent disorder in the ERK2-bound state. However, point mutations of several of these residues (e.g., K18A, L23A) have no significant effect (29) on the K_M value, suggesting that these contacts individually do not significantly stabilize the EtsΔ138•ERK2 complex.

Spectral perturbations on EtsΔ138 (for the P9–I14 segment) in the presence of ERK2, and vice versa, were used to generate ambiguous interaction restraints (AIRs) that were combined with unambiguous methyl-based intramolecular and intermolecular distance restraints between EtsΔ138 (Ile, Thr) and ERK2 (Leu, Val). These AIRs were used in docking calculations using HADDOCK (30, 31) to generate the structure of the EtsΔ138 D_hyd-site bound to the ERK2 DRS (Fig. 4A).

Fig. 4.

Docking interactions between EtsΔ138 and ERK2. (A) The best HADDOCK cluster for the D_hyd-site of EtsΔ138 (residues 9–14; in green) docked against the ERK2 DRS. The ERK2 surface is shown only for the lowest-energy solution, in gray. The key hydrophobic pockets, ϕ_A, ϕ_B, and ϕ_L, are in purple, pink, and blue, respectively. Sidechains of EtsΔ138 that make important contacts with ERK2 are labeled and shown in stick representation. ERK2 residues making contacts with EtsΔ138 are also labeled (in yellow) and underscored. (B) The best HADDOCK cluster for a C-terminal fragment of EtsΔ138 (green, ribbon representation; residues 40–138) docked against the ERK2 FRS (gray surface shown only for the lowest-energy solution). The equivalent regions of ppERK2 that accommodate the two Phe residues of a canonical F-site sequence are labeled ϕ₁ and ϕ₂ and colored in blue and pink, respectively [defined according to Lee et al. (8)]. ERK2 residues that form each of these pockets are labeled in the same color as the corresponding pocket. EtsΔ138 sidechains that contact ERK2 are labeled and are shown in stick representation. ERK2 residues making contacts with the EtsΔ138 are also labeled (in yellow if not part of the canonical ϕ₁ and ϕ₂ pockets) and underscored.

In the lowest-energy ensemble, the T10–I14 segment of EtsΔ138 binds ERK2 in a largely extended conformation (as expected based on the chemical shift analysis described above) in a parallel (N→C) orientation and occupies an area between loop 11 and loop 8 on ERK2, a region targeted by a number of canonical D-site peptides (Fig. 4A) (10). I13 interacts with the ϕ_A hydrophobic pocket by assuming an orientation similar to that of a number of other DRS ligands (SI Appendix, Fig. S8). Residues L11 and I14 are characterized by less canonical poses, with the former binding between the ϕ_L and the ϕ_A pockets and the latter partially occupying a hydrophobic pocket comprising ERK2 residues Leu113, Gln117, and Leu119. The δ1 position of Leu113 displays a large CSP, and the δ2 position of Leu119 broadens out (SI Appendix, Fig. S9) confirming the engagement of this novel pocket on ERK2. The ϕ_B pocket remains unoccupied.

Measurements of sidechain dynamics of the methyl groups of I13 and I14 in the EtsΔ138•ERK2 complex suggest a greater spatial restriction of I13 compared with I14, and indeed a far greater degree of burial of the former (SI Appendix, Fig. S10) compared with the latter. This is also reflected in the ∼2.5-fold difference in methyl CSPs, as well as in the absence of significant Cα/Cβ shift changes for I14 (SI Appendix, Fig. S7). The occupancy of the noncanonical hydrophobic pocket by I14 is also enabled by the spatial constraints imposed by the burial of I13 in ϕ_A, as well as by the fact that EtsΔ138 carries a Φ-Φ (I13-I14) motif rather than a canonical Φ-X-Φ motif.

Interestingly, the HADDOCK models also show interactions between the ERK2 ϕ_L pocket and P9 on EtsΔ138 (for which the Cα and Cβ resonances are broadened out in the presence of ERK2). Replacement of His123 on ERK2, which contacts P9, by Ala leads to an approximate fivefold increase in the K_M value (29). K8 at the N terminus of EtsΔ138 displays a large amide CSP in the presence of ERK2 (Δδ = 0.24 ppm). Inclusion of this residue in the HADDOCK calculations leads to less well-defined separation of the structure clusters (although the overall conformations are not substantially altered). Nevertheless, these clusters indicate that the K8 sidechain can assume an orientation that enables electrostatic interactions with Asp316 on ERK2. Of note, the backbone amide resonance of Asp316 is severely broadened in the presence of EtsΔ138 (SI Appendix, Fig. S6, Top Left), and is the only ϕ_chg residue of ERK2 substantially affected. We suspect that this interaction, although present, occurs only transiently. This scenario is supported by the fact that cysteine footprinting studies show no significant protection of Asp316 in the presence of EtsΔ138 and only a ∼1.7-fold increase in K_M for the phosphorylation of EtsΔ138 by an ERK2 construct carrying a Asp316Ala/Asp319Ala double mutation (32).

The C-Terminal Region of EtsΔ138 Engages the ERK2 FRS.

The finding that Lig-F, but not Lig-D, can inhibit the phosphorylation of Δ24EtsΔ138, together with the CSPs seen for the CTR of EtsΔ138 (Fig. 1) in the presence of ERK2 and the corresponding CSPs at the ERK2 FRS (12) (SI Appendix, Fig. S6) in the presence of EtsΔ138, indicates that the EtsΔ138 CTR engages the ERK2 FRS. Clearly, because the FRS is not fully formed in inactive ERK2 (8), the EtsΔ138-binding mode is expected to be different from that adopted by canonical F-site sequences, e.g., one derived from Elk-1 that was structurally characterized previously (15).

A number of EtsΔ138 amide resonances are broadened beyond the threshold of detection in the presence of both ERK2 and ppERK2 (Fig. 1), indicating the presence of exchange on the microsecond to millisecond timescale. These residues are localized mostly at the C-terminal end of H0 and on H5. Individual residues at the N-terminal ends of H0 and H1, on the H4-H5 loop, and on H4 are broadened as well. Of these residues, F120, which lies on the H4-H5 loop, has been predicted to be critical for ERK2 recognition. An F120A mutation leads to a ∼20-fold increase in the K_M value for the phosphorylation of EtsΔ138 (22, 29). In addition to line-broadening, significant CSPs can be seen in the first half of H1; at the C-terminal end of H2; on the H2-H3, H3-H4, and H4-H5 loops; and on most of H5 (Fig. 1). A structural analysis of these amide CSPs and those of the Ile, Val, Leu, and Thr methyl resonances (Fig. 3B) suggests that all of these perturbations (CSPs and quenching) do not result from direct contacts between the PNT domain and ERK2, but rather are the result, at least in part, of some structural rearrangement within the PNT domain induced by the docking event. Several EtsΔ138 residues that show spectral perturbations on H1, H4, and H5 are not substantially solvent-exposed in the free state, but are coupled to one another through the structural core (SI Appendix, Fig. S11). This suggests that a slight rearrangement in the relative orientation of H1 and H5 induced by binding is likely responsible for some of the perturbations mentioned above. The observed perturbations for several buried methyl groups (e.g., L125 and L129, which show significant CSPs; I124, which is broadened out) of EtsΔ138 support this hypothesis (SI Appendix, Fig. S11). A comprehensive analysis of the chemical shifts of EtsΔ138 in the presence of ERK2 using TALOS+ (33) indicates that the secondary structure elements for the PNT domain are unchanged from free EtsΔ138. Taken together, these data support a binding mode in which the N terminus of H1, the H4-H5 loop, and H5 are involved in ERK2 recognition, with the latter two regions playing a more dominant role.

We previously noted that extensive line-broadening in part of the ERK2 FRS and the activation loop leads to missing amide cross-peaks for this region (12, 26). We had also shown that the presence of a canonical F-site ligand partially quenches these motions at the FRS and at the activation loop, leading to the reappearance of several amide resonances corresponding to this region in ppERK2 (15) (SI Appendix, Fig. S6, Middle). In contrast, no additional backbone resonances appear in the presence of EtsΔ138, suggesting that the EtsΔ138 PNT domain only partially engages the ERK2 FRS and fails to stabilize the dynamic parts of this region in ERK2 (and ppERK2).

Methyl CSPs (SI Appendix, Fig. S9) for the ERK2 FRS localize mainly on helix αG and generally are of smaller magnitude than those seen at the DRS, as in case of the backbone resonances (12). Moderate shifts are seen for Leu242 and at least one of the methyl positions of Leu235, whereas the methyl groups of Leu232 and Leu239 are broadened out. Interestingly, whereas Leu232 and Leu235 form part of the FRS pocket (ϕ₂) that accommodates the second phenylalanine of the F-X-F motif (15), Leu198, which is also part of the same pocket, is not affected, confirming the partial occupancy of this pocket. Although the activation loop Leu/Val residues (Leu168, Val171, Leu182, and Val186) could not be assigned (SI Appendix, Fig. S9), no additional resonances appear in the presence of EtsΔ138, reinforcing the idea (as discussed above) that EtsΔ138 fails to stabilize the activation loop and the remainder of the FRS.

To supplement the ambiguous constraints from the CSPs described above, we obtained unambiguous distance restraints using the measurement of paramagnetic relaxation enhancement (PRE) (SI Appendix, Materials and Methods and Fig. S12). These PRE measurements used ERK2 variants in which the seven native cysteines were replaced and single cysteines were reintroduced, one at a native site (Cys252) and another through an engineered mutation (Gly240Cys). These two ERK2 variants, each containing a single cysteine, were chemically modified by an S-(1-oxyl-2,2,5,5-tetramethyl-2,5-dihydro-1H-pyrrol-3-yl)methyl methanesulfonothioate spin-label. These spin-labeled species were used together with an EtsΔ138 construct that lacks all four native cysteines to obtain PREs. These various mutations lead to minimal changes in the enzymatic parameters (32), confirming that native interactions between ERK2 and EtsΔ138 are preserved. The spin-label at position 240 produced a number of significant PREs (indicative of shorter distances); a smaller effect was seen for the spin-label at position 252. (Position 240 lies on the same face as the FRS, whereas position 252 lies on the opposite face.) A majority of the PREs obtained in the latter case involved methyl groups that could be detected with high sensitivity. The unambiguous distances from the PRE measurements (SI Appendix, Materials and Methods) were combined with the CSP-derived AIRs (as above) to perform docking calculations using HADDOCK (34).

The lowest-energy structural ensemble (Fig. 4B) obtained from the HADDOCK calculations suggests that the bulk of the interface between the EtsΔ138 CTR and the ERK2 FRS is defined by the N terminus of H5 and the H4-H5 loop (which bears F120) of EtsΔ138 and helix α2L14 on ERK2, in line with the spectral perturbations described above. Additional interactions also include the H0-H1 loop and the N terminus of H1 on EtsΔ138 and the N terminus of αG on ERK2, which bears Lys229 and His230. Alteration of the latter two ERK2 residues (Lys229Thr/His230Asp) leads to an ∼15-fold increase in the K_M value for EtsΔ138 phosphorylation (29). Given the substantial amount of dynamics in this region of EtsΔ138 on both the picosecond-nanosecond (SI Appendix, Fig. S5B) and microsecond-millisecond time scales, the docking calculation does not allow us to unambiguously identify the specific EtsΔ138 partners that pair with Lys229 and His230 on ERK2. Additional points of contact are also found between the C terminus of H2 together with the H2-H3 loop on the PNT domain of EtsΔ138 with helices αL14 and αG on ERK2.

In ppERK2, the hydrophobic pockets that accommodate the two Phe rings of the canonical F-site sequence are formed by phosphorylated Tyr185, Arg192, Ile196, and Met197 (ϕ₁) and by Leu198, Tyr231, Leu232, Leu235, and Tyr261 (ϕ₂). The PNT domain of EtsΔ138 engages Tyr261 through F120, with Tyr261 and Leu262 displaying the largest amide perturbations in this region of ERK2 (SI Appendix, Fig. S6, Top Left) (12). Tyr231 is engaged through F88; an F88A mutation leads to an ∼2.2-fold increase in K_M value (29), compared with an ∼20-fold increase with an F120A mutation (22). Thus, based on our structure, it appears that there is no occupancy of ϕ₁, and only partial occupancy of ϕ₂, by EtsΔ138. It is important to note that both ϕ₁ and ϕ₂ are fully formed only on activation loop dual-phosphorylation (8). However, several residues that form the ϕ₂ pocket on activation are still sufficiently solvent-exposed in the inactive kinase and available for interaction. These findings are in accordance with our observation that EtsΔ138 binds both active and inactive forms of ERK2 with similar affinity.

Finally, we tested whether the HADDOCK-derived structures of EtsΔ138 docked at the ERK2 DRS and the FRS are compatible with each other. We note that an intervening loop (this loop is highly dynamic; SI Appendix, Fig. S5B) can be modeled, and that the two sets of interactions involving the N-terminal and C-terminal regions of EtsΔ138 can be readily merged into a single structure without introducing any distortions, confirming their mutual compatibility (SI Appendix, Fig. S13).

The Phospho-Acceptor Is Sensitive to ERK2 Dual-Phosphorylation and Activation.

As mentioned earlier, the CSPs induced on EtsΔ138 by both active and inactive forms of ERK2 exhibit similar overall patterns (Fig. 1). To further probe the local structural differences in EtsΔ138 in the context of the inactive (EtsΔ138•ERK2) and active (EtsΔ138•ppERK2) complexes, we extrapolated the expected chemical shifts for EtsΔ138 when fully bound to either form of ERK2 using transverse relaxation-optimized spectroscopy (TROSY)-based titrations and the ITC-determined dissociation constants. We then determined the CSPs for EtsΔ138 in the active complex, using the chemical shifts of EtsΔ138 in the inactive complex as a reference. The largest differences in backbone chemical shifts between the active and inactive complexes are localized near the phosphorylation site (Fig. 5A); in fact, the largest difference is seen for the phospho-acceptor T38 (0.23 ppm). These data suggest that the phosphorylation site on EtsΔ138 senses changes resulting from dual phosphorylation and activation of ERK2. In addition, the CSP seen for the γ2 methyl group of T38 (0.24 ppm) is significantly larger than those of the other Ile and Thr methyl positions in EtsΔ138 (Fig. 5C).

Fig. 5.

Chemical shift differences among the inactive, active, and prechemistry states of EtsΔ138. Chemical shift differences between the extrapolated fully bound states of the inactive (EtsΔ138•ERK2) and active (EtsΔ138•ppERK2) complexes of EtsΔ138 with ERK2 are plotted against the EtsΔ138 sequence. The magnitudes of the perturbations for the amide (A) and Ile (δ1)/Thr methyl positions (C) are shown by the orange bars, except for the phospho-acceptor T38, which is represented by the blue bar. Corresponding differences between the active state (EtsΔ138•ppERK2) and prechemistry state (EtsΔ138•ppERK2•Mg²⁺/AMPPCP) are also shown for the amide (B) and Ile (δ1)/Thr methyl (D) positions. In C and D, residues for which the Ile/Thr resonances could not be extrapolated because of excessive line broadening during the titrations are indicated by asterisks.

As discussed above, H0, the dynamics of which show significant changes on engaging ERK2, also exhibits substantial chemical shift differences between the inactive and active complexes (Fig. 5A). We suspect that these differences result from the fact that H0 makes transient contacts with the activation loop of ERK2 and thus experiences an alteration in its local environment induced by the dual phosphorylation of the ERK2 activation loop and the resulting conformational changes (35). However, these interactions are transient and evidently not as robust as those seen in a canonical F-site sequence (with ppERK2), where interactions N-terminal to the F-X-F motif (and C-terminal to the phosphorylation site) dampen conformational exchange in the activation loop and in part of FRS on the microsecond-millisecond timescale (as discussed above) (15). Finally, another cluster of residues exhibiting significant differences in backbone chemical shifts between the inactive and active complexes is seen around F120 (Fig. 5 A), a key residue in the PNT/FRS interactions. This again may reflect local effects of dual phosphorylation on the ERK2 activation loop together with its resulting conformational rearrangements felt at the spatially proximal F120 that occupies a part of the FRS.

Notably, the formation of a prechemistry complex (EtsΔ138•ppERK2•Mg²⁺/AMPPCP) induces minimal additional perturbations on the EtsΔ138 backbone (Fig. 5B). However, at the level of Ile and Thr sidechains, a modest but significant CSP is seen for the T38 γ2 position (0.09 ppm) (Fig. 5D). A comparable CSP is also seen for the I13 δ1 position, raising the possibility of cross-talk between the active site and the DRS in ERK2 transmitted to the docked EtsΔ138. The precise origin of this effect is not immediately evident and merits further investigation. It should be noted that previous studies have suggested long-range coupling between the DRS and the active site of ERK2 (36).

The Phosphorylation Site Remains Dynamic in the Prechemistry Complex.

As discussed above, in EtsΔ138 the phospho-acceptor T38 shows a substantial chemical shift response to dual-phosphorylation and activation of ERK2, suggesting that it senses the active site of the kinase (and changes therein) to a significant extent. In addition, the substantial k_cat (14 s⁻¹; Table 1) and specificity constant (k_cat/K_M = 0.6 ×10⁶ M⁻¹s⁻¹) values suggest that the phospho-acceptor is in a favorable orientation for catalysis. This is in contrast to the poor k_cat (0.6 s⁻¹) and low k_cat/K_M (0.04 × 10⁶ M⁻¹s⁻¹) values measured for an Elk-1–derived substrate peptide (Elk_387–399, which carries a canonical F-site) (15), in which the phospho-acceptor was found to be highly dynamic in the prechemistry complex (Elk_387–399 •ppERK2•Mg²⁺/AMPPCP). We asked whether the higher efficiency of phospho-transfer to EtsΔ138 results from an enhanced rigidity of the T38 sidechain enabling the formation of a stable Michaelis complex. To answer this question, we compared the sidechain dynamics of methyl positions of Thr (and Ile) residues in free EtsΔ138 with that in the corresponding prechemistry complex.

The order parameters (S²_axis) for the Ile and Thr methyl groups calculated from the ¹H-¹H intramethyl cross-correlated relaxation rate, η, (37) in IT-labeled EtsΔ138 (τ_C = 8.6 ± 0.3 ns; determined from backbone relaxation rates) (38) indicates the existence of significant disorder at the N terminus, in line with the backbone relaxation data. T10, T12, I13, I14, and T16 all show S²_axis values ≤0.17, whereas all residues of the PNT domain show S²_axis values ≥0.5 with the exception of I124 (0.39) (SI Appendix, Fig. S14). The phospho-acceptor T38 is highly dynamic, with an S²_axis = 0.29 ± 0.01. Estimating the rotational correlation time for the fully bound prechemistry state based on extrapolated relaxation rates (e.g., in the case of the {¹H}-¹⁵N NOE, discussed above) is problematic, given the substantial exchange contributions to the R₂s in the partially saturated complex. Therefore, to assess changes in methyl sidechain dynamics on complex formation, we used the ratio of the η values measured at two different populations of the prechemistry complex (η_bound) against those for the free state (η_free) of EtsΔ138. Because η values scale linearly with both the τ_C and S²_axis (SI Appendix, Eq. S6), a uniform increase in the η values and therefore also in the η_bound/η_free ratio would imply that the local dynamics (i.e., S²_axis) are unchanged, and that the increase is the result of a larger τ_C due to complex formation. A value higher or lower than the mean would imply an increase or a decrease in order, respectively. As shown in Fig. 6 for two ratios (45% bound and 77% bound), the η_bound/η_free values of T10, T12, and I14 suggest increased ordering on complex formation. As discussed above, these residues belong to the D_hyd-site that interacts with the ERK2 DRS. (I13 was extensively broadened, and data for this residue could not be analyzed with significant precision for the prechemistry complex.) However, T38 remains disordered on formation of the prechemistry complex.

Fig. 6.

Dynamics of EtsΔ138 sidechains. Ratio of the intramethyl ¹H-¹H dipolar cross-correlation rates (η) of the prechemistry state to that of free EtsΔ138 shown for two sets of bound populations (77%, in black, and 45%, in red). The shaded regions represent the 10% trimmed mean ± SD for the two populations (77%, in gray, and 45%, in light red). Residues for which data could not be analyzed with high precision are indicated by asterisks.

Increased Catalytic Efficiency Toward EtsΔ138 Results from Optimal Localization of the Phospho-Acceptor at the ERK2 Active Site.

Biochemical data indicate that the spatial position of T38 is well optimized with respect to both docking interactions. A peptide encoding the first 52 residues of Ets-1 (1–52) is phosphorylated by ppERK2 with a similar k_cat value as EtsΔ138, albeit with a ∼28-fold higher K_M value (22). The Δ24EtsΔ138 construct also registers a similar k_cat value for phosphorylation by ppERK2, but with an approximate fivefold difference in K_M value (Tables 1 and 2). Thus, the disorder around the phospho-acceptor (as discussed above) ensures that as long as it is in a particular position (likely evolutionarily optimized) with respect to either docking site, the phosphorylation occurs normally once the appropriate Michaelis complex is formed. Therefore, any modifications in the positioning of T38 with respect to either docking site would be expected to lead to alterations in the k_cat value and thus in the overall specificity.

Table 2.

Kinetic parameters for the phosphorylation of Δ24EtsΔ138 positioning mutants by ppERK2

Sequence	Threonine position	kcat, s−1	KM, µM	kcat/KM, µM−1s−1
32ADVPLLTPSSKEMMSQ47	-38TP39-	13 ± 0.9	105 ± 17	0.12
32ADVPLTPASSKEMMSQ47	-37TP38-	9 ± 0.4	53 ± 7	0.17
32ADVPTPAASSKEMMSQ47	-36TP37-	6.7 ± 0.8	181 ± 47	0.04

Parameters are the best fits to Eq. S12 in SI Appendix. The first row represents the wild-type position of the phospho-acceptor in Δ24EtsΔ138.

Given the persistent flexibility at the N-terminal region of EtsΔ138 in the complex with ERK2, an N-terminal shift of the phospho-acceptor would be expected to be better tolerated than a shift toward the C terminus, closer to the structured PNT domain. This is because in the latter case, additional structural changes, such as the further destabilization of H0 or an alteration in the location of this disorder with respect to the PNT domain, would be necessary to restore the optimal positioning of T38 at the active site (SI Appendix, Fig. S13). In extreme cases, disruption of the PNT domain itself may be required. To test this hypothesis, we measured the capability of several EtsΔ138 positioning variants to be phosphorylated by ppERK2. In these variants, the phosphorylation motif (³⁸TP³⁹) was moved toward or away from the PNT domain by introducing alanine spacers while leaving the overall length of the loop constant (Table 1). Consistent with our hypothesis, we found that moving the TP motif closer to the PNT domain led to significant reductions in the k_cat (2.5- to 4.6-fold decrease), with only modest effects on K_M (Fig. 7), resulting in lower values of the specificity constant. Also in line with our prediction, the effect of moving the motif away from the PNT domain by two residues led to a smaller reduction in k_cat. However, moving the motif away from the PNT domain by a single residue toward the N terminus (creating a PLTP motif rather than the wild-type PLLTP motif) decreased the k_cat value as expected, but also led to a threefold decrease in K_M, resulting in an approximate twofold increase in specificity. A similar trend was noted in the absence of the N-terminal docking site, as reflected in the Δ24EtsΔ138 positioning variants shown in Table 2. This curious result could reflect further optimization in the binding interactions within the Michaelis complex. Of note, a PXTP sequence was identified as an optimal motif for phosphorylation by ppERK2 (39). Conversely, we found that replacing ³⁸TP³⁹ by ³⁸TA³⁹ led to an approximate threefold increase in K_M, along with a fourfold decrease in k_cat (Table 1), suggesting that the increased flexibility of the substrate at the P+1 position destabilizes both the Michaelis complex and the transition state for T38 phosphorylation.

Fig. 7.

Phosphorylation of EtsΔ138 positional mutants by ppERK2. Data for the steady-state kinetic assays for wild-type EtsΔ138 and five unique positional mutants that move the ³⁸TP³⁹ phosphorylation motif away from (underscored) or toward the PNT domain are shown. The circles and solid lines represent the experimental data and fits, respectively, to SI Appendix, Eq. S12.

Discussion

In this study, we used NMR approaches to build on our previous kinetic analyses (18, 20, 40) to define the molecular mechanism by which ERK2 binds and phosphorylates the transcription factor Ets-1 on T38.

Molecular Recognition of ERK2 by Ets-1 Involves Noncanonical Interactions at the DRS and the FRS.

As mentioned above, Ets∆138 targets ERK2 by engaging parts of both its DRS and FRS in a unique bipartite fashion. The N terminus of EtsΔ138 contains a set of residues (D_hyd-site) that partially mimic the hydrophobic sequence (Φ_A-X-Φ_B) of canonical D-site sequences. The lack of stable electrostatic contacts involving the Ets∆138 N terminus at the ERK2 DRS, as in canonical D-site/DRS interactions, results in low intrinsic affinity and persistent dynamics. Nonetheless, there are some similarities between the interactions involving the N-terminal D_hyd-site of Ets∆138 and those involving the hydrophobic segment of canonical D-sites at the ERK2 DRS (SI Appendix, Fig. S8). Among the several canonical D-site ligands that we inspected, the orientations of key EtsΔ138 sidechains that contact the ERK2 DRS closely resemble those of HePTP, which binds in an N→C orientation as in the case of EtsΔ138 (36), and PEA-15 (17), which binds in a reverse orientation, i.e., C→N (SI Appendix, Fig. S8). Notably, I13 (which occupies the ϕ_A pocket) on EtsΔ138 shows a similar orientation to L123 in PEA-15 and L29 in HePTP. In addition, P9 on EtsΔ138 contacts the ERK2 surface in an orientation that partially mimics that of P126 on PEA-15 and of V25 on HePTP. Indeed, overall similarities between the backbone CSPs induced at the ERK2 DRS by HePTP, PEA-15, and EtsΔ138 have been reported previously (12, 28).

The mode of engagement of the ERK2 FRS by the EtsΔ138 CTR (H1+PNT) deviates significantly from that of a canonical F-site. Unlike in the latter case, which involves the insertion of the two Phe rings composing the F-X-F motif deep into the fully formed FRS (15) in ppERK2, the interaction surface of the EtsΔ138 CTR (SI Appendix, Fig. S15A), although significantly larger in spatial extent, does not involve contacts that can be easily parsed as essential (except perhaps F120). This is the likely cause of the reduced affinity (by ∼15-fold) of the EtsΔ138 CTR (Δ51EtsΔ138) toward ERK2 compared with that of a canonical F-site sequence for ppERK2 (19). In the case of canonical F-site sequences, additional stabilization is obtained through transient contacts with the dual-phosphorylated activation loop (8, 15). These interactions appear to play a less significant role in the present case, as detailed above. Despite these differences, however, there is some overlap between the regions of ERK2 that engage a canonical F-site sequence (e.g., ³⁹⁵FQFP³⁹⁸ in Elk_387–399) and the CTR of EtsΔ138. Most notably, P398 in Elk_387–399 clashes with F120 (SI Appendix, Fig. S15A), and thus it is not surprising that a ligand encompassing a canonical F-site sequence (Lig-F) can inhibit the phosphorylation of both EtsΔ138 and Δ24EtsΔ138 by ppERK2 (26).

Interestingly, there are several points of similarity between the ERK2-binding modes of EtsΔ138 and the ERK2 inhibitor, PEA-15 (17). Both species engage ERK2 in a bidentate manner, with the globular domains [the PNT domain for EtsΔ138 and the death effector domain (DED) for PEA-15] engaging the FRS (SI Appendix, Fig. S15B), and a flexible tail (the N terminus for EtsΔ138 and the C terminus for PEA-15) interacting with the hydrophobic portion of the DRS (although the PEA-15 C terminus engages the ERK2 DRS in an antiparallel C→N fashion; SI Appendix, Fig. S8). As mentioned above, the N-terminal D_hyd-site of EtsΔ138 assumes an extended conformation and engages the DRS through a limited number of hydrophobic contacts, similar to the C terminus of PEA-15. Higher B-factors for the PEA-15 C-terminal residues that contact the ERK2 DRS, and the lack of electron density for the rest of the C-terminal fragment that connects it to the globular DED, suggest that this interaction also is likely dynamic in nature, mirroring the EtsΔ138 N terminus. Like EtsΔ138, PEA-15 engages the ERK2 FRS (primarily through the C- and N-terminal ends of two consecutive, antiparallel helices of its DED; SI Appendix, Fig. S15B) in noncanonical fashion, for both inactive as well as ppERK2 (41). However, it does not engage identical regions of the FRS like the EtsΔ138 PNT domain. The most significant overall difference between the PNT- and DED-binding modes is the fact that the DED shows a more intimate interaction with the ERK2 FRS and forms a larger network of contacts by also engaging the ERK2 activation loop and its contiguous αEF helix, and by doing so induces a number of structural rearrangements in ERK2. The fact that these differences between the two ERK2 ligands are real and not a manifestation of the lower precision of the structure presented here is supported by the much tighter binding affinities measured for PEA-15 (∼50 nM) (41) than for EtsΔ138. This intimate contact between PEA-15 and ERK2 is also supported by our previous NMR measurements, where the CSPs induced by PEA-15 on ERK2, although spatially similar to those induced by EtsΔ138, were significantly larger in magnitude (approximately twofold higher backbone CSPs for PEA-15). In addition, the ¹³Cα CSPs were also larger in the case of PEA-15, suggesting larger conformational rearrangements in ERK2 in the presence of PEA-15 (12).

Proximity-Mediated Catalysis.

As evident from the foregoing relaxation analyses, the sidechain of the phospho-acceptor T38 in EtsΔ138 remains highly dynamic when bound to ERK2 and does not form a well-ordered complex with the catalytic elements of ppERK2, suggesting that phosphorylation is “proximity-mediated” (20). In this mechanism, the binding affinity is determined primarily by remote docking interactions at the DRS and FRS that serve to enhance the local concentration of the substrate ³⁸TP³⁹ moiety near the active site (Fig. 8). Efficient phosphorylation is enabled by the generation of a structural ensemble in which the phospho-acceptor T38 samples various conformations, a significant fraction of which place the phosphorylatable sidechain in optimal alignment with the ppERK2 catalytic elements.

Fig. 8.

Schematic representation of the mechanism of proximity-mediated phosphorylation of EtsΔ138 by ERK2. EtsΔ138 docks onto ERK2 using two noncanonical docking interactions: a fuzzy interaction involving a part of the ERK2 DRS and a mostly rigid body interaction involving a portion of the ERK2 FRS. The docking generates an ensemble of conformations of the dynamic phospho-acceptor near the ERK2 active site. This ensemble comprises states in which the phospho-acceptor and surrounding regions (including the Pro in the P+1 position) are in appropriate conformation (along with the catalytic elements of ERK2) for chemistry (high-activity state) and those in which they are not (low-activity state). We use two discrete states here for illustrative purposes, but multiple states of varying intrinsic activity could be involved. Docked complexes with substantial populations of high-activity states are efficiently phosphorylated.

To gain insight into this process, we estimated the fraction of docked complexes that go on to produce a phosphorylated product. Our earlier rapid quench studies had revealed a first-order rate constant for the phosphorylation of EtsΔ138 within the encounter complex of k_p = 109 s⁻¹ (18, 40). We performed transient kinetics measurements using a fluorescence competition assay (SI Appendix, Fig. S16) and obtained an EtsΔ138 dissociation rate from the complex with ppERK2, k_off ∼280 s⁻¹. This finding suggests that despite the transient nature of the encounter complexes between ppERK2 and EtsΔ138, a significant fraction (possibly as large as ∼40%) of collisions result in productive phosphorylation on T38. As might be expected, the generation of an optimal ensemble appears to be sensitive to the location of the phospho-acceptor with respect to the docking sites (as noted by the lower k_cat values reported in Tables 1 and 2 for mutants with displaced TP motifs, and by the requirement for a Pro at the P+1 position). The latter likely provides a greater degree of rigidity to stabilize both the Michelis complex and the transition state, in line with predictions from a QM/MM study (42) that provides a basis for the preference of a Pro at the P+1 position in MAP kinases.

Based on our present results and on our earlier work on Elk_387–399 (15)_, it is likely that many, if not all, ERK2 substrates (and possibly all MAP kinase substrates as well) are phosphorylated through the proximity-mediated mechanism (20) (Fig. 8). In the case of Ets-1, the mechanism underlies a highly specific and efficient phosphorylation on a single residue within the protein. Phosphorylation of other sites close to T38 is inefficient within the Michaelis complex, presumably because of their suboptimal sampling of the kinase active site owing to a combination of factors. These include the spatial distribution of the ensemble that limits the active site proximity (e.g., for an appropriate substrate motif such as ²⁶SP²⁷) and inefficient phosphorylation of a “poor” substrate, such as S41 (43), where a Lys (instead of a Pro) at the P+1 position prevents optimal orientation of the phospho-acceptor within the active site.

In addition to the location of T38 with respect to the two docking interactions, we hypothesize that the destabilization of H0 plays a role in generating the optimal ensemble to allow efficient phosphorylation. In contrast to the persistent dynamics at the N terminus of EtsΔ138, the rigid body interaction of the CTR (H1+PNT) necessitates some degree of conformational rearrangement to allow both docking and appropriate sampling of the ERK2 active site by the proximal phospho-acceptor. The fact that H0 is already dynamic and makes minimal contributions to the stability of the EtsΔ138/ppERK2 interface perhaps provides an ideal means of achieving proper placement of the phospho-acceptor at the ERK2 active site without significant deleterious effects on docking. Although our positioning mutants provide some evidence of this through the finding that moving T38 closer to the PNT domain has a more significant effect on catalytic efficiency, tuning the flexibility of this helix through mutations and supporting NMR and kinetics studies are needed to fully test this hypothesis.

In summary, in this work we have provided a molecular basis for how ERK2 phosphorylates Ets-1 through a proximity-mediated mechanism (Fig. 8). First, Ets-1 docks onto ERK2 using two unique suboptimal docking interactions, and in doing so localizes the phospho-acceptor near the ERK2 active site. After docking, the dynamic phospho-acceptor samples a variety of conformations, some in which the appropriate elements within the ppERK2 active site, ATP, and the phosphorylatable sidechain of Ets-1 (and surrounding regions including the Pro at the P+1 position) are appropriately aligned to enable chemistry (high-activity states), and others in which the relevant elements are not correctly aligned (low-activity states). Indeed, states of higher and lower intrinsic activity have been demonstrated in single-molecule studies of the phosphorylation of a peptide substrate by protein kinase A (PKA) (44). Thus, higher turnover is obtained in cases with a substantial population of high-activity states in the docked ensemble. We suspect that this is case for EtsΔ138, in contrast to Elk_387–399, explaining the large differences in efficiency (15). The persistence of dynamics around the phospho-acceptor may be a general feature of the bound states of protein kinase substrates irrespective of whether the primary recognition sequence is proximal to the phospho-acceptor, as for PKA substrates (45), or distal, as in Ets-1 (and other MAP kinase substrates). The lack of an organized substrate moiety could be one reason why kinases are “average enzymes” in terms of catalytic efficiency compared with their metabolic counterparts (46).

Materials and Methods

Protein Expression and Purification.

Various constructs of EtsΔ138 and ERK2 were expressed and purified, as described in detail in SI Appendix, Materials and Methods.

Resonance Assignments and NMR-Based Titrations.

Resonance assignments for EtsΔ138 were carried out at 25 °C in NMR buffer and used in NMR-based titrations, as described in detail in SI Appendix, Materials and Methods.

Backbone and Sidechain Relaxation Experiments and Data Analysis.

TROSY-based ¹⁵N relaxation experiments (47) were carried out for EtsΔ138 and analyzed using the reduced spectral density functions (48), as described in detail in SI Appendix, Materials and Methods. Intramethyl ¹H-¹H dipole-dipole cross-correlation rates (η) (37) were measured and analyzed, as described in detail in SI Appendix, Materials and Methods.

Determination of the Structural Models of the Complex of EtsΔ138 with ERK2.

Chemical-shift based ambiguous interaction restraints and unambiguous distance restraints obtained from paramagnetic relaxation enhancement rates (49) (backbone amides, Ile, Leu, and Val methyl groups for EtsΔ138 using spin-labeled ERK2) and intramethyl and intermethyl (Ile and Thr for EtsΔ138, Leu and Val for ERK2) to obtain structural models for the complex of EtsΔ138 with ERK2 using HADDOCK (31, 50). The procedures are described in detail in SI Appendix, Materials and Methods.

Measurement of Kinetic Parameters.

Kinase assays were performed at 28 °C in assay buffer to obtain the Michelis–Menten kinetic constants, as described in detail in SI Appendix, Materials and Methods. In addition, the rate of dissociation of the Ets∆138 from its ppERK2 complex was measured using fluorescence measurements, as described in detail in SI Appendix, Materials and Methods.