Substrate binding to Src: A new perspective on tyrosine kinase substrate recognition from NMR and molecular dynamics

Abstract

Most signal transduction pathways in humans are regulated by protein kinases through phosphorylation of their protein substrates. Typical eukaryotic protein kinases are of two major types: those that phosphorylate‐specific sequences containing tyrosine (~90 kinases) and those that phosphorylate either serine or threonine (~395 kinases). The highly conserved catalytic domain of protein kinases comprises a smaller N lobe and a larger C lobe separated by a cleft region lined by the activation loop. Prior studies find that protein tyrosine kinases recognize peptide substrates by binding the polypeptide chain along the C‐lobe on one side of the activation loop, while serine/threonine kinases bind their substrates in the cleft and on the side of the activation loop opposite to that of the tyrosine kinases. Substrate binding structural studies have been limited to four families of the tyrosine kinase group, and did not include Src tyrosine kinases. We examined peptide‐substrate binding to Src using paramagnetic‐relaxation‐enhancement NMR combined with molecular dynamics simulations. The results suggest Src tyrosine kinase can bind substrate positioning residues C‐terminal to the phosphoacceptor residue in an orientation similar to serine/threonine kinases, and unlike other tyrosine kinases. Mutagenesis corroborates this new perspective on tyrosine kinase substrate recognition. Rather than an evolutionary split between tyrosine and serine/threonine kinases, a change in substrate recognition may have occurred within the TK group of the human kinome. Protein tyrosine kinases have long been therapeutic targets, but many marketed drugs have deleterious off‐target effects. More accurate knowledge of substrate interactions of tyrosine kinases has the potential for improving drug selectivity.

1. INTRODUCTION

Tyrosine and serine/threonine protein kinases (PKs) are involved in the regulation of a myriad of eukaryotic signaling networks in response to extracellular or endogenous stimuli.1 Stringent control of signaling demands that PKs recognize and phosphorylate the correct substrates to ensure the proper cellular response while avoiding spurious signaling. Kinases with mis‐regulated activities that lead to aberrant signaling pathways continue to be identified in many cancers2, 3, 4, 5, 6, 7 and thus PKs have long been targeted for therapeutic intervention. Most ligands developed to modulate kinase activity are ATP analogs, with some recent efforts to develop allosteric inhibitors targeting pockets outside the ATP binding site. Nevertheless, one region that is relatively unexplored is the substrate‐binding site. But to this end, the structural information on substrate recognition of protein kinases is limited by comparison to ligands overlapping the ATP site. Furthering our knowledge on kinase interactions with their substrates could benefit the development of therapeutics with appropriate specificity.

The structures of protein kinases in complex with peptide substrates have contributed important knowledge about substrate recognition in the vicinity of the phosphorylation site, even though missing information on distal contacts that occur with protein substrates.8 The first protein kinase structure determined was the crystallographic structure of cAMP‐dependent protein kinase (PKA), a serine/threonine kinase, bound to a substrate analog peptide.9 In the PKA complex, and as subsequently shown for the serine/threonine kinases phosphorylase kinase (PhK),10 cyclin‐dependent kinase (CDK),11 Akt/protein kinase B12 and Pim‐1 kinase,13 the recognition of peptide substrate includes residues C‐terminal to the acceptor serine/threonine contacting the P + 1 loop located at the C‐terminus of the activation loop, with the peptide exiting the catalytic site along the cleft between the two lobes of the catalytic domain (CD) and positioned on top of the activation loop as viewed in Figure 1a. By contrast, the crystal structure of a tyrosine kinase complex between insulin receptor kinase (IRK) and a peptide substrate14 shows the larger tyrosyl sidechain is accommodated by differences in interactions at the P + 1 loop near a conserved proline (P1172 in IRK), so that the peptide residues C‐terminal to the acceptor tyrosine exit the catalytic site along the C‐lobe, below the activation loop as viewed in Figure 1b. Structures of peptide substrate complexes from the tyrosine kinases insulin‐like growth‐factor receptor (IGF1R),15 fibroblast growth‐factor receptor 2 (FGFR2),16 ephrin receptor A3 (EphA3)17 and epidermal growth factor receptor (EGFR)18 show a substrate orientation similar to that seen for IRK. To date, all crystallographic reports of substrate recognition in kinases find this distinction in substrate recognition between serine/threonine kinases positioned in the cleft and tyrosine kinases positioned on the C‐lobe.14, 19

Figure 1.

Substrate recognition in threonine/serine kinases compared to tyrosine kinases defined from currently known structures of kinase‐peptide complexes. The peptides are represented as ribbons with the phospho acceptor position in red. The kinase activation loop and P + 1 loop are highlighted (green). (a). Protein kinase A bound to PKI (yellow) (PDB ID: 1ATP) illustrates the cleft binding orientation on one side of the activation loop observed for serine/threonine kinases. (b). Insulin receptor kinase bound to a peptide substrate (magenta) (PDB ID: 1IR3) illustrates the C lobe binding orientation on the opposite side of the activation loop observed for tyrosine kinases. (c). Schematic representation of protein kinase peptide substrate complexes to compare binding orientations for cleft binding versus C lobe binding. The two orientations differ in the position of the substrate residues C terminal to the acceptor residue as the substrate exits the catalytic site

We investigated the association of a peptide substrate, acet‐AEEEIYGEFEA‐NH₂ (named here SSP),20 with cytosolic Src tyrosine kinase catalytic domain (SrcCD) using NMR spectroscopy and molecular dynamics (MD). A highlight of the current study is that paramagnetic relaxation measurements from NMR are consistent with substrate peptides binding SrcCD along the cleft in a manner similar to that observed for serine/threonine kinases, which is in contrast to other tyrosine kinases known to date.

1.1. SSP interaction with SrcCD monitored by NMR chemical shifts

¹⁵N HSQC spectra of isotopically‐labeled SrcCD (residues 255:533)21, 22 were measured in the absence and presence of saturating amounts of SSP to try to identify the interaction region from relatively larger chemical shift perturbations $(\text{CSP}=\sqrt{{(0.154\Delta {\delta }_{\text{N}})}^2+\Delta {\delta }_{\text{H}}^2})$ of resonances of residues localized near the binding site. Nevertheless, the residues with larger CSP values (>0.1 ppm, Figure 2a) are distributed throughout the structure (mapped in purple on the SrcCD surface, Figure 2b), and do not readily pinpoint the SSP binding site on SrcCD. Furthermore, the ¹⁵N‐HSQC spectrum of apo SrcCD is missing more than 100 resonances out of the 269 expected based on sequence, and more than 60 of these missing peaks appear in the spectrum of SSP‐bound SrcCD (Figure S3). A CSP value cannot be determined for these resonances, but the residues are shown as red stars in the CSP profile in Figure 2a, and colored magenta on the surface representation of SrcCD in Figure 2b. These residues are also broadly distributed in both the N and C lobes and not sequestered at a specific location. Similar long‐range effects on chemical‐shift were reported for ligands that bind Src near the ATP site.23 Together, the chemical shift changes of SrcCD indicate that SSP binding has a global effect on the conformational equilibrium of SrcCD, without notable localized effects from direct binding contacts. Therefore, the more targeted approach of paramagnetic relaxation was taken to elucidate SSP interaction with SrcCD.

Figure 2.

SrcCD SSP NMR chemical shift perturbation and paramagnetic relaxation. (a). Chemical shift perturbation of backbone amide peaks in the trosy‐HSQC spectra of SrcCD between the SSP bound and free states (black bars). Residues whose amide peaks appear only in the peptide‐bound spectrum are shown with red stars. A CSP value greater than 0.05 ppm is a reliable measurement based on linewidths and digital resolution of the spectrum. (b). Perturbation mapped to SrcCD surface for residues with CSP > 0.1 ppm (purple) or resonances observed only in SSP‐bound state (magenta). (c). The ratio of integrated peak volumes measured from HSQC spectra acquired for the oxidized and reduced forms of 3 (2 iodoacetamido) PROXYL plotted against the residue number. (d). PREs calculated from Vox/Vred = exp(−PRE × 2τ_INEPT), where τ_INEPT is the time for which the proton magnetization is transverse during the INEPT transfer sequence in the 15 N trosy HSQC. Residues with higher values of PRE are closer to the spin label. G300, whose amide peak is absent in the spectrum with an oxidized spin label, is shown as a dotted arrow. (e). Volume ratios mapped on the surface of SrcCD in red: Vox/Vred <0.25; green: 0.25 < Vox/Vred <0.50; slate: Vox/Vred >0.50; gray: unassigned residues

1.2. Paramagnetic relaxation for orienting bound SSP

The binding location for SSP was probed using paramagnetic relaxation enhancement (PRE), which causes a reduction in the ¹⁵N‐HSQC peak intensity depending inversely on the proximity of the amide proton to the paramagnetic spin label24, 25 (details in SI). Peptide substrates in the two orientations determined from crystallographic structures of protein kinase complexes differ most in the position of their C‐terminal residues (Figure 1c); a pY + 5 residue for a peptide bound on the C‐lobe would be approximately 10–15 Å away from that residue in a peptide bound in the cleft of the catalytic domain. To examine which of the two known substrate orientations SSP adopts when bound to SrcCD, SSP was modified by substituting the C‐terminal Ala residue with Cys and attaching the paramagnetic spin label 3‐(2‐iodoacetamide) PROXYL to generate SSP with residue CYP11, acet‐AEEEIYGEFE(CYP)‐NH₂ (Figure S2). The attenuation in peak intensities was monitored by trosy‐HSQC spectroscopy to identify SrcCD residues proximal to the C‐terminus of SSP and so obtain information on the peptide orientation.

That the spin label did not substantially perturb the bound conformation was verified by comparing the bound‐state HSQC spectrum using unlabeled SSP and reduced spin‐labeled SSP. Both peptides induced similar spectral changes relative to unbound SrcCD, and no significant differences in the bound‐state spectra were observed (Figure S5).

Peak volume ratios, Vox/Vred, determined from ¹⁵N‐HSQC spectra with the spin‐label either oxidized or reduced by addition of ascorbic acid (Figure S4), were used to estimate PREs. The volume ratio rather than intensities or differences in relaxation rates are used based on arguments given by Reference 26 and described in SI. Results are shown in Figure 2c, d and mapped in Figure 2e onto the surface representation of SrcCD. The residues colored light grey, which includes the activation loop, are not assigned (see SI), so no PRE‐rate information is available for them. The residues on the rear face of the CD, opposite the active site (Figure 2e, right‐most panel), have negligible PRE effect as expected.

Examination of Figure 2e finds the residues with strong PREs (Vox/Vred < 0.25, red) map to regions of the catalytic domain consistent with the two known orientations of bound peptide substrates, but are too broadly distributed on the surface of SrcCD for the spin‐labelled residue CYP11 to be simultaneously close to all of them. These strong‐PRE residues are located both in the N‐lobe near the active site (G300, T301, E305, A306, G406 and L407), and also in the C‐lobe (Y436, R438). E305 and A306 are in the first turn of helix C, a key helix in conformational activation,27, 28 while G300 and T301 are in the loop leading into helix C. G406 is the glycine in the well‐known DFG motif, and L407 is at the start of the activation loop. Y436 and R438 are positioned in the C‐lobe in contact with the activation loop. The broad distribution of strong PRE residues indicates CYP11 contacts SrcCD at multiple sites, and SSP adopts alternative conformations when associated with SrcCD. We therefore turned to computational modeling and simulations to interpret the PRE data for orienting SSP bound to SrcCD.

1.3. Distance‐restrained ensemble molecular dynamics

Distance‐restrained ensemble molecular dynamics (r‐eMD) was used to search for SSP orientations consistent with the PRE measurements (Figure 2e), with the assumption that SSP binds in one of the two known orientations of protein kinase substrates. Starting coordinates were modeled using the activated form of SrcCD (PDB: 1Y57) and with SSP based on the peptide in either the tyrosine kinase complex of IGF1R (PDB: 1K3A) for C‐lobe binding, or on the peptide in the serine/threonine kinase complex of PKA (PDB: 1ATP) for cleft binding. Four complexes, two with SSP modeled on the C‐lobe and two with SSP modeled along the cleft (details in SI), were energetically relaxed with unrestrained MD (2.2‐16.2 ns) before initiating r‐eMD simulations computed with the ensemble module29, 30 of CHARMM.31, 32 Restraint terms were included for all residues exhibiting strong PREs [Figure 2c‐e, red], which are the data measured with highest certainty. The PRE restraint distance d _PRE,i is between the CYP11 nitroxide oxygen atom of SSP and the amide proton of SrcCD residue i, and was set to an upper limit of 14 Å (details in SI). A more quantitative analysis of the PRE data is unnecessary to orient bound SSP. Each of the four computed ensembles, whether initiated with SSP modeled on the C‐lobe or in the cleft, satisfy all PRE restraint distances; that is, ${〈d_{\mathit{PRE},i}^{-6}〉}_{\mathit{ens}}^{-1/6}\le 14$ Å for all PRE residues (Table S1).

1.4. Cluster analysis of the global ensemble from r‐eMD

To characterize the SSP orientation and the position of the spin‐labelled CYP11 in the complexes, a cluster analysis of the Src‐SSP complexes from the r‐eMD ensembles was performed by combining the four ensembles into a global ensemble (4,000 conformers collected over 8 ns) and clustering on the distances corresponding to the PRE measurements, d_PRE. Fifteen clusters were obtained; Table 1 lists the PRE‐distance values that define each of the cluster centers. The values for a given cluster range from 5–9 Å for short distances, to as much as 30–40 Å for long distances. Center values ≤14 Å (boldface in Table 1) indicate the cluster accounts for the experimental PRE for that residue. No single cluster has centers that satisfy all PRE distance restraints, which confirms the premise that multiple conformations of bound SSP are needed to fit the NMR data in totality.

Table 1.

Cluster center values from clustering the global ensemble (4,000 conformers) computed with PRE‐restrained eMD

Cluster	Cluster center, d (Å)
	PRN1		PRN2				PRC
	G300	T301	E305	A306	G406	L407	Y436	R438
Cleft orientation
1	17	18	8	8	13	13	26	23
2	23	22	21	19	18	15	8	7
3	7	6	12	11	15	14	25	23
4	9	8	7	5	10	9	22	21
5	12	11	8	6	9	8	20	18
6	14	12	12	10	10	8	19	17
7	25	24	23	21	20	17	7	7
Clobe orientation
8	39	38	37	35	34	31	11	14
9	38	36	36	34	33	30	10	13
10	41	39	38	37	35	32	12	15
11	35	34	34	32	32	29	10	13
12	30	28	29	28	28	25	10	13
13	43	41	40	38	36	34	13	17
14	33	32	31	30	30	27	9	12
15	31	30	31	30	29	27	12	15

The ensemble was clustered on the distances between the nitroxide oxygen of the spin‐labeled CYP11 residue of SSP and the main chain amide hydrogen of the indicated SrcCD residues, which exhibit strong PREs. Values ≤14 Å are consistent with PRE measurements and shown in boldface. The cleft and C‐lobe orientation of SSP is determined by examining the cluster average structures, Figure 3a).

The orientation of SSP in the 15 clusters of the global ensemble is shown in Figure 3a. We represent each cluster with the nearest‐to‐average member of the cluster, which is the member with the minimum RMSD to coordinates averaged over all cluster members (see Methods/SI). In the cluster‐average complexes, SSP is either oriented along the cleft and above the activation loop as visualized in Figure 3a (blue main chain ribbon traces, clusters 1–7), or SSP lies along the C‐lobe and below the activation loop (gold traces, clusters 8–15). An analysis finds that certain members of the cleft‐positioned clusters derive from trajectories initiated with SSP on the C‐lobe, and therefore the C terminus of SSP traveled more than 15 Å in response to the PRE distance restraints. None of the C‐lobe cluster members derive from cleft‐initiated trajectories.

Figure 3.

SSP association with SrcCD derived from PRE r eMD, unrestrained MD and probed by mutagenesis. (a) The 15 cluster average complexes of the global ensemble of r eMD have SSP oriented either in the cleft (blue) or on the C‐lobe (gold) on opposing sides of the SrcCD activation loop (green). Complexes were aligned on SrcCD (surface representation) main chain heavy atoms. SSP is shown by a backbone trace and a red sphere for the PROXYL oxygen atom of the spin labelled residue CYP11. A dashed line is drawn between CYP11 Cα and this oxygen atom. SrcCD residues that exhibit strong PREs (red) are broadly distributed in the N lobe/cleft (PRN1 = G300, T301; PRN2 = E305, A306, G406, L407) and C lobe regions (PRC = Y436, R438). Two or more cleft SSP orientations together have spheres near all PRE residues and account for the PRE data in totality, whereas the C lobe orientations position the oxygen atom near only PRC residues, with none near PRN1 or PRN2 residues. (b) Conformational transition of cleft bound SSP (stick representation) from unrestrained MD involving displacement of the spin labelled CYP11 residue (orange) resulting in all ${〈d_{\mathit{PRE}}^{-6}〉}^{-1/6}<14$ Å. Top: conformation early in the simulation with CYP11 PROXYL oxygen atom (red sphere) near PRN1 and PRN2 (dotted lines). Bottom: conformation after 60 ns with the oxygen atom near PRC (dotted line). (c) Time series from unrestrained trajectories initiated with cluster averages either in the cleft (top) or on the C lobe (bottom) for d _YD, the distance between the SSP Tyr 5 Oη and the catalytic Asp 386 Cγ as a measure of the nearness of the SSP to the catalytic site. (d) Mutagenesis results corroborate the cleft binding orientation. Left: Catalytic rate constants (K _M and k _cat/K _M) measured for WT and the variant SrcCDL407D. Right: An overlay of cleft (blue) and C lobe (gold) orientations of SSP illustrating the proximity of L407 to SSP in the cleft‐bound but not the C‐lobe position

Figure 3a demonstrates which PRE distances are fit by each cleft and C‐lobe cluster average. SSP oriented on the C‐lobe (gold) positions CYP11 near the region labeled PRC (PRC=PRE residues Y436 and R438), but none of the C‐lobe SSP structures have CYP11 near other PRE residues. In contrast, SSP oriented in the cleft (blue) has alternative conformations that position CYP11 near different PRE residues. One cleft average has CYP11 positioned near the region labelled PRN2 (PRN2 = PRE residues E305, A306 G406 and 407), several cleft averages have CYP11 near PRN1 (PRN1 = PRE residues G300 and T301) and PRN2, while two have CYP11 near PRC. Accordingly, together SSP conformers oriented in the cleft have CYP11 in close proximity to all PRE residues. Therefore, a primary outcome of the r‐eMD simulations is the cleft‐oriented set of SSP conformations together satisfy all PRE restraints, while the C‐lobe‐oriented SSP conformations satisfy only the PRE restraints to PRC

1.5. Unrestrained MD simulations reveal fluctuations of cleft‐oriented SSP consistent with the full set of PRE data

The structural variation of the cluster averages in Figure 3 raises the question of whether conformational averaging in the bound state of SSP occurs, and specifically whether the conformers oriented in the cleft undergo fluctuations that can account for all PRE measurements. To further explore the nature of the conformational sampling of SSP in the bound state, SrcCD‐SSP was simulated without the PRE‐distance restraint potential starting with coordinates of the cluster averages shown in Figure 3a and for a 100‐ns time period.

Interestingly, time‐averaged ${〈d_{\mathit{PRE}}^{-6}〉}^{-1/6}$ values computed from the unrestrained trajectories (Table S2) revealed spontaneous fluctuations can occur in the C‐terminus of SSP bound in the cleft to generate conformations consistent with the full set of PRE data. Thus, in the simulation started from cluster‐6 average, CYP11 is initially near only to PRN1 and PRN2. A transition in the SSP C‐terminus occurs during the trajectory that places CYP11 in proximity to PRC residues. The transition is illustrated in Figure 3b with the top panel showing a conformation early in the trajectory with CYP11 close to PRN1 and PRN2, and the bottom panel from the second half of the trajectory showing CYP11 near PRC. Similar fluctuations in the C‐terminus of SSP occurred in a second cleft‐oriented trajectory initiated from cluster‐3 average. While these observations from unrestrained simulations are only suggestive, and require additional study to more accurately define the actual conformational ensemble of the SrcCD‐SSP complex, the results demonstrate conformational averaging of SSP consistent with the totality of the PRE data when the substrate is oriented in the cleft and positioned above the activation loop.

1.6. Nearness to catalytic site in unrestrained MD

Protein kinases have a conserved Asp residue that is required for catalytic activity by interaction with the accepting residue hydroxyl group.1 For Src, this role lies with Asp 386. We therefore monitored the distance between SSP Tyr 5 Oη and Asp 386 Cγ (d _YD) to assess the nearness of SSP to the catalytic site during unrestrained MD. In all cleft‐oriented SSP complexes and for a subset of the C‐lobe‐oriented complexes, this distance remains 3.2–3.3 Å (Figure 3c, top), which approximates the distance of a catalytically competent complex. In contrast, in some trajectories initiated from cluster averages with SSP oriented on the C‐lobe (Figure 3c, bottom), the distance d _YD is initially less than ~3.5 Å, but within ~30 ns fluctuations displace the acceptor Tyr of SSP out of the SrcCD active site.

1.7. Mutagenesis corroborates SSP binding in the cleft

The proposal that SSP binds in a cleft orientation was tested by mutagenesis of L407, the DFG + 1 conserved hydrophobic residue, which exhibits a strong PRE and lies in the cleft in proximity to SSP oriented in the cleft but not SSP oriented on the C‐lobe (Figure 3d). L407 sidechain is solvent exposed in the absence of peptide substrate, but would be partly buried by bound SSP oriented in the cleft and not on the C‐lobe. Substitution with aspartic acid in the L407D variant is expected to disrupt cleft‐oriented binding due to desolvation of an acidic group without compensating favorable electrostatic interaction with residues of SSP. The L407D variant of SrcCD, SrcL407D, was produced and purified with methods similar to wild‐type, and the enzymatic activity tested using a continuous assay (details given in SI). The K _M for peptide substrate of wild‐type SrcCD was measured to be 494 μM, but with SrcL407D peptide saturation could not be achieved using concentrations >2 mM and therefore the K _M could not be determined accurately. The k _cat /K _M determined from the complete progress curves33 is 33‐fold lower for SrcL407D than wild‐type SrcCD. Thus, the reduced k _cat /K _M of SrcL407D further supports the cleft binding orientation for Src tyrosine kinase.

2. DISCUSSION

Until this report, past structural information on peptide substrate/inhibitor complexes of protein kinases found that tyrosine and serine/threonine kinases are distinguished by their recognition of substrates14, 34, 35; however, contrary to the current paradigm, the PRE structural analysis, MD computations and mutagenesis results for SrcCD presented here suggest that Src tyrosine kinase binds C‐terminal substrate residues in the cleft between the two lobes of the catalytic domain similar to serine/threonine kinases and distinct from other tyrosine kinases. An earlier NMR study on a peptide substrate complex of Lyn, another member of the Src‐family, also argued that substrates of Lyn could bind similarly to serine/threonine kinases,36 but the rationale was based on exchange‐transferred NOE spectroscopy that detects NMR signals from only the substrate, and not the kinase. Here, PRE data on SrcCD indicate a cleft‐binding orientation is necessary to account for the PRN1 and PRN2 paramagnetic signals, while the PRC signals are consistent with either binding orientation (Figure 3a). SSP, bound in the cleft, contacts L407, which provides a direct rational for the reduced catalytic efficiency observed with the Src_L407D variant (Figure 3d). Furthermore, spontaneous fluctuations of the SSP C‐terminus observed in unrestrained MD simulations resulted in CYP11 being proximal to all of the residues with strong PREs (Figure 3b), which illustrates how SSP oriented in the cleft can account for the totality of the PRE data, whereas a C‐lobe orientation for the substrate can only account for part of the observed PREs. All together these results find that the cleft orientation is necessary and sufficient, but the C‐lobe orientation cannot be ruled out. Nonetheless, one orientation is almost certain to be more catalytically effective, and the distances between the catalytic Asp sidechain and the acceptor Tyr in the unrestrained MD simulations (Figure 3c) suggest that the cleft orientation is more likely to be catalytically important for Src. Whereas the C‐lobe binding mode of other tyrosine kinases positions the substrate receptor tyrosine residue near the catalytic machinery in the phosphorylation site through β‐sheet contacts with the activation loop,14 the activation loop of Src is unknown to form a β structure, and C‐lobe orientation does not appear to lead to a well‐positioned tyrosine in the active site. This orientation therefore could be inhibitory rather than catalytic, and is similar to the positioning of a bisubstrate inhibitor.37

Our analysis indicates Src substrate recognition differs from other tyrosine kinases. All tyrosine kinases fall in the TK group of the eukaryotic kinome, but the Src family of non‐receptor tyrosine kinases (SFKs) lie in a different branch of the TK group than IRK, IGF1R, FGFR2, EGFR and EphA3 (Figure 4a), and it is reasonable that substrate recognition is an evolutionary distinction within the TK group rather than between TK and other kinome groups. We therefore asked about potential sequence differences among the branches of the TK group and compared the sequences of the C‐terminal end of the activation loop plus the P + 1 loop, which is the region highlighted in Figure 1, and shown for SrcCD in Figure 4b, where contact with substrate differs between the two binding orientations; in the cleft orientation, the peptide substrate interacts with the C‐terminal end of the activation loop on the side facing the cleft, whereas the interaction for the C‐lobe orientation is on the opposite side of the activation loop, and in the case of IRK, IFG1R and EGFR a short antiparallel β‐sheet structure is formed between the peptide and the C‐terminal end of the activation loop. A difference in conformation of the C‐terminal ends of the phosphorylated activation loop of IRK and the SFK member LCK was, in fact, noted previously.14 We conducted a multiple‐sequence alignment38 of the TK group catalytic domains to compare Src with IRK, IGF1R, FGFR2, EGFR and EphA3 (Figure 4c). The contact residues of the C‐terminal end of the activation loop and the start of the P + 1 loop (residues 413–425 in Src numbering) is a region where the sequence has diverged between the TK branches. For example, positions 420, 422 and 424 are E, A and F, respectively, in the SFKs, but are G/T/−, G and V/L/I in the other branches. The sequence differences support the premise that separation in evolution of the TK branches involved changes in substrate recognition associated with the C‐terminus of the activation loop.

Figure 4.

Sequence differences in the C terminal end of the activation loop among the branches of the TK group suggests an evolutionary basis for the alternative substrate recognition of Src. (a). Phylogenetic tree (HYPERTREE) of the TK group of the human kinome. Branches for which a peptide substrate kinase complex structure has been determined by crystallography are in color. (b). Close‐up view of the region of the kinase activation loop and P + 1 loop (green, residues 413–425) which interacts differentially with substrate oriented in the cleft (blue, cluster 4) or on the C‐lobe (gold, cluster 8). (c). Sequence alignment for Src residues 413–429 with proteins in the TK branches shown in color in A. Aligned residues are the C‐terminus of the activation loop and the P + 1 loop (Src 422–429). Some sequence differences between the Src branch (top) and the other three branches are highlighted in orange

While short peptides clearly bind the active site and are efficiently phosphorylated, protein substrates involve additional contacts away from the phosphorylation site that can contribute to substrate recognition.8, 35, 39, 40 The distal contacts, which occur with the CD or through docking to separate regulatory domains, could be linked to the evolution of disparate binding modes at the active site within the TK group; the relative positioning between substrate and kinase imposed by the distal contacts could impact evolution of the direction the substrate exits the phosphorylation site. In any case, the bulk of the evidence is consistent with the idea that the phosphorylated region of an intact protein would bind the kinase similarly as a peptide; primary sequence determinants of peptides are often reproduced in cellular protein substrates, suggesting that the recognition of amino acids near the active site is the same. That disparate surfaces are contacted with the two binding modes should assist in the development of substrate‐competitive inhibitors with increased selectivity and improved treatment against disease.

CONFLICT OF INTEREST

The authors declare no potential conflict of interest.

Supporting information

Appendix S1: Supporting Information

Joshi MK, Burton RA, Wu H, et al. Substrate binding to Src: A new perspective on tyrosine kinase substrate recognition from NMR and molecular dynamics. Protein Science. 2020;29:350–359. 10.1002/pro.3777