Conformational states dynamically populated by a kinase determine its function

Abstract

Protein kinases intrinsically sample a number of conformational states with distinct catalytic and binding activities. We used nuclear magnetic resonance spectroscopy to describe in atomic-level detail how Abl kinase interconverts between an active and two discrete inactive structures. Extensive differences in key structural elements among the conformational states give rise to multiple intrinsic regulatory mechanisms. The findings explain how oncogenic mutants can counteract inhibitory mechanisms to constitutively activate the kinase. Energetic dissection revealed the contribution of the activation loop, the DFG motif, the regulatory spine and the gatekeeper residue to kinase regulation. Characterization of the transient conformation to which the drug imatinib binds enabled the elucidation of drug resistance mechanisms. Structural insight into inactive states highlights how they can be leveraged for the design of selective inhibitors.

One sentence summary:

NMR data allows dissection of regulatory and drug resistance mechanisms in Abl kinase

Protein kinases are key enzymes involved in mediating cell signaling by phosphorylating tyrosine, serine or threonine residues in target proteins. The activity of protein kinases is tightly regulated typically by means of intra-molecular interactions, binding of protein partners, and post-translational modifications (1–3). Dysregulation of kinase activity by deletions or mutations often results in cancer, inflammation or neurodegeneration (4, 5) and disease-associated kinases are the target of therapeutic intervention by small molecules (6, 7). Many genetic mutations associated with cancer affect kinase domains (8).

Evidence from biophysical experiments and computational studies, as well as from structures of kinases in complex with various inhibitors, indicates that protein kinases are dynamic and sample alternate conformational states (9–17). However, it has proved challenging to monitor these transitions and structurally characterize the manifold of conformational states populated by a kinase. Knowledge of the conformational ensemble could give insight into the mechanisms of regulation. Evolutionary pressure has forced kinases to adopt similar active states so that regions required for catalysis are precisely positioned (18, 19). In contrast, inactive states can be distinct and may in principle vary significantly among kinases. Structural information of the inactive states adopted by kinases could aid in the development of selective inhibitors.

Three key structural elements typically determine whether a kinase adopts the active or an inactive state (18, 19): the catalytic Asp-Phe-Gly (DFG) motif, the activation loop (A-loop), and the αC helix (Fig. 1A). In the fully active state the DFG motif adopts the so-called “DFG-in” state, wherein the Asp (Asp400 in Abl) of DFG points inside the active site cleft and is poised to coordinate Mg-ATP; the A-loop adopts a so-called open conformation that allows substrates to dock onto it to be phosphorylated; and the αC helix adopts the so-called “αC-in” state wherein a conserved αC helix Glu residue (Glu305 in Abl) forms an ion pair with a conserved Lys residue (Lys290 in Abl) in the β3-strand that coordinates the α and β phosphates of the ATP. A significant deviation from this architecture will result in an inactive state.

Fig. 1. Characterization of the energetically excited conformational states in Abl kinase by NMR CEST experiments.

(A) Structure of the Abl kinase domain in the active conformation determined in the current work. Key structural features are highlighted. (B) Methyl groups in Abl indicating the presence of one (colored cyan) or two (colored magenta) excited conformational states in the NMR CEST experiments. (C) Representative ¹³C CEST profiles of the indicated methyl groups. The major dip corresponds to the major (ground) state whereas the minor dips correspond to the energetically excited conformational states E₁ and E₂. As explained in the main text, excited states E₁ and E₂ correspond to inactive states I₁ and I₂, respectively. (D) Energy landscape of the ground (G), E₁ and E₂ states denoting their populations and kinetics of interconversion as measured by fitting the CEST data. (E) Correlation of the CEST-derived chemical shift difference between the ground and E₁ states (Δ(E₁–G)_CEST) with the chemical shift difference between apo and PD173955-bound Abl. Methyls (marked with an asterisk) for which the dips corresponding to the ground and E₁ states are within ±0.3 p.p.m. cannot be resolved in the CEST profile and their chemical shift difference was extracted from ¹H-¹³C–correlated experiments (see materials and methods). Methyls that are close (within 6 Å) to PD173955 were excluded from the correlation because their chemical shifts are directly affected by the inhibitor. Val398^CG1 deviates from linearity because of the rearrangement of the nearby Phe401 side chain in the inhibitor complex compared to the E₁ state. (F) Correlation of the CEST-derived chemical shift difference between the ground and E₂ states (Δϖ(E₂–G)_CEST) with the chemical shift difference between apo and imatinib-bound Abl. Methyls that are close (within 6 Å) to imatinib were excluded from the correlation because their chemical shifts are directly affected by the inhibitor. Methyls colored blue deviate from linearity as discussed in the text and in fig. S8. Methyls (marked with an asterisk) for which the dips corresponding to the ground and E₂ states are within ±0.3 p.p.m. cannot be resolved in the CEST profile and their chemical shift difference was extracted from ¹H-¹³C–correlated experiments.

Abl kinase is involved in a number of signaling pathways that control cell growth, survival, invasion, adhesion and migration (20–22). The breakpoint cluster region (Bcr)-Abl, an aberrant fusion form of Abl with dysregulated activity causes leukemia and other cancers and is the target of a number of inhibitors, such as imatinib (Gleevec) (20, 23). However, a majority of patients develop mutations that confer resistance to imatinib and other inhibitors (24–28). While it has been possible to explain how some of these mutations decrease the inhibitor affinity, the mechanistic basis for the effect of mutations in sites remote to the inhibitor is not known. We have used nuclear magnetic resonance (NMR) spectroscopy to structurally characterize how Abl switches between the active and two distinct inactive states and to dissect how the conformational ensemble is exploited by mutants, ligands, post-translational modifications, and inhibitors to regulate the kinase activity and function.

Accessing alternate conformational states of Abl kinase

NMR has provided a wealth of information on the role of dynamics in the regulation of kinases (14, 29–37). The ¹H-¹⁵N– and ¹H-¹³C–correlated NMR spectra of the Abl kinase domain (hereafter Abl; residues 248–534) labeled in methyl-bearing (Ala, Ile, Met, Leu, Thr and Val) and aromatic residues were of high sensitivity and resolution (fig. S1). In the spectra of unliganded Abl recorded at 20 °C or higher, the resonances of the activation loop (A-loop) and nearby residues at the αC helix (Fig. 1A) broaden beyond detection (fig. S1B,C). Such extreme line broadening is commonly observed in kinases studied by NMR in their unliganded state (11, 29) and has typically prevented a detailed characterization of the A-loop. The A-loop encompasses the DFG motif, forms the docking site for substrate binding, and, in tyrosine kinases, includes the tyrosine residue(s) that are phosphorylated to regulate activity (38). We obtained near-complete methyl assignment of Abl by lowering the temperature to 10 °C where all of the expected resonances were visible (fig. S1B,C).

The observed line broadening is due to millisecond time-scale conformational transitions indicating the presence of alternate conformational states (39). To access the conformational states sampled by Abl, we used the chemical exchange saturation transfer (CEST) NMR technique (40, 41). In this experiment, a weak radiofrequency field is applied to monitor magnetization transfer in sites that undergo conformational exchange. We recorded ¹³C-based CEST experiments on Abl samples labeled (¹H,¹³C) specifically in the 104 methyl-bearing residues (Fig. 1B and fig. S1A), which afforded high sensitivity even at low temperatures. A CEST profile was obtained for each one of the methyl probes in Abl. If exchange is present (with an exchange rate, k_ex, between ~20−500 s⁻¹), a large dip at the chemical shift of the dominant, energetically ground state (G) and a smaller dip at the chemical shift of the alternate, energetically excited state are observed (Fig. 1C and fig. S2). Methyl probes reporting on the conformational transitions are distributed throughout the protein with many of them being located in or near the A-loop and the αC helix (Fig. 1B).

Forty methyl probes (colored cyan in Fig. 1B) showed one minor dip in their CEST profile (e.g. Val318^CG1 and Ile333^CD1; Fig. 1C), while 43 methyl probes (colored pink in Fig. 1B) showed two minor dips indicating the presence of two distinct energetically excited conformational states (e.g. Leu267^CD2 and Leu389^CD2; Fig. 1C) referred to as E₁ and E₂. CEST data fitting (see materials and methods) indicated a linear equilibrium, G↔E₁↔E₂ (Fig. 1D). The exchange rate k_ex for the G↔E₁ transition is 46.8±4.3 s⁻¹ and for the E₁↔E₂ transition is 88.7±13.5 s⁻¹. The population of the ground state (p_G) is 88% whereas the population of either E₁ (p_I1) or E₂ (p_I2) is 6% indicating that the two excited states are equally stable (Fig. 1D).

Structure of the Abl ground state

The structure of unliganded Abl is not known. We used NMR spectroscopy to determine the structure of the ground state (population 88%; Fig. 1D and fig. S1B, C) of the isolated Abl kinase domain in solution. The structural data demonstrate that unliganded Abl kinase domain adopts a catalytically active conformation in solution (Fig. 2A and fig. S3A–C) with the A-loop adopting an open conformation and the DFG motif and αC helix both in the “in” state (Fig. 2A). Phe401 of the DFG motif fits snugly into a hydrophobic pocket lined by Val308, Met309, Ile312, Leu317, Val318, and Leu373, and these interactions further stabilize the DFG motif in the “in” state. Tyr412, the residue that when phosphorylated stimulates Abl activity (42), forms hydrogen bonds through its side chain to Arg381 and Arg405, and these interactions stabilize the open conformation of the A-loop. The phosphate-binding loop (P-loop; also referred to as the Gly-rich loop) is an important region in kinases because it directly affects ligand binding in the active site (19, 43). Tyr272, the bulkiest residue in the P-loop, points inwards and occupies the cleft where the phosphate group of ATP is located (Fig. 2A). Thus, Abl adopts predominantly the active state in solution and inhibitors, such as dasatinib, that selectively target the Abl active state bind with high affinity (fig. S3D, E) (44).

Fig. 2. Structures of the ground (active), E₁ (I₁) and E₂ (I₂) states of Abl.

(A) NMR solution structure of the ground state of Abl shows that the Abl kinase domain inherently adopts the fully active state. The zoomed view shows the hydrophobic residues surrounding Phe401. (B) Structure of the Abl E₁ state (green) superimposed on the structure of the active state (blue). Because Abl E₁ adopts an inactive state is referred to as I₁. The zoomed views highlight similarities (e.g. αC helix) and differences (e.g. DFG motif and A-loop) in the disposition of key structural elements between the two states. (C) Structure of the Abl E₂ state (orange) superimposed on the structure of the active state (blue). Because Abl E₂ adopts an inactive state is referred to as I₂. The major structural rearrangement undergone by the αC helix, the A-loop and the P-loop between the two states is indicated by an arrow. (D) Superposition of the structures of the Abl I₂ state (orange) and the Abl–imatinib complex (cyan; PDB ID 1IEP). The zoomed view highlights the structural rearrangement upon imatinib binding in the residues lining the pocket. (E) ¹³C CEST profiles of Abl and Abl^E311K demonstrate that the E311K substitution destabilizes the Abl I₁ state but has no apparent effect on the I₂ state. A denotes active state. The E311K substitution abrogates the ion pair between Glu311 and Arg405 in the I₁ state (panel B).

Structure of the Abl excited state 1 (E₁)

To gain structural insight into the low-populated conformational states that Abl samples, we directly measured the ¹³C chemical shift of the methyls in these two states from the corresponding CEST profiles (Fig. 1C). We next compared the chemical shifts of these two states to the chemical shifts of Abl in complex with twelve different inhibitors (fig. S4). Structural data (28, 45–52) have shown that these inhibitors can capture Abl in distinct conformational states and thus we asked whether any of the low-populated conformational states sampled by unliganded Abl resembles an inhibitor-bound state.

The ¹³C methyl chemical shifts of the Abl E₁ state showed remarkable similarity to the chemical shifts of Abl in complex with inhibitor PD173955 (Fig. 1E and fig. S5A). To structurally characterize Abl E₁ we sought to increase its population at the expense of the ground state as shown before for other systems (53–55). The double amino acid substitution M309L/H415P combined with adjusted pH buffer conditions collectively increased the population of E₁ from 6 to 50% (fig. S5B–D). NMR data showed that the double amino acid substitution M309L/H415P shifts the equilibrium towards E₁ without eliciting other structural changes (fig. S5B,E). We used the Abl^M309L/H415P variant to obtain NOE-based distance restraints to gain insight into the structure of Abl E₁ (fig. S5F,G). The structural data showed that Abl in E₁ adopts a structure (Fig. 2B) that is distinct and uncommon among unliganded protein kinases with known structures. Specifically, the DFG motif is flipped 180°, with respect to the active conformation, and it adopts the DFG-out conformation. However, the A-loop remains in an open conformation similar to the active conformation. Because of the DFG flip, the five residues following the DFG motif (Leu403-Met407) of the A-loop are also flipped by 180°. Thus, although the overall structural disposition of the N-terminal part of the A-loop is very similar in the active and in E₁ states, the backbone dihedral angles are very different so that the side chains of the residues point in opposite directions (Fig. 2B). Asp400 is in the “out” conformation and swaps positions with Phe401 of the active state. Thus, Asp400 is now located inside a hydrophobic pocket and to compensate for this unfavorable environment its side chain forms a hydrogen bond with the carbonyl oxygen of Val318. With Phe401 flipping inside the ATP-binding pocket, Leu403 swings in the opposite direction. As a result, Arg405, which in the active conformation forms a hydrogen bond with Tyr412 (Fig. 2A), pivots outwards and forms a salt bridge with Glu311 of the αC helix. The αC helix adopts the “in” conformation where Glu305 forms an ion pair with Lys290, similarly to the active state (Fig. 2B). Overall, Abl in E₁ adopts the DFG-out, αC-helix-in conformation and thus it is catalytically inactive. We will refer to E₁ as inactive state 1 (I₁). Structural and NMR data analyses are consistent with Abl in state I₁ adopting a structure that is very similar to the crystal structure of the complex with the PD173955 inhibitor (46). Thus, PD173955 binds to and stabilizes a conformational state that already preexists in the conformational ensemble of unliganded Abl (fig. S5H).

Structure of the Abl excited state 2 (E₂)

The ¹³C methyl chemical shifts of Abl E₂ show an excellent correlation with the chemical shifts of Abl in complex with imatinib (Fig. 1F). The crystal structure of Abl in complex with imatinib showed that a hallmark is the closed conformation of the A-loop (46). All of the methyl probes in or around the A-loop experience large chemical shift changes (Δϖ) as a result of the A-loop undergoing a large conformational change upon imatinib binding. For example, the two methyl groups of Leu406 have very characteristic chemical shifts; methyl CD1 has a ¹³C Δϖ of approximately −2.8 p.p.m. and methyl CD2 a ¹³C Δϖ of approximately 2.1 p.p.m. (Fig. 1F). The excellent correlation of the Δϖ values between E₂ and the imatinib-bound state observed indicates that the A-loop in E₂ adopts the closed conformation. Of note, a number of methyl probes on or nearby the αC helix deviate significantly from linearity (blue spheres in Fig. 1F) suggesting that the αC helix adopts a different conformation in Abl E₂ than that observed in the complex with imatinib.

To gain more detailed structural insight into the Abl E₂ state, we sought to increase its population at the expense of the ground state as we did above for the Abl E₁ state. From the analysis of the crystal structure of the Abl-imatinib complex (46), augmented by structural modeling of the unliganded Abl aided by the determined chemical shifts in the E₂ state, we reasoned that the T408Y substitution should stabilize the closed A-loop conformation by forming a hydrogen bond between the hydroxyl of Tyr408 and the side chain of Asp344 or Asn341 (fig. S6A). NMR analysis of the Abl^T408Y variant showed that indeed the substitution increases the population of E₂ from 5 to 70% (fig. S6B). Because the exchange between the active and inactive states is slow on the NMR time scale, two sets of peaks are present in the NMR spectra of Abl^T408Y: the chemical shifts of one of the sets correspond to those of the ground, active state, while the chemical shifts of the second set correspond to the chemical shifts of the E₂ state as measured by CEST (fig. S6B). The two sets of peaks are connected in a magnetization exchange experiment (fig. S6C) further corroborating that the two states are in dynamic equilibrium. To further increase the population of the E₂ state we produced the M309L/T408Y double substitution (fig. S6D). The NMR data clearly show that the single T408Y or the double M309L/T408Y substitution shift the equilibrium towards E₂ without eliciting other structural changes (fig. S6B,E). The Abl^M309L/T408Y and similar variants (Materials and Methods), which have E₂ populated at ~90–100%, allowed us to measure a large number of NOE-based distance restraints (fig. S7A) that were used to determine the structure of the Abl I₂ state (fig. S7B).

The structure of the Abl E₂ state is depicted in Fig. 2C. Compared to the active state, the A-loop in the E₂ state rotates with the middle of the A-loop translating by as much as 35 Å (see arrow in Fig. 2C). A segment in the C terminus of the A-loop folds back and masks the substrate docking site, while Tyr412 occupies the substrate Tyr position thereby preventing substrate binding altogether. Similar to the Abl I₁ state, the DFG motif in Abl E₂ adopts the “out” conformation. However, Phe401 of the DFG in the E₂ state does not only flip into the catalytic pocket as was observed in I₁ (Fig. 2B), but in addition it translates by ~11 Å to occupy a hydrophobic pocket that is lined by Leu267, Val275, Ala288, Phe336, and Leu403 (Fig. 2C). This hydrophobic pocket is formed as a result of the void created by Tyr272 (P-loop) swinging outwards and is further stabilized and enlarged by the repositioning of A-loop residues Leu403 and Leu406 (Fig. 2C). The positioning of Phe401 into this hydrophobic pocket directly blocks binding of ATP or competitive inhibitors.

In the E₂ state the entire P-loop undergoes a pronounced rotation and translation with Tyr272, located in the middle of the P-loop, shifting by ~16 Å towards the αC helix (Fig. 2C). The repositioned Tyr272 acts as a wedge to push the αC helix away from the C-lobe. As a result, Glu305 in the αC helix rotates outwards by ~90° and breaks its ion pair with Lys290, an interaction that is required for catalysis (9) (Fig. 2C). Thus, the αC helix transitions from the “in” conformation in the active state to the “out” conformation in the E₂ state. Taken together, Abl in the E₂ state adopts a fully inactive conformation with the four key structural elements, A-loop, DFG motif, αC helix and P-loop, all being in conformations that are not compatible with either substrate binding or ATP binding or hydrolysis. We will refer to Abl E₂ state as inactive state 2 (I₂).

Structural differences between Abl I₂ and the Abl−imatinib complex

Comparison of the structure of Abl in the I₂ state and the crystal structure of Abl in complex with imatinib (46) indicate similarities but also key differences (Fig. 2D). The most important similarity is the conformation of the A-loop that adopts a closed conformation in both the I₂ state and the imatinib complex. Imatinib pushes Phe401 towards the C-lobe by ~3 Å to make room for its pyrimidine ring inside the hydrophobic pocket. The DFG motif is in the “out” conformation both in I₂ and the imatinib complex. However, imatinib binding elicits a pronounced change to the structure and disposition of the P-loop and the αC helix. Strong imatinib binding to Abl entails the formation of a hydrogen bond between its secondary amino group and the carboxyl side chain of Glu305 (46). This interaction forces the αC helix to the “in” conformation. In addition, the P-loop switches from the stretched conformation in I₂ that blocks imatinib from entering into the binding pocket to the kinked conformation observed in the structure of the imatinib complex (Fig. 2D). Thus, imatinib apparently binds selectively to the Abl I₂ state, a low-populated conformational state that resembles the final imatinib complex structure in the A-loop and DFG motif but is very different in the αC helix and P-loop (Fig. 3A). These structural differences explain (fig. S8) the discrepancy between the CEST-derived chemical shifts of certain residues in the Abl I₂ state and their observed chemical shifts in the Abl−imatinib complex (Fig. 1F). Our data suggest that imatinib expends energy to elicit the structural changes required for its binding to Abl likely resulting in a lower affinity. By contrast, the PD173955 inhibitor binds to the I₁ state with minimal structural changes (fig. S5H), which could explain the higher affinity and potency of PD173955 compared to imatinib for Bcr-Abl despite the fewer contacts between the inhibitor and Abl in the PD173955 complex (46).

Fig. 3. Structural and energetic dissection of imatinib-resistance Abl variants.

(A) Structures of the three Abl conformational states and binding of imatinib to the I₂ state highlighting the disposition of key structural elements. (B) Patient-derived imatinib-resistance mutants studied here shown on the structure of Abl in complex with imatinib and on the structure of the Abl I₂ state. (C) Change in affinity for imatinib by the Abl variants relative to wild type Abl as determined by ITC (fig. S9). Error bars are standard deviation determined from a triplicate. (D) Effect of the H415P substitution on the population of the Abl states measured by ¹³C CEST experiments. The changes in the populations are indicated in a free energy diagram. A zoomed view of the mutation site suggests how the Pro substitution may destabilize the Abl I₂ state. (E) Effect of the Y272H substitution on the population of the Abl states measured by ¹³C CEST using the wild type Abl as reference (top) and by ¹H-¹³C–correlated experiments using the Abl^T408Y variant, which populates Abl I₂ state at 70%, as reference (bottom). A zoomed view of the mutation site indicates that substitution of Tyr272 will disrupt the hydrogen bond to Glu305 thereby destabilizing the Abl I₂ state. (F) Effect of the E274V substitution on the population of the Abl states measured by ¹H¹³C–correlated spectra using the Abl^I2M variant as reference. A zoomed view of the mutation site indicates that substitution of Glu274 will disrupt the hydrogen bond to Lys293 thereby destabilizing the Abl I₂ state. (G) Effect of the F378V substitution on the population of the Abl states measured by ¹³C CEST using the wild type Abl as reference (top) and by ¹H-¹³C–correlated spectra using the Abl^T408Y variant as reference (bottom). For additional data and discussion see fig. S10.

Mechanisms underpinning imatinib-resistance mutations

Imatinib is a selective inhibitor of Abl kinase (56, 57) and is widely used to target Bcr-Abl in chronic myelogenous leukemia (CML) patients (23). However, a serious problem is the occurrence of mutations that confer resistance to imatinib in a majority of the patients (24–27, 58). The underlying mechanism of action is straightforward to understand in mutations, e.g. T334I, that occur in positions that line the drug-binding site and sterically clash with imatinib to decrease its affinity for Abl (Fig. 3B). By contrast, the mechanisms conferring drug resistance by mutations in positions remote to the imatinib site have remained enigmatic.

Substitution of His415 by Pro confers resistance to imatinib, despite the fact that this residue is located ~18 Å from the inhibitor (Fig. 3B). We find that Abl^H415P binds to imatinib 5-fold weaker than the wild type Abl (Fig. 3C and fig. S9), consistent with a previous report (59). To understand the basis for the affinity decrease, we used NMR CEST experiments to measure the effect of the substitution on the Abl conformational ensemble. The data show that H415P destabilizes the Abl I₂ state to the extent that it is not detectable by the CEST experiments, indicating that the population of I₂ in Abl^H415P is less than 1% (Fig. 3D). Depletion of the state to which imatinib selectively binds (Fig. 2D), will result in a decrease in imatinib affinity for Abl (Fig. 3C). The destabilization of the I₂ state by the H415P substitution results in the stabilization of the I₁ state, with its population increasing 2-fold (Fig. 3D). The structural basis of these effects can be understood because a Pro residue, given its rigidity and constrained backbone dihedral angles as noted before (51), is not compatible with the sharp turn the A-loop makes at this position in the I₂ state, while it can be tolerated in the open conformation of the A-loop observed in Abl I₁ (Fig. 3D).

Several imatinib-resistance mutations cluster in the P-loop, including G269E, Q271H, Y272H and E274V (Fig. 3B). It has been suggested that CML patients with P-loop mutations have a poorer prognosis (60). Tyr272 is a key residue in stabilizing the stretched P-loop conformation in the Abl I₂ state by packing against Phe302 of the αC helix and using its hydroxyl group to form a hydrogen bond with the side chain of Glu305 (Fig. 2C). The transition of the P-loop from the kinked conformation observed in the active state to the stretched conformation in the I₂ state is necessary to accommodate the closed conformation of the A-loop in I₂. Quantitative NMR CEST and ¹H-¹³C–correlated experiments show that the Y272H substitution decreases the population of the I₂ state ~7-fold (Fig. 3E), accounting for the ~10-fold decrease in the affinity of imatinib for Abl^Y272H (Fig. 3C and fig. S9). His272 is not well poised to form a hydrogen bond with Glu305 and its aromatic π stacking with Phe302 is suboptimal compared to Tyr272, resulting in destabilization of the stretched P-loop conformation and decrease of the Abl I₂ population.

Glu274 forms an ion pair with Lys293, thereby stabilizing the stretched conformation of the P-loop in the I₂ state, is frequently mutated to a Val in patients that have developed imatinib resistance (Fig. 3F). Because the singly mutated Abl^E274V variant is not stable enough for CEST studies, we measured the effect of the E274V substitution using the Abl^{G269E/M309L/T408Y} (referred to as Abl^I2M) variant, which populates primarily the I₂ state (p_I2 ~90%), by ¹H-¹³C–correlated spectra. The data showed that the loss of the Glu274−Lys293 ion pair results in destabilization of the Abl I₂ state by ~1.4 kcal mol⁻¹, which translates to a several fold decrease in its population thus explaining the lower affinity of imatinib for Abl^E274V (Fig. 3C,F).

The F378V imatinib-resistant mutation (Fig. 3B) decreases binding to the inhibitor ~4-fold (Fig. 3C and S9). Consistent with this, NMR analysis of the Abl^F378V variant reveals that the F378V substitution decreases the population of the I₂ state ~4-fold (Fig. 3G). On the basis of NMR analysis, we surmise that Val378 exerts its effect by stabilizing the active state, thereby shifting the conformational ensemble towards the active state and depleting the I₂ state (fig. S10A). Our NMR data also explain the imatinib-resistance phenotype of two additional mutations in the P-loop, G269E and Q271H, which exhibit a more complex behavior (fig. S10B,C). Taken together, our findings illustrate how amino acid substitutions in different structural elements of the Abl kinase can use various mechanisms to destabilize the conformational state to which imatinib selectively binds giving rise to drug resistance.

The role of the regulatory spine in activity regulation

A highly conserved structural feature in protein kinases is the so-called regulatory (also referred to as the hydrophobic) spine (61). Mutations in this spine can directly activate or suppress the catalytic activity of kinases (62–64); however, direct experimental evidence on the underlying mechanisms is lacking. In Abl, the four residues making up the regulatory spine are Met309, Leu320, His380 and Phe401 (Fig. 4A). We used NMR to obtain structural and thermodynamic insight into how mutations in the regulatory spine may remodel the ensemble of the active and the inactive states in Abl. Our data show that the spine is fully assembled in the active state but broken in both the I₁ and I₂ states due to the DFG motif being in the “out” conformation (Fig. 4A). The structural deviation is more pronounced in the I₂ state than in the I₁ state because of the move of Phe401 towards the hydrophobic pocket in the nucleotide binding site. ¹³C CEST and ¹H-¹³C–correlated NMR experiments showed that substitution of Leu, a residue more frequently seen in kinases (65), for Met at position 309 increases the population of I₁ and I₂ from 6% each to 10% and ~35%, respectively, and hence increase the total population of inactive states ~4-fold (Fig. 4B). Substitution of the bulkier Ile for Leu at position 320 has an even greater effect with the I₂ state population increasing to 65% (Fig. 4B). The double M309L/L320I substitution switches almost the entire ensemble towards the I₂ state (p_I2 ~82%) and decreases the active state population to 8%, essentially deactivating the kinase, as confirmed by kinase assays (Fig. 4B and fig. S11). The energetic contribution of the M309L and L320I substitutions towards stabilizing the inactive states is 1.3 and 2.2 kcal mol⁻¹, respectively (Fig. 4C).

Fig. 4. Quantitative dissection of the effect of the regulatory spine, the gatekeeper, the DFG motif, and phosphorylation on the Abl conformational ensemble.

(A) Structure of the active state of Abl highlighting the regulatory spine, made up by Met309, Leu320, His380, and Phe401. The disposition of the regulatory spine is also shown for Abl states I₁ and I₂. (B) Effect of amino acid substitutions at the regulatory spine on the populations of the Abl conformational states as measured by NMR. Because the k_ex rates of the transitions (Fig. 1D) are slow on the NMR chemical shift time scale the resonances of the individual conformational states can be seen in ¹H-¹³C–correlated spectra if their population is above ~5%. Thus, in addition to ¹³C CEST, ¹H-¹³C–correlated experiments were also used to quantitate the populations, especially for Abl variants that are not stable over the period of time required for CEST experiments (materials and methods). There is excellent agreement in the population measurements by CEST and ¹H-¹³C–correlated experiments. The labeling scheme of the Abl variants is as shown in fig. S1. Determination of the kinase activity (fig. S11) shows that the Abl^M309L/L320I variant inhibits Abl. (C) Energy contribution to the active and I₂ states of Abl by the indicated variants. Negative values of ΔΔG indicate increased stability of the Abl I₂ state whereas positive values indicate increased stability of the Abl active state. A graph is included where the populations of the active and I₂ states are plotted as a function of the associated free energy, ΔG/RT, where R is the gas constant and T is the temperature. A 0.6 kcal mol⁻¹ change in ΔG corresponds to a change by 1 unit in ΔG/RT at room temperature (77). The populations of the two states for Abl (isolated kinase domain) and full-kinase Abl^FK are shown to help assess the effect that each one of the variants has on the activation or inhibition of Abl. (D) Effect of the Thr334 to Ile substitution at the gatekeeper position on the populations of the Abl states. Given that Abl exists predominantly in the active state (p_A ~88%), to measure the full effect we used the Abl^I2M variant (Abl^{G269E/M309L/T408Y}), which populates predominantly the I₂ state (p_I2 ~90%). (E) Effect of amino acid substitutions of the Phe residue in the DFG motif on the populations of the Abl states. Val or Tyr substitution results in an Abl kinase that adopts primarily the I₂ state. The low kinase activity of Abl^F401V is consistent with the NMR findings. (F) Effect of Tyr412 phosphorylation in the A-loop on the populations of the Abl states. Tyr412 phosphorylation eliminates both I₁ and I₂ states and stabilizes the active state as evidenced by ¹³C CEST experiments. To measure the full effect, we used the Abl^I2M variant as discussed in (D). The zoomed views of the superposition of the structure of each of the inactive states on the structure of the active state provides mechanistic insight into the effect.

Although residues with higher hydrophobicity would be expected to promote the assembly of the spine, and thus stabilize the active state, our data show that this is not the case. Both the Met to Leu (position 309) and the Leu to Ile (position 320) substitutions are toward more hydrophobic residues (66) and both stabilize the inactive state. It is likely that in addition to hydrophobicity, the size and shape of the residue side chain at these positions are important in determining the packing and thus the overall assembly of the spine. It is of interest that a Met at the residue corresponding to the Abl 309 position is conserved in ~21% of protein kinases, while a Leu is conserved in ~54% of them. At the residue corresponding to the Abl 320 position, a Leu is conserved in ~54% and an Ile in ~8% of them (65). Thus, even conservative mutations that may cause only subtle changes in hydrophobicity and size of the amino acids in the regulatory spine can give rise to pronounced changes in the population of the active and inactive states thereby strongly regulating kinase activity. This finding may explain why the regulatory spine is a hotspot of oncogenic mutations in various kinases (63).

Effect of the gatekeeper residue

The residue at the gatekeeper position, which controls accessibility to the hydrophobic region of the ATP pocket in protein kinases, is important for at least two reasons: (i) mutations to bulky residues, such as Ile and Met, at the gatekeeper position (67, 68) stimulate the intrinsic kinase activity and give rise to drastically increased cell transforming and oncogenic activity (62, 69); and (ii) mutations to bulky residues confer drug resistance in some kinases (24, 70). The mechanistic basis for the gatekeeper-induced kinase activation is not known. In Abl, the strongest effect reported is for the gatekeeper Thr334 residue substituted by Ile (62, 71). We characterized Abl^T334I by NMR and the analysis showed that the T334I substitution shifts the equilibrium towards the active state. Because the Abl kinase domain already exists predominantly in the active state (p_A ~88%), it is experimentally challenging to quantitate this effect. To do this, we used the Abl^I2M variant, which populates primarily the I₂ state (p_I2 ~90%) while the active state is only marginally stable (p_A ~10%) (Fig. 4D). T334I reverses the populations by strongly promoting the active state (p_A ~93%) (Fig. 4D). The data show that the Ile in the gatekeeper position stabilizes the active state by ~3.2 kcal mol⁻¹ (Fig. 4C) and reveal that the basis for kinase activation is a shift of the conformational ensemble towards the catalytically active state. Presumably, the Ile exerts its effect by stabilizing the hydrophobic spine (65) through favorable contacts with their nonpolar side chains (Fig. 4D).

The role of the DFG motif in activity regulation

Mutations in the DFG motif have been identified as “drivers” in human cancers in several kinases (72). Although the highly conserved nature of the Asp residue in the DFG motif is explained by its key role in the chemical steps of the catalysis (73), the reason for the conservation of the Phe residue is not well understood (74). We thus prepared the Abl^F401V variant, which NMR data show that it adopts almost exclusively (higher than 95%) the I₂ state (Fig. 4E). The F401V substitution in Abl flips the DFG motif from its preferable “in” state to the “out” state, which is stabilized by ~3.0 kcal mol⁻¹ (Fig. 4C). In agreement, kinase assays showed that the catalytic activity of Abl^F401V is at least 20-fold lower than that of the wild-type protein (Fig. 4E and fig. S11). We observed a similarly strong shift towards the DFG-”out” state in the Abl^F401Y variant (p_I2 ~90%; Fig. 4E). Thus, even a conservative substitution in this position can affect the ensemble of active and inactive conformational states giving rise to impaired function. This observation is consistent with and may explain findings on other kinases. For example, p38α loses its activity upon substitution of the Phe residue of the DFG motif to Tyr (75) and LRRK2, which features a Tyr in this position is retained in an inhibited state and substitution by Phe activates the kinase (76). Although Abl I₂ is stabilized in the Abl^F401V and Abl^F401Y variants, the I₁ state is not. The energetic balance of the contacts of the residue at position 401 with the residues in the two hydrophobic pockets where Phe401 was observed to reside into in the active and the I₂ states (Fig. 2A,C), will determine the relative populations of the two states and thus the activity of the kinase.

The Phe residue in the DFG motif is conserved in ~90% of kinases, but ~7% of them feature a Leu residue (65). In PKA, the Phe to Leu mutation has no effect on its catalytic activity (65). The NMR data of Abl^F401L show that this variant populates predominantly the active state, similarly to wild type Abl (Fig. 4E) thus explaining why a Leu residue in this position can be tolerated.

The role of A-loop phosphorylation in kinase activation

Phosphorylation of Tyr/Thr/Ser residues in the A-loop is a common mechanism for stimulating the activity of protein kinases (38), including Abl (42). We quantitated the effect of Abl Tyr412 phosphorylation (pTyr412) on the conformational ensemble of active and inactive states. NMR CEST data show that in Abl^pY412 both inactive states I₁ and I₂ are depleted and only the active state is detected (Fig. 4F). Our structural data illustrate the mechanistic basis for this effect. Both the active and I₁ states feature an open A-loop conformation with Tyr412 exposed to the solvent. In the active state the phosphate group of pTyr412 forms ion pairs with Arg381 and Arg405 (Fig. 4F). In the I₁ state, although Arg381 is well poised to form a salt bridge with pTyr412, Arg405 is pointing outwards and interacts with Glu311 (Figs. 2B and 4F). Phosphorylation of Tyr412 causes Arg405 to flip inward to form a salt bridge with pTyr412, thereby destabilizing the I₁ state.

In the I₂ state, Tyr412 binds as a pseudosubstrate and blocks the substrate-binding site (Figs. 2C and 4F). The side chain of Tyr412 is positioned in a hydrophobic pocket that is lined by Leu403, L406 and Met407. Thus, phosphorylation of Tyr412 destabilizes the I₂ state by removing the negatively charged pTyr412 from the pocket (Fig. 4F). To quantitatively measure the effect of Tyr412 phosphorylation on the energetics of the ensemble, we used the Abl^I2M variant. The population of I₂ in Abl^I2M is 90%, whereas it is only 10% in Abl^I2M-pY412 (Fig. 4F). Thus, phosphorylation of Tyr412 stabilizes the active state by ~3.0 kcal mol⁻¹ (Fig. 4C) and can activate even Abl variants that exist predominantly in the inactive state.

Allosteric regulation of Abl kinase by modulation of the conformational ensemble

Similarly to the Src family kinases (3), the docking of the SH3-SH2 regulatory module onto the kinase domain of Abl suppresses its activity (52). Because no structural data are available on an SH3-SH2-kinase Abl fragment without an inhibitor bound, the structural changes to the kinase domain elicited by the SH3-SH2 module are not known. We prepared and studied by NMR Abl fragments that encompass the SH3-SH2 module and the kinase domain, which for simplicity we refer to as the full kinase (Abl^FK). The presence of the SH3-SH2 module in Abl^FK increases the population of the I₂ state ~6-fold, from 6% in the isolated kinase domain to 34% in Abl^FK (Fig. 5A). As shown previously (77), only ~40% of the molecules in Abl^FK are found in the fully assembled state, that is where the SH3-SH2 module is docked onto the back of the kinase, while ~60% of the molecules adopt a disassembled conformation. The allosteric inhibitor GNF5, which was previously shown to promote the assembled conformational state (32, 77), increased the population of the assembled Abl^FK state to ~95% (77). NMR data on Abl^FK-GNF5 indicate that GNF5 binding stabilizes the I₂ state so that its population rises to 95% (Fig. 5A). Thus, our data clearly establish that allosteric stabilization of the I₂ state by the SH3-SH2 module is the structural basis for inhibition in Abl^FK (Fig. 5B). The I₁ inactive state is not stabilized by the docking of the SH3-SH2 module. The H415P substitution, which abrogates the I₂ state and promotes the I₁ state in the isolated kinase domain (Fig. 3D), shifts the equilibrium markedly towards the active state in Abl^FK-H415P thereby effectively counteracting the autoinhibitory mechanism (Fig. 5A,B). This explains the puzzling finding that the H415P substitution stimulates Abl kinase activity (58).

Fig. 5. Quantitative dissection of the effect of variants on the conformational ensemble of the full-length Abl kinase.

(A) Effect of the SH3-SH2 regulatory module, the addition of the allosteric inhibitor GNF5, and the H415P substitution as measured by NMR. Abl^FK corresponds to the SH3-SH2-KD Abl fragment. Because Abl^FK is not sufficiently stable for ¹³C CEST experiments, we used ¹H-¹³C–correlated spectra to quantitate the populations. For the isolated Abl kinase domain we know from CEST experiments that the population of the I₁ state is 6%. However, while we can conclude from the ¹H-¹³C–correlated spectra that the population of the I₁ state does not increase in Abl^FK and its variants, we cannot conclude if it is depleted. Thus, the populations reported for Abl^FK variants are only for the active and the I₂ states. (B) Schematic of Abl^FK showing the equilibrium between a disassembled conformation, wherein the regulatory module is not bound to the kinase domain, and an assembled conformation, wherein the regulatory module is docked onto the back of the kinase domain. In the disassembled conformation the kinase adopts the active state whereas in the assembled one it adopts the I₂ state. The populations of the active and I₂ states in the Abl variants studied are indicated. (C) Schematic of Abl^FK summarizing the effect of the gatekeeper T334I substitution, which forces Abl to adopt the active state even when the regulatory module is docked onto the kinase. ¹H-¹³C–correlated spectra showing the M362 methyl resonance. M362 is located at the interface between the SH2 and the kinase domain and provides a sensitive probe for the assembled conformation. (D) Superposition of the structures of the Abl I₂ state and the Abl^T334I–axitinib complex (PDB ID 4TWP) highlights the steric clash between the bound inhibitor and the A-loop in the I₂ state, which explains the higher affinity of the inhibitor for the T334I variant. (E) Effect of Tyr412 phosphorylation on the populations of the active and I₂ states of Abl^FK in complex with the allosteric inhibitor GNF5 measured by NMR. (F) Effect of the extended conformation, wherein the SH2 domain docks on the top of the N-lobe, on the populations of the active and I₂ states of Abl measured by NMR. The structure of the Abl^FKΔSH3 fragment (magenta; PDB ID 4XEY) is superimposed on the structure of the Abl I₂ state (orange).

Quantitative information on the conformational ensemble of Abl^FK provides a tool to assess the effect that various factors, such as the gatekeeper, phosphorylation, and allosteric inhibitors exert on the kinase. As shown above for the isolated Abl kinase domain, the T334I gatekeeper substitution stimulates Abl activity by shifting the equilibrium towards the active state (Fig. 4C,D). In Abl^FK, the T334I substitution promotes the active state even in the presence of the GNF5 allosteric inhibitor (Fig. 5C). Therefore, this single mutation is capable of activating Abl even when the kinase is held almost entirely in its assembled state (Fig. 5C). This finding is consistent with and explains the observation that the T334I variant markedly stimulates kinase activity even in the myristoylated form of Abl (62, 69). Our data also explain why allosteric inhibitors are not effective against the T334I gatekeeper substitution (78, 79), despite the fact that Ile334 is located away from the allosteric pocket. Our finding that Abl^FK-T334I adopts predominantly the active state is consistent with and accounts for the ~10-fold higher potency of axitinib, which binds selectively to the active state of Abl kinase, in inhibiting Bcr-Abl^T334I compared to Bcr-Abl (48), as the latter exists predominantly in the I₂ state and is not compatible with axitinib binding (Fig. 5D). Similar to the effect exerted by T334I, phosphorylation of the A-loop at Tyr412 counteracts the strong autoinhibitory mechanism in Abl^FK-GNF5 by destabilizing the I₂ state (Fig. 4F) and promoting the active state (Fig. 5B,E). This finding explains the much lower affinity of imatinib for the phosphorylated (pTyr412) Abl (80).

An alternative to the fully assembled, downregulated state of Abl^FK is an experimentally observed conformational state wherein the SH2 domain docks on the top of the N-lobe, thus giving rise to an extended conformation (77, 81–83). This structural arrangement is known to stimulate the activity of Abl and has been associated with increased leukemogenic activity (58, 82, 84). The structural basis for this activity stimulation is not understood. To quantitatively assess the effect, we studied by NMR the Abl^{FKΔSH3-I2M/T231R} variant, which includes the I2M triple mutant that populates predominantly the I₂ state, lacks the SH3 domain, which promotes the assembled conformation, and features the T231R substitution, which is thought to stabilize the interface between the SH2 and the N-lobe (58, 77). The results show that docking of SH2 on the N-lobe of the kinase promotes the active state by a factor of ~4, corresponding to an energetic contribution of ~1.2 kcal mol⁻¹ (Fig. 5F). In the context of Abl^FK, stabilization of the SH2−N-lobe interface is thus expected to increase the population of the active state from ~35% to ~80% (Fig. 4C). Superposition of the structure of an Abl fragment crystallized in the extended conformation (83) with the structure of the Abl I₂ state shows that the surface of the N-lobe in the I₂ state is not compatible with SH2 binding (Fig. 5F). A steric clash between the N-lobe Glu294 and the SH2 His233 and Tyr234 forces the loop connecting the αC helix and β3 strand to retract thereby disrupting the ion pair between Glu274 and Lys293 (Fig. 5F), which is essential for the stabilization of the stretched conformation of the P-loop observed in the Abl I₂ state (Fig. 3F). Taken together, our data show that binding of the SH2 domain on the top of the N-lobe disrupts the αC helix/P-loop conformation in the I₂ state and thus shifts the equilibrium towards the fully active state. Of note, the T231R mutation has been identified as a patient-derived imatinib-resistant variant. Our data explain how the T231R substitution, which strengthens the SH2−N-lobe interface, confers resistance to imatinib (58).

pH changes redistribute the populations between the inactive states

Given the distinct environment of Asp400 in the three conformational states (Fig. 6A–C), we asked whether subtle changes in the electrostatic environment coupled to variations in Asp400 pKa could affect the conformational ensemble and thus the kinase activity. We measured the populations of the three conformational states of Abl as a function of pH (Fig. 6D,E) and found that even relatively small changes in the pH can have a substantial effect on the relative stability of the two inactive states. Specifically, although at pH 7.1 the I₁ and I₂ states have the same population, at pH 6.5 the I₁ state is several-fold higher than the I₂ (Fig. 6F). Conversely, at pH 7.7 the I₂ state is much more stable than the I₁ state. Because of the high energy (85) required to bury the charged side chain of Asp400 in the hydrophobic pocket in the I₁ state, Asp400 needs to be protonated (86). This is in agreement with our observation that lower pH values favor the I₁ state. Of interest, pH changes within this range (6.5–7.7) do not seem to affect the relative stability of the active vs. the inactive states but rather only redistributes the population within the two inactive states.

Fig. 6. pH effect on the Abl conformational ensemble and activation energy of the DFG transition.

(A to C) Close-up view of the DFG motif conformation in the (A) active state, (B) the I₁ state and (C) the I₂ state. The positioning of Asp400 of the DFG motif differs among the various conformational states. In the active state Asp400 is exposed to the solvent whereas Phe401 is buried inside a hydrophobic pocket. The DFG motif flips 180° as it transitions from the active to the I₁ state with the two residues swapping positions: in the I₁ state Asp400 is now positioned inside the hydrophobic pocket while Phe401 is exposed to the solvent. Although the DFG motif remains in the “out” conformation in the I₂ state, the A-loop undergoes a major rearrangement and as a result Asp400 is found in a polar environment, similarly to the active state, and Phe401 is buried inside the hydrophobic pocket located in the nucleotide binding site. (D) ¹³C CEST profiles of representative residues as a function of pH. The results show that lower pH (pH 6.5) stabilizes selectively the I₁ state while higher pH (pH 7.7) stabilizes selectively the I₂ state. (E) ¹H-¹³C–correlated spectra of Abl^M309L and Abl^M309L/H415P variants as a function of pH. In Abl^M309L/H415P at pH 7.7 there is no detectable inactive state because the H415P substitution eliminates the I₂ state and the higher pH depletes the I₁ state. (F) Populations of the I₁ and I₂ states as a function of pH. The relative populations of the I₁ and I₂ states fluctuate antagonistically but the ratio of populations between the active state and the inactive states collectively is not affected by changes in pH within this range. (G) ¹³C CEST profiles of Abl^H415P as a function of temperature. (H) Arrhenius plot of the k_ex measured by the CEST experiments in (G) for the three temperatures indicated used to determine the activation energy of the DFG transition from the active to the I₁ inactive conformation. (I) Energy diagram indicating the activation enthalpy (~36.4 kcal mol⁻¹) of the DFG flip from DFG-in to DFG-out.

Activation energy of the DFG transition

The kinetics of the DFG flip determine how fast a kinase is activated or inhibited (86, 87) as it toggles between active and inactive states in response to stimuli. The kinetics of the DFG flip may also determine the rate of ADP release (74), which is typically the rate-limiting step in kinase catalysis (73). To measure the activation energy of the DFG transition, which governs its kinetics, we measured the exchange rate (k_ex) between the active state and the I₁ state as a function of temperature (Fig. 6G,H). The data showed that the activation energy of the DFG transition from the active to the inactive state is ~36 kcal mol⁻¹ (Fig. 6F). This large energy barrier indicates that the DFG flip in Abl kinase entails pronounced structural rearrangement in the transition state and explains why the DFG flip is associated with intrinsically slow kinetics.

DISCUSSION

Here we describe in structural and energetic terms the transition of Abl kinase between its active and two inactive conformational states. All of these states are intrinsic to the isolated kinase domain and they determine how the kinase responds to regulatory stimuli, substrate and ligand binding, allosteric interactions, and post-translational modifications. Abl kinase populates two distinct inactive states that are quite different from each other. Whereas the DFG motif is in the “out” conformation in both I₁ and I₂ states, the A-loop and the αC helix adopt very different conformations. Thus, there are various mechanisms and pathways for a kinase to switch to an inactive conformation.

The Abl I₂ state, which adopts a fully inactive conformation is selectively stabilized in the full-length kinase (Fig. 5B). The biological significance of the I₁ state is not apparent. Although our data suggest that I₁ lies in the pathway between the active and the I₂ state, depletion of the I₁ state has no effect on the population of the I₂ state (Fig. 2E) and therefore is likely not an obligatory intermediate species. Moreover, a single mutation (H415P) that abrogates the I₂ state converts the kinase to its fully active conformation with the I₁ state being very poorly populated even in the full length Abl kinase. That is, the inhibitory mechanism in Abl imposed by the docking of the regulatory module onto the kinase operates only on the I₂ state and not the I₁ state (Fig. 5A,B). Similarly, mutations in the DFG motif that inhibit the kinase exert their effect by selectively stabilizing the I₂ state and not the I₁ (Fig. 4E). It is possible that the I₁ state is a by-product of the evolutionary process that established the active and the I₂ state, the two functional states. Although conformational states such as the I₁ may not be associated with a known biological function they could be leveraged for the design of selective inhibitors. Characterization of a number of kinases employing the approaches described here could provide evidence of the existence of low-populated inactive conformational states in the kinome.

The discovery and structural insight into the Abl I₂ state as the one to which imatinib selectively binds provides the mechanistic basis for explaining a number of patient-derived imatinib-resistance variants located at sites remote to imatinib. Our data demonstrate that all of these amino acid substitutions confer resistance to imatinib by depleting the population of the I₂ state. The I₂ state can be destabilized by altering the arrangement of any of several structural elements in this state, such as the A-loop, the P-loop and the αC helix, either directly or allosterically (Figs. 4, 5F and fig. S10). These mutations decrease the affinity not only of imatinib but also of other inhibitors, such as nilotinib, ponatinib and rebastinib (28), that also select the Abl I₂ state (fig. S4). The structural data of the Abl I₂ state reported here provides an opportunity for optimizing imatinib analogs for higher affinity.

A question that remains largely unaddressed is how many different inactive conformations intrinsic to kinases are present in the kinome (18). This is significant not only for better understanding the variety of mechanisms used by kinases to regulate their activity but also for the design of selective inhibitors. Only a limited number of unliganded kinases have been crystalized in an inactive conformation and only in a fraction of them the intact A-loop, P-loop and αC helix are all visible (65). Among kinases with a known structure, the Abl I₁ state is unique. The Abl I₂ state is similar to the structure that the insulin receptor kinase (IRK) adopts in its inactive, non-phosphorylated form (88). Superposition of the structures of the Abl I₂ state with that of IRK highlights the striking similarities in the disposition of the A-loop, the DFG motif, the P-loop and the αC helix (fig. S12A). Structural and sequence analysis of the two kinases indicated that the residue at position 408 in Abl (1158 in IRK) may be crucial in stabilizing the closed conformation of the A-loop. The single T408Y substitution in Abl increases the I₂ state population by more than an order of magnitude, from 6% to 70% (fig. S6). IRK has a Tyr in this position (fig. S12A). Four more receptor kinases, TrkA, TrkB, IGF1R and MuSK, adopt a similar structure in their inactive state (fig. S12B). All of them feature a Tyr residue in this position. Conversely, three receptor kinases, FLT3, KIT and FMS, that also adopt a similar structure in their inactive state, do not bear a Tyr in this position; however, all of these three kinases require the presence of the juxtamembrane (JM) domain to stabilize their inactive state (fig. S12B). It should be noted that these receptor tyrosine kinases must adopt an inactive conformation in their resting state (89) and thus have evolved to stabilize such a conformation. Taken together, these findings underscore two key observations. First, a structurally similar inactive state among different kinases can be elicited by various mechanisms. For example, Abl accomplishes this by the allosteric action of the regulatory module. Some kinases, such as IRK, feature a crucial Tyr residue in the A-loop that directly stabilizes the inactive state. Other kinases, such as KIT, require the docking of the JM domain against the A-loop. Second, several kinases, both in the receptor and non-receptor tyrosine kinase families, share a structurally similar inactive state. This observation suggests that there might be a limited number of structurally-divergent inactive states intrinsic to kinases (18). More structural data on the intrinsic inactive conformations of kinases in the absence of inhibitors are needed to discover the full repertoire of their architecture and the mechanisms used to induce these conformations.

Our findings reveal additional mechanisms, other than allosteric interactions by intra-molecular domains and direct binding of inhibitory proteins, that kinases use to induce their inactive states. The regulatory spine can switch the kinase between its active and inactive states even with conservative changes in the amino acid residues that make up the spine. In Abl, the double substitution M309L/L320I in the spine switches the kinase from a fully active state (p_A ~90%) to an almost fully inactive state (p_A ~8%) (Fig. 4B). This finding explains why many oncogenic mutations map on the regulatory spine (63). Another mechanism for a kinase to stabilize its inactive state is to feature an amino acid other than Phe in the DFG motif (Fig. 4E). Finally, our data highlight and explain how a single mutation at the gatekeeper position can completely neutralize autoinhibitory mechanisms to give rise to a fully activated form of a kinase.

Materials and methods

Expression and preparation of proteins

Abl constructs used in this work were cloned from full-length Abl isoform 1b. The coding sequences were cloned into the pET16b vector for expression as maltose binding protein (MBP)-His₆ fusion proteins including a tobacco etch virus (TEV) protease cleavage site. Various Abl mutants were generated using the QuikChange site-directed mutagenesis kit (Agilent) and their sequences were confirmed by DNA sequencing. The following Abl constructs were prepared: Abl^FK (residues 1–534 or 83–534), Abl kinase domain (referred to as Abl; residues 248–534), Abl^G269E, Abl^Q271H, Abl^Y272H, Abl^M309L, Abl^E311K, Abl^L320I, Abl^T334I, Abl^F378V, Abl^F401V, Abl^F401L, Abl^F401Y, Abl^T408Y, Abl^H415P, Abl^G269E/R405E, Abl^G269E/M309L, Abl^Y272H/T408Y, Abl^M309L/E311K, Abl^M309L/L320I, Abl^M309L/F401L, Abl^M309L/T408Y, Abl^M309L/H415P, Abl^E311K/T408Y, Abl^L320I/T408Y, Abl^F378V/T408Y, Abl^L389M/T408Y, Abl^T408Y/H415P, Abl^{G269E/M309L/T334I,} Abl^{M309L/L320I/T334I,} Abl^{G269E/M309L/T408Y,} Abl^{G269E/E274V/M309L/T408Y,} Abl^FK-M309L, Abl^FK-T334I, Abl^FK-T408Y, Abl^FK-H415P, Abl^FKΔSH3 (residues 138–534), Abl^{FKΔSH3-T231R, and} Abl^{FKΔSH3- T231R/G269E/M309L/T408Y}. Abl constructs were expressed and purified as described previously (77). Unlabeled protein samples were produced in cells grown in Luria-Bertani (LB) medium at 37 °C in the presence of ampicillin (100 μg ml⁻¹) to an absorbance at 600 nm (A₆₀₀) of 0.8. Protein induction was induced by the addition of 0.2 mM isopropyl-β-D-1-thiogalactopyranoside (IPTG) and cells were allowed to grow for 48 h at 16 °C. Cells were harvested at A₆₀₀ ~1.8 and resuspended in lysis buffer (50 mM Tris-HCl, 500 mM NaCl, 1 mM phenylmethylsulfonyl fluoride (PMSF), 5 mM β-mercaptoethanol (BME), pH 8). Cells were disrupted by a high-pressure homogenizer and centrifuged at 50,000 × g. Proteins were purified using Ni Sepharose 6 Fast Flow resin (GE Healthcare), followed by tag removal by TEV protease at 4 °C (incubation for 16 h) and gel filtration using Superdex 75 16/60 or 200 16/60 columns (GE Healthcare). Protein concentration was determined spectrophotometrically at 280 nm using the corresponding extinction coefficient. CrkII was cloned into the pET42a vector, and expression and purification were performed as previously described (90). Hck kinase used for Abl Tyr412 phosphorylation was cloned into the pET16b vector and was expressed and purified using a similar protocol to Abl.

Inhibitors

All Abl inhibitors (Selleck Chemicals) used in this study were solubilized in DMSO-D₆ (Cambridge Isotope Laboratories) at a concentration of 10–50 mM and titrated to Abl solutions from these concentrated stocks.

ITC experiments

ITC experiments were performed on MicroCal iTC200 or Auto-iTC200 calorimeters (Malvern Instruments Inc.). ITC titrations for each Abl-inhibitor pair were typically performed at 25 °C in 25 mM sodium phosphate buffer (pH 7.1) including 75 mM NaCl and 2.0 mM tris(2-carboxyethyl)phosphine (TCEP). Proteins were purified by size exclusion chromatography using corresponding ITC buffers before use. Concentrations of protein and inhibitors were measured spectrophotometrically and by weight, respectively. Proteins were placed in the cell, while the inhibitors were in the syringe with concentrations in the range 8 to 60 μM and 80 to 650 μM, respectively. The small quantity of DMSO remaining in the inhibitor buffer after the transfer from the stock solution was matched in the protein buffer. All solutions were filtered using membrane filters (pore size, 0.45 μm) and thoroughly degassed. The data were fitted with Origin 7.0 (OriginLab Corporation).

NMR Sample Preparation

Isotopically labelled samples for NMR studies were prepared by growing the cells in minimal (M9) medium. Cells were typically harvested at A₆₀₀ ~1.2. U-[¹³C,¹⁵N]-labelled samples were prepared for the backbone assignment by supplementing the growing medium with ¹⁵NH₄Cl (1 g l⁻¹) and ¹³C₆-glucose (2 g l⁻¹) (CIL and Isotec). The ¹H–¹³C methyl-labelled samples were prepared as described (91, 92). For Phe and Tyr selective labeling, 50 mg l⁻¹of U-[²H] ε−¹H,¹³C-Phe and Tyr were used. NMR samples of Abl variants were typically prepared in 25 mM sodium phosphate buffer (pH 7.1) containing 75 mM NaCl, 3.0 mM BME, in 100% ²H₂O. The concentration of protein samples ranged from 30 μM to 300 μM. Abl-inhibitor complexes for NMR studies were prepared by mixing the protein at low concentration (~10 μM) with the inhibitor in the NMR buffer followed by concentration via ultrafiltration.

NMR Spectroscopy

All NMR experiments were acquired on Bruker 1.1 GHz, 900, 850, 800, 700 and 600 MHz spectrometers equipped with cryogenic probes. Spectra of unliganded Abl were typically collected at 10 °C whereas spectra for Abl-inhibitor complexes were collected at 10 or 20 °C. All NMR data were processed using NMRPipe (93) and analyzed using NMRFAM-Sparky (94). Backbone ¹H, ¹⁵N, and ¹³C resonance assignment for Abl variants was achieved using 3D HNCACB, CBCA(CO)NH, HNCA, and HNCO experiments. Sidechain methyl and aromatic resonances were assigned using 3D-¹³C,¹⁵N-NOESY-HMQC and ¹³C-HMQC-NOESY-HMQC spectra (95) aided by the MAGIC software (96).

CEST experiments

0.15 to 0.3 mM. The labeling scheme for the methyl-bearing residues was as follows: (i) [U-²H; Leu,Val-¹³CHD₂/¹³CHD₂], (ii) [U-²H; Iled1-¹³CHD₂], (iii) [U-²H; Met-¹³CHD₂], (iv) [U-²H; Ala-¹³CHD₂], and [U-²H; Thr-¹³CH₃]. A series of 2D spectra in a pseudo-3D mode were recorded with a mixing time (T_mix) of 500 ms and a weak B₁ radiofrequency fields of 15 and 25 Hz applied at the ¹³C dimension in steps of 20–25 Hz. A recycle time of 2.0 s was used in all experiments. Peak intensities were measured using NMRFAM-Sparky (94). CEST profiles were made by plotting the intensity ratios with and without the T_mix period.

CEST data fitting

CEST data were analyzed using ChemEx (https://github.com/gbouvignies/chemex), which numerically simulates the experiment by solving the Bloch-McConnell equations and minimizes the standard χ² equation to get best-fit exchange parameters (41, 97). Errors in peak intensity for the ChemEx input were measured from the background noise of the spectra. Uncertainties in the exchange parameters were estimated by Monte Carlo simulations. Fitting data in a three-state exchange process is typically more challenging. For this reason, we sought to assign the individual minor dips to the corresponding excited state (E₁ and E₂). The strong and opposite dependence of the populations of the two excited states on pH facilitated this assignment, further confirmed by the CEST data on the Abl^H415P variant, which eliminates the I₂ state. We first proceeded to fit the CEST data for the Abl^H415P variant in which the exchange process is a two-state. Methyls were included in a global fitting to a two-state model of conformational exchange $(A\stackrel{k_{ex}}{\iff }{E}_1)$ based on the following criteria: 1) no overlap with other peak; 2) minor dip could be discerned in CEST profile; 3) no affect by other local dynamics. The calculated k_ex and P_E1 from fitting were 46.8 ± 4.3 s⁻¹ and 12.1 ± 0.5 % respectively, with a reduced χ² (χ_red²) value of 1.38. Next, we fitted the wild-type Abl CEST data by globally fitting to a three-state exchange model as follows:

$$A\stackrel{k_{ex1}}{\iff }{E}_1\stackrel{k_{ex2}}{\iff }{E}_2$$ model 1

$$E_1\stackrel{k_{ex1}}{\iff }A\stackrel{k_{ex2}}{\iff }{E}_2$$ model 2

$$A\stackrel{k_{ex1}}{\iff }{E}_2\stackrel{k_{ex2}}{\iff }{E}_1$$ model 3

The χ_red² values obtained from models 1, 2, and 3 were 1.15, 1.29, and 1.41, respectively, suggesting model 1 is the most appropriate model to fit the data. The exchange parameters obtained from model 1 were k_ex1 = 12.3 ± 2.1, k_ex2 = 145.5 ± 22.6 s⁻¹, p_E1 = 10.6 ± 1.2%, p_E2 = 9.6 ± 1.1%, parameters obtained from model 2 were k_ex1 = 14.7 ± 3.6, k_ex2 = 18.7 ± 2.7 s⁻¹, p_E1 = 9.9 ± 1.7%, p_E2 = 7.0 ± 0.7%, and parameters obtained from model 3 were k_ex1 = 11.2 ± 2.4, k_ex2 = 136 ± 23.2 s⁻¹, p_E1 = 21.2± 4.0%, p_E2 = 9.2 ± 1.2%. Populations of the excited states larger than ~10% were not consistent with the CEST profiles. Furthermore, we couldn’t detect any peaks of the excited states in ¹H-¹³C-correlated spectra of Abl recorded over a period of several days, indicating lower populations. Because of the increased number of fitting parameters in a three-state exchange model data fitting can yield multiple minima. We thus globally fit the data to both model 1 and model 2 by fixing k_ex1 at 46.8 s⁻¹, as determined by the two-state CEST data fitting on Abl^H415P. The χ_red² values were 1.42 and 1.41 for model 1 and model 2, respectively. The exchange parameters obtained from model 1 were k_ex1 = 46.8 ± 4.3, k_ex2 = 88.7 ± 13.5 s⁻¹, p_E1 = 5.9 ± 0.1%, p_E2 = 6.1 ± 0.7%, and parameters from model 2 were 46.8 ± 4.3, k_ex2 = 19.8 ± 2.9 s⁻¹, p_E1 = 5.4 ± 0.1%, p_E2 = 6.8 ± 0.7%. Model 1 is in agreement with the population shift observed in Abl^H415P in which E₂ is eliminated and the population of E₁ is doubled.

Measurement of the populations of the Abl conformational states

Populations of the active (A) and the two inactive states (I₁ and I₂) in the several Abl variants studied in this work were measured by fitting the CEST data as described above and by integrating the resonances corresponding to each state using ¹H-¹³C–correlated spectra. Because the interconversion of states is slow on the NMR time scale a distinct set of resonances are present in a 2D correlated spectrum provided that its population is higher than ~5%. This latter approach was used for Abl variants that did not remain stable over a period of several days typically needed for recording the CEST data. It was successfully used to measure the populations in Abl^FK, which gave very poor CEST data because of low solubility and stability. Two variants were produced and tested for Abl^FK, one that encompassed residues 1–534 and the other residues 83–534. Both include the full regulatory module consisting of the SH3-SH2 tandem. Given that both variants gave very similar results, and because of the much higher stability of the 83–534 fragment, the latter one was primarily used. Changes in populations as a result of a mutation were expressed in free energy terms as ΔΔG= −RT(lnR_W–lnR_M) in kcal mol^–1, where R_W and R_M is the ratio of the populations of the two states in the wild-type and the mutant, respectively.

Structure determination

Structures were calculated from experimental methyl-methyl, methyl-aromatic, aromatic-aromatic, methyl-NH and aromatic-NH NOEs (Table S1). Dihedral angles and hydrogen bond restraints were derived from NOE patterns. Complete NOESY spectra assignment conducted iteratively from sparse restraint sets (98) using the automated NOESY cross-peak assignment protocol and dihedral angle dynamics simulated annealing within CYANA 3.98 (99). NOE peak assignments were manually curated and refined with several CYANA runs. The lowest target function CYANA-derived structure ensemble was subjected to an all-atom restrained molecular dynamics energy refinement in water bath using CNS (100). The active conformation is the predominant (~90%) one populated by Abl in solution and hence its structure was determined using standard approaches. By contrast, I₁ and I₂ are marginally stable with either one a population of 6%. As described in the main text, to structurally characterize these states by NMR we sought to increase their population by amino acid substitutions to shift the equilibrium towards these states without eliciting other structural changes. Because we had determined the chemical shifts of the I₁ and I₂ states by CEST, it was straightforward to assess the effect that the amino acid substitutions had on their structure. The M309L/H415P double substitution was found to markedly stabilize the I₁ state with a population of ~50% (Fig. S5). NOESY data were then collected using this sample. Detailed chemical shift analyses showed that the structure of the I₁ state is similar to the one of Abl in complex with the PD173955. Especially in protein parts that are remote to the catalytic site, the A-loop and the αC helix, the two structures were essentially identical. For this reason, we used restraints from the Abl– PD173955 for areas that do not change in Abl I₁, whereas we determined de novo the structure of Abl I₁ in all other using the NMR restraints collected on the Abl^M309L/H415P variant. To determine the structure of the Abl I₂ state we followed a similar approach. We initially collected a large number of NOEs on the Abl^M309L/T408Y variant, which populates the I₂ state at ~85%. Using the preliminary structural data, we gained more insight into the effect of several other mutants and some of them were observed to further increase the population of the I₂ state, but even more importantly, they exhibited much sharper peaks and were more stable. The Abl^{G269E/M309L/T408Y} variant (also referred to as the I2M variant in the main text) increases the I₂ population to 90% and was amenable to higher resolution NMR characterization.

Abl Phosphorylation by Hck

To prepare Abl specifically phosphorylated at the Tyr412 residue we used Hck kinase. 2 μM Hck was auto-phosphorylated in 50 mM Tris buffer (pH 7.5) containing 2 mM ATP, 10 mM MgCl₂, 50 mM NaCl, and 3.0 mM BME for 1.5 hours at 22 °C. Abl (~5 μM) was incubated with 250 nM of auto-phosphorylated Hck for 14–16 hours at 15 °C, followed by extensive dialysis against 50 mM Tris buffer (pH 7.5), 50 mM NaCl, 3.0 mM BME to remove ATP, ADP and MgCl₂ and ion exchange purification (Source™ 15Q, GE Healthcare) to separate the phosphorylated Abl from any traces of unphosphorylated protein. Abl phosphorylation was monitored by western blot analysis with an antibody specific to Abl pTyr412 (2865S, Cell Signaling Technology). Western blot indicated phosphorylation of Abl by Hck led to the phosphorylation of Tyr412 and mass spectrometry and NMR spectra confirmed that samples were homogeneous.

Kinase assays

The kinase assays were conducted in 50 mM Tris (pH 7.5), 0.5 mM MgCl₂, 100 mM KCl, and 3.0 mM BME at room temperature with CrkII as substrate. 10 μM CrkII was incubated with 0.2 μM Abl in the reaction buffer, with the addition of ATP to a final concentration of 0.1 mM to initiate the reaction. Reactions were stopped at 0, 30, 60, 120, 180, 300, and 600 sec by adding SDS-containing loading buffer. Proteins were resolved in a 4–20% SDS-PAGE gel (Bio-Rad Laboratories) and probed by western blot with antibody to CrkII pTyr221 (3491S, Cell Signaling Technology). All blots were loaded with SuperSignalTM West Pico Plus Chemiluminescent Substrate (34577, Thermo Scientific), visualized using an Amersham Imager 600 (GE Life Sciences) and quantified with ImageJ. Activity of Abl proteins on CrkII are shown as the n-fold activation above the activity at 0 s for each time point.