Cumulative mechanism of several major imatinib-resistant mutations in Abl kinase

Significance

One obstacle for the prolonged success of the wonder drug imatinib in leukemia has been the emergence of resistance mutations within the Abl kinase domain. Here, we elucidate the molecular mechanism for resistance of three major mutations in patients treated with imatinib. Unpredictably, the single-site resistance mutations act via a cumulative effect of an only modest decrease in drug affinity, combined with an increase in enzyme activity and altered substrate affinity/cooperativity. Strikingly, this combination indeed leads to at least an order-of-magnitude higher IC₅₀ values for imatinib, but only under cellular ATP concentrations. Our findings settle a longstanding controversy, and concepts found here are likely to play a role in drug resistance in other targets.

Abstract

Despite the outstanding success of the cancer drug imatinib, one obstacle in prolonged treatment is the emergence of resistance mutations within the kinase domain of its target, Abl. We noticed that many patient-resistance mutations occur in the dynamic hot spots recently identified to be responsible for imatinib’s high selectivity toward Abl. In this study, we provide an experimental analysis of the mechanism underlying drug resistance for three major resistance mutations (G250E, Y253F, and F317L). Our data settle controversies, revealing unexpected resistance mechanisms. The mutations alter the energy landscape of Abl in complex ways: increased kinase activity, altered affinity, and cooperativity for the substrates, and, surprisingly, only a modestly decreased imatinib affinity. Only under cellular adenosine triphosphate (ATP) concentrations, these changes cumulate in an order of magnitude increase in imatinib’s half-maximal inhibitory concentration (IC₅₀). These results highlight the importance of characterizing energy landscapes of targets and its changes by drug binding and by resistance mutations developed by patients.

Kinases are key enzymes in many crucial cellular-signaling processes and, when aberrant, often lead to the development of cancer (1–3). Thus, it is not surprising that they have been major targets for modern drug design. However, a major obstacle in designing orthosteric drugs for protein kinases is their high structural similarity. As each kinase catalyzes the same phosphoryl-transfer reaction between the γ-phosphate of adenosine triphosphate (ATP) and its protein substrate (usually a tyrosine, serine, or threonine residue), designed orthosteric drugs are subject to off-target effects. Yet, highly selective, orthosteric drugs are possible, as shown by the success story of the anticancer drug imatinib, a potent inhibitor of Abl kinase. Imatinib binds with 3,000-fold higher affinity to its target Abl kinase relative to its closest homolog, with a high degree of sequence (54%) and structural similarity, Src kinase (4). This high specificity of imatinib for Abl has made it an effective treatment for chronic myelogenous leukemia (CML), a pathological condition that is caused by a fusion product of the gene for Abl (located on chromosome 9) with the BCR gene on chromosome 22, thereby forming a shortened chromosome termed the Philadelphia chromosome. As a consequence of the formation of this Philadelphia chromosome, parts of the regulatory N terminus of Abl are replaced with BCR, which then leads to the expression of a constitutively active BCR-Abl (5–9). Since imatinib is one of the few drugs on the market that is highly selective with minimizing side effects (10), it has been the frontline treatment for most CML patients and has proven to be very effective. Imatinib is also highly effective against the kinase c-kit in gastrointestinal stromal cancer (11).

However, one problem BCR-Abl–positive patients have faced over last decade has been the emergence of resistance to imatinib (12, 13). Both BCR-Abl–independent and BCR-Abl–dependent mechanisms are possible ways patients can acquire resistance toward drug treatment. Examples of BCR-Abl–independent mechanisms include mutations in transporters that mediate drug influx or persistence of drug-resistant cancer stem cells that retain their ability to self-renew (14, 15). Alternatively, BCR-Abl–dependent mechanisms involve either BCR-Abl amplification, mutations in the Abl’s regulatory domains that exert their resistance via allostery (16), or mutations in the Abl kinase domain, which are thought to modulate the kinase–drug interaction (17, 18). The development of mutations in the Abl kinase domain is the most commonly reported mechanism for resistance toward imatinib treatment. For many kinase-domain mutations, the cause for resistance has been assumed to be a direct interference with drug binding (19–21). While this hypothesis is compelling, no quantitative analysis has been performed, and the exact rationale for imatinib resistance remains obscure. Food and Drug Administration-approved second- and third-generation inhibitors are effective against several of the imatinib-resistant mutants, but bear the same problem of becoming ineffective due to new resistance mutations. A promising allosteric inhibitor Abl001 that binds to the myristoylation site is in clinical trial (22).

Here, we experimentally investigated three mutations that are among the most commonly found resistance mutations in patients treated with imatinib. We found that, contrary to previous hypotheses for these mutations, where resistance was suggested to originate simply by disrupting the kinase–drug interaction, it is a combination of multiple effects that lead to resistance. Surprisingly, single mutations result only in small changes in drug affinity, but simultaneously alter Abl’s turnover rate, affinities of ATP, and its target substrate. The cumulative effect of all four effects reflects a substantially changed energy landscape and explains the resistance to imatinib by these mutations.

Results

Resistance Mutations Overlap with Dynamic Hot Spots Essential for Imatinib Selectivity.

Previously, we had studied the origin of imatinib specificity using ancestral sequence reconstruction. By studying the evolutionary trajectory of Abl from its closest homolog, Src, we identified 15 key residues that are responsible for the tight affinity and high selectivity of imatinib to Abl (23). These 15 residues are distributed across the N-terminal lobe of the kinase (Fig. 1A) and were suggested to alter the flexibility of the P loop, resulting in the observed conformational changes after imatinib binding that are crucial for high affinity and selectivity (23, 24). Interestingly, when we now compared the location of these 15 residues with known resistance mutations found in cancer patients that were treated with imatinib, many of the resistance mutations overlapped with the 15 dynamic hot spots in Abl (Fig. 1).

Fig. 1.

Drug-resistance mutants in Abl kinase overlap with residues important for induced-fit step in imatinib binding. (A) Wild-type Abl bound to imatinib shown in black [Protein Data Bank ID code 1OPJ (33)]. (A, Upper) Residues important for tight binding of imatinib (23) are highlighted as red spheres. Dark red residues are overlapping with resistance mutations that have been isolated from cancer patients. The three mutations investigated in this study are highlighted with colored arrows and labeled. (A, Lower) Common kinase domain resistance mutations isolated from patients treated with imatinib are shown as brown spheres. Dark brown residues are mutations that overlap with the 15 residues important for imatinib binding. (B) Multiple sequence alignment of Src (weak imatinib binder; top rows), the ancestral protein (ANC-SA; middle rows) previously characterized to be an intermediate imatinib binder (23), and Abl (tight imatinib binder; bottom rows). Coloring of residues corresponds to A.

Intrigued by the observation that nature seems to escape drug treatment by amino acid changes in these dynamic hot spots that are responsible for the high affinity of imatinib to Abl, we set out to investigate the underlying molecular mechanism for their drug resistance. Three imatinib resistance mutations were selected (Fig. 1A) with the following criteria in mind: 1) they are major resistance mutations found in patients; 2) they are from the subset of the 15 dynamic hot spots; and 3) their mechanism is not understood, with either controversial or inconclusive prior literature studies (19, 20, 25, 26). F317L is located at the back of the P loop, and the phenylalanine was suggested to be important for van der Waals interactions and polar stacking with the aromatic ring of imatinib (21, 24, 27–30). G250E is located in the P loop, and its mechanism for resistance has not been well characterized (31, 32). Y253F, part of the P loop, forms a hydrogen bond with N322, which is thought to stabilize the kinked P-loop conformation in the Abl drug-bound form (20, 23, 33–35). There are additional conflicting reports for the Y253F mutant indicating a possible effect on catalytic activity; both an increase (36, 37) and no effect (38, 39) on catalytic activity was reported.

Thermodynamics and Kinetics of Imatinib Binding to Resistance Mutants.

While there have been reports that epidermal growth factor receptor (EGFR) resistance mutations can lead to an increase in kinase activity or altered ATP affinity (40), the bulk of the literature on resistance mutations observed for Abl come to the conclusion that resistance mutations in Abl’s kinase domain mainly work by affecting binding of the drug to the kinase. To test this assumption, we first measured the affinity of imatinib to the three resistance mutants of Abl via intrinsic tryptophan fluorescence at 5 °C (Fig. 2A; see SI Appendix, Fig. S1 for data at 25 °C). Since formation of the BCR-Abl fusion product fully removes the autoinhibition by the N terminus of Abl, we decided to use the Abl kinase domain only constructs as a suitable mimic of the fusion product. To our surprise, none of these mutations decreased binding by more than threefold (K_D = 50 ± 10 nM, 78 ± 12 nM, and 54 ± 4 nM for F317, G250E, and Y253F, respectively) compared to wild type (K_D = 26 ± 4 nM). It seemed unlikely that such minor changes in overall affinity would result in resistance against imatinib and opposes the resistance mechanism proposed in the literature. What are we missing? The measured binding affinities are overall K_D values comprising all microscopic steps for drug binding (Fig. 2B). As shown previously, the high specificity of imatinib toward its target Abl stems from an induced-fit step after drug binding, and not from the conformational-selection mechanism for aspartate–phenylalanine–glycine-in/out, or subsequent binding (Fig. 2B) (35). The induced-fit step entails kinking of the P loop for extensive interactions between the protein and the drug. Since all three resistance mutations overlap with the residues important for this P-loop kinking (23), the expectation is that these mutations alter the conformational dynamics after drug binding.

Fig. 2.

Thermodynamics and kinetics of imatinib binding to wild-type (WT) and mutant forms of Abl showing imatinib resistance. (A) Macroscopic dissociation constant (K_D), measured by monitoring intrinsic tryptophan fluorescence at 5 °C. (B) Imatinib binding scheme to Abl comprising a conformational selection and an induced-fit step after drug binding. (C) Representative traces for stopped-flow fluorescence experiments at 5 °C illustrating double-exponential decay for imatinib binding (data in color, with double-exponential fit shown in black). a.u., arbitrary units. (D and E) Plot of observed rate constants for the fast binding step (k_bind^obs) (D) and the slow induced-fit step (k_conf^obs) (E) versus the concentration of imatinib. (F) Dissociation of imatinib from Abl wild-type, F317L, G250E, and Y253F. (G) Comparison between measured overall K_D (K_D^obs) from A and binding affinities calculated from microscopic rate constants (K_D^kin); see also Table 2. (n = 3 experiments; mean ± SD.)

To test this hypothesis, stopped-flow experiments were performed to measure the kinetics of association and dissociation. In this experiment, Abl was rapidly mixed with varying concentrations of imatinib, and quenching of intrinsic tryptophan fluorescence due to drug binding was measured over time (Fig. 2C). All experiments were performed at 5 °C, since the initial binding of the drug to enzyme was too fast to be observed at 25 °C, as noted earlier (35). Kinetic traces were recorded for 1 and 120 s to capture both the fast-binding and subsequent slower conformational changes. After correction for photobleaching, the short time traces were fit to a single-exponential function (SI Appendix, Figs. S2–S5, Insets) with the observed rate constants for the mutants increasing linearly with imatinib concentration corresponding to the second-order binding step (Fig. 2D). The bimolecular rate constant (k_on,Binding) obtained from the slope and dissociation of imatinib (k_off,Binding) determined from the intercept (Fig. 2D) were comparable to wild type (k_on,Binding = 1.0 ± 0.1 μM⁻¹⋅s⁻¹, k_off,Binding = 25 ± 4 s⁻¹) The observed rate constants for the second slow phase obtained from the longer time traces (SI Appendix, Figs. S2–S5) were nonlinearly dependent on imatinib concentration, with the plateau defined by the sum of forward and reverse rates of the induced-fit step (k_fwd + k_rev; Fig. 2E).

The dilution experiments were performed for the enzyme/drug complex to monitor the dissociation and complete the kinetic characterization of imatinib binding. Since imatinib binds tightly to Abl and the setup of our stopped-flow machine only allows for an 11-fold dilution, the experiments were performed on a fluorometer instead. The kinase was preincubated with twofold excess of drug at 5 °C for 30 min and rapidly diluted 30-fold into buffer. Dissociation of the complex, characterized by an increase in intrinsic tryptophan fluorescence, was monitored over a time of 1,600 s. The kinetics of this process was very similar for all protein forms (Fig. 2F and Table 1) and can be assigned to the k_rev of the induced-fit step.

Table 1.

Rapid kinetics of imatinib binding to wild-type and mutant forms of Abl

	Binding	Conformational change
WT	kon = 1.0 ± 0.1 μM−1⋅s−1	kfwd = 1.5 ± 0.1 s−1
	koff = 25 ± 4 s−1	krev = (17 ± 5) × 10−4 s−1
F317L	kon = 1.1 ± 0.1 μM−1⋅s−1	kfwd = 1.3 ± 0.1 s−1
	koff = 25 ± 3 s−1	krev = (25 ± 3) × 10−4 s−1
G250E	kon = 1.3 ± 0.1 μM−1⋅s−1	kfwd = 0.5 ± 0.1 s−1
	koff = 25 ± 4 s−1	krev = (23 ± 3) × 10−4 s−1
Y253F	kon = 1.2 ± 0.1 μM−1⋅s−1	kfwd = 0.9 ± 0.1 s−1
	koff = 25 ± 3 s−1	krev = (22 ± 3) × 10−4 s−1

Results from fitting of rapid kinetics from stopped-flow experiments (Fig. 2 and SI Appendix, Figs. S2–S5). Rate constants for the binding step of wild type (WT) and mutants are shown in the second column: physical binding (k_on) and dissociation of imatinib (k_off). Rate constants for slow conformational change after imatinib binding are shown in the last column (forward, k_fwd; and reverse, k_rev).

The kinetics experiments revealed that F317L, G250E, and Y253F altered the conformational dynamics in the drug-bound state and, in particular, the forward rate of the induced-fit step since the slow-off rate (k_rev) obtained from the dilution experiment was the same. This qualitatively agrees with what was observed for the well-characterized gatekeeper mutant reported earlier (23); however, the effect for these newly characterized mutants was much smaller. The qualitative agreement for F317L, G250E, and Y253F with the gatekeeper mutant suggests that changing the equilibrium of the induced-fit step might be a more general mechanism of how mutations alter binding of imatinib to the kinase. Taken together, our kinetics data (Fig. 2 C–F) enabled a rigorous testing of the underlying binding model (Fig. 2B) by comparing the measured overall K_D with the calculated macroscopic K_D from the individual rate constants. The excellent agreement between the experimental (K_D^obs) and calculated (K_D^kin) overall dissociation constants for imatinib (Fig. 2G and Table 2) verifies the binding mechanism.

Table 2.

Comparison between measured overall K_D^obs and the calculated macroscopic K_D^kin from kinetic scheme

		KDkin, nM	KDobs, nM
WT	KDbind = 25 ± 4 μM	27 ± 9	26 ± 4
	Keqconf = (11 ± 3) × 10−4
F317L	KDbind = 25 ± 4 μM	42 ± 9	50 ± 10
	Keqconf = (19 ± 2) × 10−4
G250E	KDbind = 25 ± 2 μM	88 ± 26	78 ± 12
	Keqconf = (46 ± 11) × 10−4
Y253F	KDbind = 21 ± 3 μM	53 ± 12	54 ± 4
	Keqconf = (26 ± 5) × 10−4

Comparison of the overall K_D calculated from kinetic data (K_D^kin) with the experimentally measured macroscopic K_D (K_D^obs). Kinetic data are used to calculate individual equilibrium constants for binding (K_D,^bind) and the induced-fit step (K_eq^conf). The overall, kinetic K_D^kin is calculated according to $K_{\mathrm{D}}^{\text{kin}}=K_{\mathrm{D}}^{\text{bind}}\ast K_{\text{eq}}^{\text{conf}}/\left(1+K_{\text{eq}}^{\text{conf}}\right)$ and can be compared to the macroscopic, experimentally determined value, K_D^obs (Fig. 2A) to validate the binding scheme. WT, wild type.

Effect of Resistance Mutation on IC₅₀ Values for Imatinib Depends Strongly on ATP Concentration.

These relatively small changes of the induced-fit step lad only to a twofold to threefold weaker overall affinity for imatinib. The small change in imatinib affinity for the resistance mutants alone is unlikely to be responsible for development of resistance. Why does drug affinity not define drug resistance? In patients, the effective outcome of imatinib treatment only depends on the drug’s ability to efficiently inhibit the kinase’s activity. In other words, what matters is how much drug is needed to reach wild-type level of activity for the mutated proteins in the cellular context. Therefore, we next measured the effect of imatinib on kinase activity at 25 °C, using conditions that are routinely used in the literature when reporting half-maximal inhibitory concentration (IC₅₀) values (50 μM ATP; Fig. 3A). The increase in IC₅₀ values for the mutant proteins relative to wild-type were the same as the increases in K_D values for imatinib (Fig. 3A). Strikingly, when the IC₅₀ experiments were repeated at cellular ATP concentration (5 mM), much larger effects were observed (Fig. 3B). The IC₅₀ values were 6- to 15-fold larger for the respective mutant proteins than wild type (Fig. 3C), potentially explaining why patients harboring any of these mutations are not responsive to imatinib treatment (34, 41–43). Since cancer cells have elevated intracellular ATP concentrations, we further tested whether the IC₅₀ values changed at 10 mM ATP (SI Appendix, Fig. S7). The effect on IC₅₀ values was very small, and the ratios between wild type and mutants were not affected.

Fig. 3.

Kinase inhibition by imatinib (IC₅₀) for resistance mutants exposes puzzle. (A and B) IC₅₀ curves were measured at low (50 μM; A) and cellular (5 mM; B) ATP concentrations using 2 mM peptide substrate at 25 °C. (A, Right and B, Right) IC₅₀ values are plotted showing the effect of ATP concentration on IC₅₀. (C) Comparison of changes in K_D (Fig. 2) with changes in IC₅₀ for the imatinib-resistant mutants of Abl relative (rel.) to wild type at low and high ATP concentrations. Errors are SD of triplicate experiments.

Quantitative Enzyme Kinetics Analysis of Resistance Mutations Reveals Underlying Mechanism for Resistance.

One possible explanation for the bigger difference in drug affinity (IC₅₀) at high ATP concentrations between wild type and mutants would be a mechanism where resistance mutations tighten ATP binding, thus making it harder for imatinib to outcompete the nucleotide at cellular concentrations. To determine if resistance mutations affect ATP binding affinity, we decided against a pure measure of ATP binding. In the cell, the kinase not only has to bind ATP, but also the second substrate, its target protein. Therefore, a full steady-state analysis of enzyme activity was performed to determine the binding of both substrates (K_D^Peptide and K_D^ATP), their cooperativity (α), and the catalytic turnover (k_cat). To extract these parameters, activity was measured at varying ATP/peptide concentrations, and curves were globally fit to the equation for a sequential random Bi–Bi mechanism (SI Appendix, Eq. 1 and Fig. 4A). The α parameter reports on cooperativity for substrate binding. It was shown previously that peptide and ATP exhibit negative cooperativity for wild-type Src and Abl [α > 1 negative cooperativity; α < 1 positive cooperativity (44)].

Fig. 4.

Resistance mutations cumulatively change the energy landscape of Abl kinase. Kinase activity for wild-type and resistant mutant forms of Abl at varying peptide and ATP concentrations (symbols, raw data; dashed line, fit to general velocity equation for a sequential random Bi–Bi reaction [see also SI Appendix, Eq. 1]). (B–E) The K_D for ATP (B), K_D for peptide substrate (C), kinase turnover rate (k_cat) (D), and cooperativity between substrates (α) (E) are altered by these individual mutations. (F) Fold change in IC₅₀ of mutant forms relative (rel.) to wild type (WT) at 5 mM ATP, dissected into the individual contributions from K_D imatinib, k_cat, K_m^ATP, and K_m^Peptide. Note that the latter two also contain α. Errors shown in A are SD of mean for triplicate experiment; errors in B–D are uncertainties of the global fit for each enzyme form. Errors in E are propagated errors from A–D. The Welch t test was used to determine significance of the differences, and bars are marked. *P < 0.05; **P < 0.01.

Compared to the wild-type affinity for ATP (K_D^ATP = 69 ± 4 μM), only G250E showed a significant tighter binding to ATP (K_D^ATP = 47 ± 2 μM), while F317L (K_D^ATP = 76 ± 1 μM) had near identical, and Y253F (K_D^ATP = 98 ± 5 μM) even weaker binding to ATP (Fig. 4B and Table 3). For the peptide substrate compared to wild type (K_D^Peptide = 1,109 ± 89 μM), both F317L and G250E, however, showed increased affinity, with a ∼30% tighter binding for F317L (K_D^Peptide = 755 ± 11 μM) and, remarkably, almost threefold tighter affinity for G250E (K_D^Peptide = 438 ± 43 μM) (Fig. 4C). In addition, kinase activity was increased for all resistance mutants compared to the wild type (k_cat = 19.0 ± 0.5 s⁻¹): F317L (k_cat = 25.6 ± 0.1 s⁻¹) and G250E (k_cat = 24 ± 0.3 s⁻¹) were slightly more active, while Y253F (k_cat = 38 ± 2 s⁻¹) was twice as fast (Fig. 4D). Finally, cooperativity between the different substrates was also substantially affected (Fig. 4E).

Table 3.

Steady-state kinetic parameters for wild-type and mutant forms of Abl

Construct	KDATP, μM	KDPeptide, μM	kcat, s−1	α	Calc. KmATP @ 2 mM peptide, mM	Calc. KmPeptide @ 2 mM ATP, mM
WT	69 ± 2	1,107 ± 57	19.0 ± 0.5	1.8 ± 0.2	97 ± 8	1,975 ± 236
F317L	76 ± 1	755 ± 11	25.6 ± 0.1	1.48 ± 0.02	100 ± 13	1,343 ± 176
G250E	47 ± 2	438 ± 43	24.0 ± 0.3	2.2 ± 0.3	96 ± 13	952 ± 157
Y253F	98 ± 5	1,164 ± 50	38 ± 2	1.4 ± 0.3	104 ± 14	1,617 ± 344

Measured and derived steady-state parameters. Shown are fits of activity assays with varying concentrations of both peptide substrate (0.25 to 2 mM) and ATP (25 μM to 2 mM) shown in Fig. 4 and SI Appendix, Fig. S6. Data are fitted to the general velocity equation for a sequential random Bi–Bi reaction (see also SI Appendix, Eq. 1). This yields catalytic activity (k_cat), binding affinity of ATP (K_D^ATP), and binding affinity of peptide substrate (K_D^Peptide), as well as proportionality factor α. α quantifies the degree that binding of substrate one either increases (a < 1) or decreases (a > 1) affinity of the other substrate to the enzyme. These parameters can then be used to calculate the K_m^ATP as well as K_m^Peptide at fixed concentrations of the other substrate (see also SI Appendix, Eqs. 2 and 3). Calc., calculated.

Markedly, these different changes of Abl’s properties taken together fully recapitulate the observed increase in IC₅₀ at cellular concentrations of ATP for all resistance mutations (Fig. 4F). The quantitative analysis revealed the contribution of each parameter/step to the overall increase of IC₅₀, illustrating how single mutations in Abl can lead to imatinib resistance for patients by relatively small alterations to each individual parameter.

Discussion

Imatinib, and a number of other small-molecule drugs against protein kinases, has been very successful in the initial treatment of cancer. However, this original success has been diminished by the occurrence of drug resistance in response to the treatment. In the IRIS (International Randomized Study of IFN and STI571) trial, ∼17% of patients exhibited imatinib resistance (45). Furthermore, Soverini et al. (12) showed that 12 to 63% of imatinib-resistance patients across 11 studies exhibited mutations in the BCR-Abl kinase domain. While primary resistance (failure of initial treatment) is found in 21 to 48% of identified mutations, secondary resistance (failure of response after prolonged treatment) makes up for a larger percentage, with 10 to 68% of imatinib-resistant cases (12, 13). To open the door for solving the problem of resistance to treatment, a first step is to understand the molecular mechanism by which mutations in the target protein cause resistance to the drugs. This question has, of course, been asked in the literature due to the enormous importance and the obvious nature of this question. Surprisingly, literature reports have been highly controversial, conflicting, or puzzling. A number of the kinase domain resistance mutations have not even been experimentally investigated, but mechanisms have been assumed from static structural data, such as crystal structures and molecular-dynamics simulation-based studies (13, 17, 24, 27, 33, 46, 47). We started this project of characterizing the molecular mechanism of the three major resistance mutations with an interesting observation that many resistance mutations overlap in position with the dynamic hot spots essential for the induced-fit step of imatinib binding (23). From this finding, we reasoned that the induced fit will be hampered by the mutations, and, hence, the drug affinities will be reduced.

While this hypothesis qualitatively held true, we were surprised to find that this effect was much too small to explain drug resistance for any of the mutants. Nonetheless, the results reinforce that 1) these mutations do not alter the physical binding step of imatinib, but 2) alter the induced-fit step. This supports the hypothesis that changes of these 15 dynamic hot spots act cumulatively in causing high affinity of imatinib to Abl via this altered induced-fit step.

To solve this apparent puzzle of drug resistance, we analyzed the effect of these three resistance mutations on altering the free-energy landscape of the kinase and its effect on catalysis, as well as on the competition of the drug with ATP. Only at cellular ATP concentrations does the combination of a 1) small alteration in the induced-fit step of imatinib binding with 2) an increase in enzymatic activity and 3) tighter binding to their substrate peptide combined with substrate cooperativity yield an order-of-magnitude increased IC₅₀ value for imatinib. In other words, subtle changes in the energy landscape of Abl by single point mutations, when added together, lead to resistance in patients.

What are the implications of our findings in respect to published results on this question? Y253F is located at the P loop of the kinase and forms a hydrogen bond with N322. This led to the hypothesis in the literature that substitution of the tyrosine by phenylalanine would disrupt a hydrogen bond, thus removing crucial interaction between kinase and drug and impair binding of the drug (21, 30, 48), leading to resistance. In addition to this structural interpretation, there have been conflicting reports of activities and transformation efficiencies for Y253F. Allen and Wiedemann (49) reported an increased transformation potency for Y253F, as well as an increased total phosphotyrosine content in cell lysates from cells expressing full-length BCR-Abl–Y253F. However, they were unable to see a higher activity in in vivo assays when comparing autophosphorylation of full-length c-Abl with full-length c-Abl–Y253F (49). In a later paper, Roumiantsev et al. (38) were also unable to show increased phosphorylation, neither for autophosphorylation nor for in vitro phosphorylation assays using synthetic peptides as substrates. The authors concluded that for increased catalytic activity of Y253F, expression of BCR-Abl–Y253F in cells was necessary (38). Corbin et al. (39) investigated Y253F in the context of a kinase domain-only construct, but could not detect an increase in activity for Y253F; however, they observed an increase in binding affinity to ATP. Studies by Griswold et al. (36), also utilizing a kinase domain-only construct, showed increased transformation potency, as well as increased catalytic activity in cellular-outgrowth assays and substrate-phosphorylation assays; however, their measured activities are extremely low and only measured at 100 μM ATP. Griswold et al. (36) concluded that differences to published results might stem from insensitive assays using Western blotting of substrate stained with pY antibodies. Our data not only resolve the controversy about the underlying mechanism of resistance mutations by Y253F, but they demonstrate the need to characterize all changes to the enzyme that contribute to the overall resistance of patients’ mutant proteins to imatinib. The increase in activity or the changes in drug affinity alone are not sufficient for the observed resistance in patients.

Similarly, the proposed simple mechanism of disruption of van der Waals interactions with the aromatic ring of imatinib by the F317L mutation (21, 24, 27–30) is not the sole mechanism by which this mutation causes drug resistance. Finally, we describe the mechanism underlying resistance for one of the most commonly isolated resistance mutations (G250E) that had not been quantitatively studied before. We are able to show that G250E has fundamental effects on the energy landscape of the kinase and shows the strongest effect on both drug and ATP affinity.

Considering the findings in this study with previously published data from Yun et al. (40) for EGFR suggests that the cumulative mechanism for resistance is a more broadly applicable one than thought. Our results expose the necessity for careful quantitative analysis of the enzymatic properties, along with the drug-binding mechanism of target proteins to understand their respective resistance causing mutant forms. Enzymatic kits are commonly used in research for probing catalytic activity of kinases and other targets and the effect of inhibitors. However, it is crucial for the user of such kits to keep in mind that many of these assays rely on low ATP or substrate concentrations and could lead to misinterpretation.

Materials and Methods

Expanded methods are available within SI Appendix. In brief, different constructs of Abl kinase were expressed and purified in Escherichia coli as described (23, 35). Rapid kinetics of imatinib binding to Abl kinase were measured via stopped-flow fluorescence, similar to that reported in ref. 23. Kinase activity was measured by using a peptide substrate in a coupled enzymatic assay that detects adenosine diphosphate production through coupling with pyruvate kinase and lactate dehydrogenase, allowing for monitoring of oxidation of NADH to NAD⁺. All activity assays were performed in 50 mM Tris⋅HCl (pH 8.0), 500 mM NaCl, 1 mM Tris(2-carboxyethyl)phosphine, 5% dimethyl sulfoxide, 20 mM MgCl₂, and 0.3 mg/mL bovine serum albumin. Data fitting was performed with the Python package lmfit (50).

Data Availability.

All data and scripts are available in the paper and SI Appendix.