Structural Mechanism for STI-571 Inhibition of Abelson Tyrosine Kinase

See allHide authors and affiliations

Science  15 Sep 2000:
Vol. 289, Issue 5486, pp. 1938-1942
DOI: 10.1126/science.289.5486.1938


The inadvertent activation of the Abelson tyrosine kinase (Abl) causes chronic myelogenous leukemia (CML). A small-molecule inhibitor of Abl (STI-571) is effective in the treatment of CML. We report the crystal structure of the catalytic domain of Abl, complexed to a variant of STI-571. Critical to the binding of STI-571 is the adoption by the kinase of an inactive conformation, in which a centrally located “activation loop” is not phosphorylated. The conformation of this loop is distinct from that in active protein kinases, as well as in the inactive form of the closely related Src kinases. These results suggest that compounds that exploit the distinctive inactivation mechanisms of individual protein kinases can achieve both high affinity and high specificity.

A hallmark of CML is a reciprocal chromosomal translocation involving the long arms of chromosomes 9 and 22 (1). This somatic mutation fuses a segment of thebcr gene, from chromosome 9, to a region upstream of the second exon of the c-abl gene from chromosome 22.c-abl encodes a nonreceptor tyrosine kinase that has tightly controlled activity in normal cells. In contrast, Bcr-Abl fusion proteins have constitutive catalytic activity, despite the fact that the amino acid sequence of the Abl segment of Bcr-Abl is identical to that of c-Abl. The reason for the elevated catalytic activity of the Bcr-Abl fusion protein is poorly understood, but it is clear that this activity of the kinase domain is necessary for the ability of the Bcr-Abl protein to transform cells and cause malignancy.

A series of inhibitors, based on the 2-phenylaminopyrimidine class of pharmacophores, have been identified that have exceptionally high affinity and specificity for Abl (2). The most potent of these, STI-571 (Fig. 1A, formerly referred to as Novartis compound CGP 57148), has been successfully tested in clinical trials as a therapeutic agent for CML. The compound has led to complete hematological response in 96% of patients treated for more than 4 weeks at a dose level of 300 mg, and is well tolerated (3).

Figure 1

Crystal structure of the catalytic domain of Abelson tyrosine kinase complexed with a variant of STI-571. (A) Structural formula of the Abl inhibitor STI-571 (panel 1) and the variant (panel 2) used in this crystallographic study. (B) Ribbon representation of the three-dimensional structure of Abl kinase domain in complex with the STI-571 variant shown in (A). The molecular surface of the inhibitor is shown. A central conserved region of the kinase, the catalytic segment, is shown in green and the activation loop in magenta. (C) Ribbon representation of the activation loop of Abl. The polypeptide backbone of the activation loop is shown in magenta. Hydrogen-bonding interaction are depicted by dashed lines. Tyr393 is the site of phosphorylation within the activation loop. (D) The polypeptide region in the vicinity of the Tyr393 is shown. Superimposed is the peptide substrate (green), as seen in the structure of insulin receptor tyrosine kinase (IRK) complexed with peptide substrate (14), and the activation loop of IRK in the inactive form (light pink) (32). The figure was generated by superimposing the catalytic segments of the two kinases.

Protein kinase inhibitors typically bind at the highly conserved nucleotide-binding pocket of the catalytic domain. Specific inhibitors of protein kinases take advantage of limited sequence variation surrounding the ATP-binding site [e.g. (4, 5)], as well as conformational differences between inactive and active forms of kinases (6). STI-571 has a high affinity for Abl kinase, while being essentially inactive against Ser/Thr-kinases and most of the tyrosine kinases [notable exceptions are two related receptor tyrosine kinases, the platelet-derived growth factor (PDGF) receptor (7) and c-kit (8)].

To determine how STI-571 achieves this high specificity, we solved the crystal structure of a variant of STI-571 (Fig. 1A) bound to the catalytic domain of Abl at 2.4 Å resolution (Fig. 1B) (9,10). There is strong electron density in difference electron density maps for the STI-571 variant, which occupies the site where the adenine base of ATP is normally bound (Fig. 2A). When compared to the pyrazolo-pyrimidine inhibitor PP1 bound to the Src-kinase Hck (11) STI-571 extends much further into the catalytic domain, and its pyridinyl group is inserted underneath helix αC in the NH2-terminal lobe of the kinase. The compound is kinked at the secondary amino group, and it straddles the highly conserved NH2-terminal region of the “activation loop” (Fig. 1B). The inhibitor that we have crystallized differs from STI-571 in that it lacks a piperazinyl group that is attached to the phenyl-ring of STI-571 (Fig. 1A). The piperazinyl group increases the solubility of the parent compound, but does not alter target discrimination significantly (2); it is likely to lie along a solvent accessible and partially hydrophobic groove on the back of the kinase that is left unfilled by the STI-571 variant.

Figure 2

Mode of binding of the STI-571 variant binding to Abl tyrosine kinase. (A) Stereoview of the nucleotide-binding pocket of Abl. The side chains of residues that interact with the inhibitor are shown, as are main-chain atoms and water molecules participating in hydrogen bonds. Carbon atoms are yellow (protein) and green (inhibitor), oxygen atoms are red, nitrogen atoms are blue, and sulfur atoms are green. Hydrogen bonds are shown as dashed lines. Electron density, calculated using (∣F o∣ – ∣F c∣) coefficients, is shown for the inhibitor only (contoured at 2.5 σ), and was computed using phases from a model obtained after simulated annealing with the inhibitor omitted. (B) Schematic drawing of Abl kinase interactions with the STI-571 variant, generated by LIGPLOT (33). Residues forming van-der-Waals interactions are indicated, those participating in hydrogen bonds are shown in a ball-and-stick representation. Hydrogen bonds are depicted as dotted lines with the donor-acceptor distance given in Å.

The activation loop controls catalytic activity in most kinases by switching between different states in a phosphorylation-dependent manner (12). In fully active kinases, the loop is stabilized in an open conformation by phosphorylation on serine, threonine or tyrosine residues within the loop, and in this conformation a β-strand in the loop provides a platform for substrate binding. Three highly conserved residues in the NH2-terminal region of this loop (an Asp-Phe-Gly motif, residues 381 to 383 in Abl) are thereby held in a conformation that is appropriate for metal ion ligation by the aspartic side chain. This “active” conformation of the loop is very similar in all known structures of active kinases. There is, however, great diversity in the conformations of this loop in inactive protein kinases, in which the loop often occludes substrate binding. Additionally, crankshaft-like displacements in the NH2-terminal region of the loop change the conformation of the Asp-Phe-Gly triad, thereby inhibiting the ability of the kinase to bind ATP productively (13).

Tyr393 in the activation loop is the major site of phosphorylation in Abl, but the form of Abl that we have crystallized is not phosphorylated. The activation loop is folded into the active site of the kinase, and Tyr393 forms a hydrogen bond with Asp363, a strictly conserved side chain that is crucial for catalysis (Fig. 1C). Tyr393 is presented to the active site by a small antiparallel β-sheet that is formed by a portion of the activation loop. Interestingly, the activation loop mimics the binding mode of substrates, as was also found in the insulin receptor tyrosine kinase (IRK, Fig. 1D) (14).

Is the structure of the activation loop in Abl a natural conformation of this region of the protein, or is it induced by the binding of STI-571? Except for the NH2-terminal anchor region, STI-571 does not interact directly with the activation loop (Fig. 1C). The striking similarity between the conformation of the activation loop and the manner in which peptide substrates bind to tyrosine kinases suggests that the loop is in a natural auto-inhibitory conformation (Fig. 1D). Comparison of the catalytic domains of Abl and inactive IRK shows that the central part of the activation loop in both kinases occludes the mouth of the catalytic domain and interferes with the productive binding of peptide substrates in a similar manner (Fig. 1D). Although Tyr393 is positioned exactly as in a substrate peptide, the kinase domain is not in a conformation that is competent for phosphate transfer to the tyrosine, since the inward movement of the activation loop is coupled to displacement of the Asp-Phe-Gly motif away from the active conformation in both kinases [Asp381points away from the active site (Fig. 1C)]. Despite these similarities, STI-571 is inactive against IRK, most likely because a side chain which forms a critical contact with the inhibitor (Thr315, see below) at the periphery of the nucleotide binding site of Abl is not conserved in IRK.

In contrast to its interactions with the activation loop, where STI-571 is likely to recognize a natural conformation of the Abl molecule, its interactions with the NH2-terminal lobe of the kinase appear to involve an induced fit mechanism. The loop between the first two β-strands, which normally binds the phosphate groups of ATP, folds down to increase surface complementarity with the drug. This flap is held in place by a water-mediated hydrogen bond between Tyr253, a residue in the NH2-terminal lobe of the kinase that immediately follows the β1-β2 loop, and the side chain of Asn322. A similar conformation for this loop was found in the fibroblast growth factor receptor tyrosine kinase domain bound to a high-affinity oxindole-based inhibitor (4). This loop is known to be an extremely mobile element in protein kinases, and the induced fit is likely to be readily accommodated.

The inhibitor also interacts with the kinase through hydrogen bonds, some of which confer specificity (Fig. 2). The nitrogen in the pyridinyl ring that is attached to the pyrimidine moiety accepts a hydrogen bond from the amide of Met318, which is normally hydrogen bonded to the nitrogen N1 in ATP. The side chain of Thr315 forms a hydrogen bond with the secondary amino group in the inhibitor. This residue is replaced by a methionine in many protein kinases, e.g., IRK. Methionine cannot form this hydrogen bond, and its side chain would also interfere with the binding of the phenyl-moiety of STI-571. The presence of Thr315 is therefore a key requirement for the ability of this class of compounds to inhibit Abl.

An ion-pair between two strictly conserved side chains (Lys271 and Glu286 in Abl) is a characteristic feature of the active conformations of protein kinases. This ion pair is disrupted in the inactive conformations of many protein kinases, such as the Src and cyclin-dependent kinases, but not in the STI-571 complex of Abl. Instead, a network of hydrogen bonds involving the side chain of residues Lys271 and Glu286, as well as the main chain of Asp381, the acid amide group of the inhibitor and two water molecules further stabilizes binding (Fig. 2).

There are a number of van der Waals interactions between protein residues Tyr253, Leu370, Phe382, Met290, and Ile313, and the aromatic rings of the inhibitor, resulting in an exceptional level of surface complementarity. The snug fit hardly allows for any modification on either the inhibitor or the kinase domain without compromising binding affinity. Conversely, alterations in the sequences of other protein kinases in the regions that make up the binding site, such as the replacement of Thr315 by methionine, would interfere with binding.

The most interesting aspect of the interaction between STI-571 and the Abl kinase is that specificity is also achieved at a level beyond simple sequence requirements. The residues that contact STI-571 in Abl kinase are either identical in the Src-family tyrosine kinases, or are substituted conservatively. Nevertheless, the phenylamino-pyrimidines are virtually inactive against the Src-family tyrosine kinases (2). The residues that are not identical in Abl and the Src kinases are also variant in the c-kit and PDGF receptor tyrosine kinases, which are the only other kinases sensitive to inhibition by STI-571. These alterations in sequence therefore do not seem to account for lack of sensitivity of the Src-kinases for inhibition by STI-571 and its derivatives.

Like Abl, the Src-family tyrosine kinases are also inactivated by an inward movement of the activation loop when it is not phosphorylated (11, 15). Although the loop blocks substrate binding in inactive Src kinases, it does not do so by mimicking a substrate, and its conformation is quite different from that seen in Abl (Fig. 3). The inactive conformation of the activation loop in Src kinases is coupled to a particular “swung-out” conformation of helix αC, which is not observed in our Abl structure. The most important consequence of this for the binding of STI-571 is that the conformation of the NH2-terminal anchor of the activation loop (containing the conserved Asp-Phe-Gly motif) is quite different in the inactive Src kinase and Abl structures. The conformation of this region in the Src kinases would block the binding of STI-571 (Fig. 3).

Figure 3

STI-571 exploits the unique conformation of the activation loop in the down-regulated form of Abl. Conformation of the activation loops of Abl and the Src kinases Lck (active) (16) and Hck (inactive) (11). Also shown is a space-filling model of the Abl-specific inhibitor. The figure was generated by superimposing the catalytic segments of the displayed kinases. The structures of Lck and Hck are representative for the active and the inactive state of Src-family tyrosine kinases, respectively. The active form of Abl is expected to resemble that of Lck. The activation loop is magenta, the catalytic segment green. The conserved side chains of the Asp-Phe-Gly motif and the tyrosine residue in the activation loop are shown in a ball-and-stick representation.

Phosphorylation of Tyr393 in Abl would destabilize the closed conformation of the activation loop because of electrostatic repulsion between the phosphoryl group and the side chain of Asp363. We expect that phosphorylation on Tyr393 would stabilize the activation loop in the open conformation seen in active protein kinases, such as the Src kinase Lck (16) (this appears to be a structurally conserved feature of activated protein kinases). Interestingly, the conformation of the NH2-terminal anchor of the activation loop in active protein kinases is also inconsistent with the binding of STI-571 (Fig. 3). We therefore predict that the phosphorylated form of Abl will be less susceptible to inhibition by STI-571.

We tested this by comparing the catalytic activity of the unphosphorylated and phosphorylated forms of the Abl kinase domain in the presence of different concentrations of STI-571 (Fig. 4) (17). Because Abl is slow to autophosphorylate, we used catalytic amounts of the Src kinase Hck to generate the phosphorylated Abl kinase domain. Hck phosphorylates the kinase domain of Abl specifically at Tyr393 in the activation loop (17). Activation of c-Abl by Src kinases plays a role in the cellular response to PDGF (18) and Hck has been implicated in the Bcr-Abl–induced transformation of cells (19).

Figure 4

STI-571 preferentially inhibits the unphosphorylated form of the kinase domain of Abl. Dose response of phosphorylated and unphosphorylated Abl. The kinase activity measured in a continuous spectrophotometric assay is plotted as a function of concentration of STI-571 (17). Abl kinase was phosphorylated in the activation loop by pre-incubation with Hck. The reaction rates are corrected for the rate of a control reaction in the absence of Abl kinase. The amount of Hck used did not give a significant signal in the spectrophotometric assay.

We found that the activity of unphosphorylated Abl was essentially ablated at an STI-571 concentration of 0.5 μM, whereas Abl treated with Hck retained more than 50% of its initial activity under these conditions (Fig. 4). Even at a 10-fold higher concentration of STI-571 the Hck-treated Abl retained 30% activity. The inhibition constant, K i, of STI-571 for the unphosphorylated form of Abl was 37 ± 6 nM. The dose-response curve of the phosphorylated form is complex. It can be analyzed by assuming that inhibition occurs in two steps, one with a K i value of 43 ± 2 nM, which is virtually identical to the K i obtained for the unphosphorylated form, and another with a much higherK i of 7 ± 0.2 μM. The biphasic inhibition suggests that the phosphorylation of Abl by Hck did not go to completion, with the low K icomponent reflecting inhibition of a remaining population of unphosphorylated kinase molecules.

Interestingly, the catalytic activity of the Abl kinase domain was not increased significantly by phosphorylation in the activation loop (Fig. 4). The inactivation of Abl relies on interactions between the catalytic domain and the SH3 domain of Abl (which is lacking in the construct used here) (20–22). The unphosphorylated full-length Abl protein is indeed activated upon autophosphorylation (23, 24). Presumably, the activation loop of the isolated Abl kinase domain is flexible and can adopt the active conformation without requiring the additional stabilization provided by phosphorylation.

In summary, we have shown that although STI-571 targets the relatively well conserved nucleotide-binding pocket of Abl, it can still achieve high specificity by recognizing a distinctive inactive conformation of the activation loop of Abl. The ability of the catalytic domains of protein kinases to adopt characteristic inactive conformations is proving to be a hallmark of these proteins. That STI-571 takes advantage of this feature of its target is encouraging news for the further development of specific protein kinase inhibitors.

  • * To whom correspondence should be addressed. E-mail: kuriyan{at}


View Abstract

Navigate This Article