Engineering the Substrate Specificity of the Abl Tyrosine Kinase*

c-Abl is a non-receptor tyrosine kinase that is involved in a variety of signaling pathways. Activated forms of c-Abl are associated with some forms of human leukemia. Presently, no high resolution structure of the tyrosine kinase domain of Abl is available. We have developed a structural homology model of the catalytic domain of Abl based on the crystal structure of the insulin receptor tyrosine kinase. Using this model as a guide, we selected residues near the active site predicted to play a role in peptide/protein substrate recognition. We expressed and purified 15 mutant forms of Abl with single amino acid substitutions at these positions and tested their peptide substrate specificity. We report here the identification of seven residues involved in recognition of the P−1, P+1, and P+3 positions of bound peptide substrate. Mutations in these residues cause distinct changes in substrate specificity. The results suggest features of Abl substrate recognition that may be relevant to related tyrosine kinases.

c-Abl is a non-receptor tyrosine kinase that is involved in a variety of signaling pathways. Activated forms of c-Abl are associated with some forms of human leukemia. Presently, no high resolution structure of the tyrosine kinase domain of Abl is available. We have developed a structural homology model of the catalytic domain of Abl based on the crystal structure of the insulin receptor tyrosine kinase. Using this model as a guide, we selected residues near the active site predicted to play a role in peptide/protein substrate recognition. We expressed and purified 15 mutant forms of Abl with single amino acid substitutions at these positions and tested their peptide substrate specificity. We report here the identification of seven residues involved in recognition of the P؊1, P؉1, and P؉3 positions of bound peptide substrate. Mutations in these residues cause distinct changes in substrate specificity. The results suggest features of Abl substrate recognition that may be relevant to related tyrosine kinases.
The c-abl proto-oncogene encodes a multidomain non-receptor tyrosine kinase that is expressed ubiquitously in human tissues (reviewed in Refs. [1][2][3][4]. Mutant forms of c-abl are found in patients with Philadelphia chromosome-positive chronic myelogenous leukemia and acute lymphocytic leukemia (1)(2)(3)(4). In these diseases, a chromosomal translocation event produces a chimeric oncogene consisting of 5Ј-sequences of bcr fused to abl. The BCR-Abl fusion protein has elevated tyrosine kinase activity relative to c-Abl, and the tyrosine kinase activity of the BCR-Abl fusion protein is necessary for disease progression. Similarly, tyrosine kinase activity is necessary for transformation of fibroblasts or hematopoietic cells by BCR-Abl (5,6).
In addition to its tyrosine kinase catalytic domain, c-Abl has a short amino-terminal unique domain followed by SH3 and SH2 domains (1)(2)(3)(4). This domain organization is found in many non-receptor tyrosine kinases. Abl also possesses a large carboxyl-terminal region that includes a DNA-binding domain, an F-actin-binding domain, a nuclear localization signal, and a proline-rich region implicated in mediating protein-protein interactions. Studies aimed at understanding the normal physiological role of c-Abl have shown the enzyme to be involved in signal transduction, cytoskeletal rearrangement, RNA polymerase II activation, DNA repair, and cell cycle control (1)(2)(3)(4). c-Abl has been shown to physically associate with at least seven unique proteins, including p53 and the nuclear Rb protein (4). Mice with targeted disruptions in the c-abl gene have high neonatal mortality rates and are more susceptible to infection, suggesting a role for c-abl in B-lymphocyte development (7).
At least eight in vivo substrates for Abl have been identified (4). The amino acid sequences surrounding the phosphorylation sites for two of these proteins, RNA polymerase II (8) and c-Crk (9), have been described. These sequences do not share a common primary sequence motif, suggesting that Abl may have a broad range of substrate specificity. Studies using synthetic peptides have been used to examine the substrate specificity of Abl and to define any primary sequence determinants for substrate recognition (10,11). These studies suggest that, although Abl does not have an absolute consensus sequence for phosphorylation, the best in vitro peptide substrates for Abl contain the sequence Ile-Tyr-Ala-Xaa-Pro, where Xaa is any amino acid. These studies indicate that the PϪ1 (Ile) and the Pϩ3 (Pro) positions are most important for substrate recognition.
The molecular basis of peptide/protein substrate recognition for tyrosine kinases is not well understood. Presently, a single high resolution crystal structure of a tyrosine kinase in complex with peptide substrate is available: the tyrosine kinase domain of the insulin receptor (IRK) 1 complexed with a peptide substrate (12). This structure reveals interactions between enzyme and substrate that govern substrate specificity. Two adjacent hydrophobic pockets on the surface of the C-terminal lobe of IRK accommodate Met side chains C-terminal to the phosphorylated tyrosine on the peptide substrate. The crystal structure of the activated IRK-peptide complex provides a structural basis for understanding the primary signaling specificity of IRK and serves as a general model for tyrosine kinase substrate recognition.
In this paper, we have developed a molecular homology model of the kinase catalytic domain of Abl (Abl-CAT) to help identify amino acids that may be important in substrate recognition. A similar approach was used to propose a molecular model of the Bruton tyrosine kinase and to provide a structural basis for understanding mutations in this enzyme associated with the disease X-linked agammaglobulinemia (13). Our molecular homology model is based on the crystal structure of the ternary complex of IRK with peptide substrate and AMP-PNP bound (12). Using the model as a guide, we have targeted seven residues in Abl-CAT for amino acid substitutions to examine effects on substrate specificity. Site-directed mutants of Abl-CAT were engineered and tested with a series of peptide substrates to monitor changes in specificity. Kinetic analyses of these mutants with the peptide substrates show distinct changes in substrate preferences. primary model of the Abl catalytic domain was prepared using the spatial coordinates of ␣-carbons from the crystal structure of the ternary form of IRK with peptide substrate and AMP-PNP bound (12). Gaps in the sequence alignment were ligated manually, minimizing large steric clashes. Energy minimization and loop insertions were carried out using the Swiss-Model automated modeling system (16). Additional rounds of energy minimization were carried out using Sculpt for Power Macintosh (17). The stereochemical quality of the model was checked using PROCHECK Version 3.3 (18), which reports no distorted main-chain bonds, five distorted main-chain angles, and no distorted planar groups. The distorted main-chain angles were outside of the predicted substrate-binding region and do not interfere with our interpretation of the model. Solvent-accessible surface area was calculated and visualized using Web Lab Viewer (Molecular Simulations Inc.) with a 1.4-Å probe. Figs. 1A and 2 were prepared using Strata Studio Pro (Strata Inc., St. George, UT).
Mutagenesis, Expression, and Purification-Our experiments were carried out on the isolated catalytic domain of v-Abl, expressed in Escherichia coli as described previously (19). The sequence numbering used is from the gag-Abl fusion protein of the Abelson murine leukemia virus (20). Mutagenesis of the Abl catalytic domain was carried out using a QuikChange mutagenesis kit (Stratagene). Mutagenesis primers complementary to wild-type template were designed with single, double, or triple nucleotide substitutions. The DNA sequences encoding the entire catalytic domains of the mutants were confirmed by DNA sequencing on an ABI373 automated DNA sequencer. Wild-type and mutant proteins were expressed as glutathione S-transferase fusion proteins in E. coli strain NB42 and purified using glutathione-agarose (19). All proteins expressed to similar levels, were of the expected size, and purified to Ͼ98% homogeneity.
Peptides-Synthetic peptides were prepared by solid-phase synthesis using standard Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry on an Applied Biosystems automated 431A peptide synthesizer. Peptides were purified using semi-preparative reversed-phase high performance liquid chromatography. Matrix-assisted laser desorption ionization time-of-flight mass spectrometry was used to confirm the identity of the final products.
Kinase Assays-Tyrosine kinase activity assays were carried out using two methods. In both cases, the intact glutathione S-transferase-Abl fusion proteins were used; we showed previously that the specific activity and substrate specificity of glutathione S-transferase-Abl are similar to those of full-length Abl and that velocity versus [enzyme] plots are linear over the concentrations used here (19). For initial comparisons of several substrates, phosphocellulose binding assays were used to measure incorporation of [ 32 P]ATP into peptide substrates (19,21). These reactions were carried out in triplicate at saturating concentrations of ATP (250 M) and Mg 2ϩ (10 mM). The reactions were carried out in volumes of 25 l, using peptide concentrations of 50 M or 1.5 mM and 1 g of enzyme. Picomoles of phosphate incorporated into peptides were measured after 20 min as described (19).
A continuous spectrophotometric assay (22) was used to measure initial rates of phosphorylation and to determine kinetic constants for some peptides. Reactions were carried out in volumes of 100 l using 5 g of purified enzyme. Saturating concentrations of ATP (500 M) and Mg 2ϩ (10 mM) were used, and peptide concentrations ranged from 50 M to 2 mM. Data were sampled every 30 s to determine rates of phosphorylation. Initial rates (Ͻ10% of the reaction) were measured in triplicate. The initial rate values were averaged, and V max and K m were determined by fitting to the hyperbolic velocity versus [substrate] curves using the program MacCurve Fit. For some combinations of peptides and mutants, initial experiments established that their K m values were in the millimolar range (Ͼ2 mM). In addition, we observed that high concentrations of these peptides were inhibitory. Thus, concentrations over 2 mM were not employed, and we were unable to determine K m values accurately using initial rate kinetics. For these peptides, the complete time courses for phosphorylation were measured using peptide concentrations less than K m . In these cases, we analyzed the data graphically as described (23) to determine V max /K m .

RESULTS
Homology Model-The overall topology of the Abl structural homology model reflects the typical bilobal structure shared by all eukaryotic protein kinases (24) with a five-stranded ␤-sheet and a single ␣-helix in the amino-terminal lobe, responsible for MgATP binding, and a highly helical carboxyl-terminal lobe (Fig. 1). Based on results for other protein kinases, the Cterminal lobe is predicted to make most of the contacts with peptide/protein substrates. The total root mean square deviation of the polypeptide backbone between the Abl model and the IRK structure is 1.8 Å. The crystal structure of the activated insulin receptor catalytic domain with peptide bound (12) served as a model for the orientation and structure a peptide may adopt in the Abl active site. Comparison of the C-terminal peptide-binding domains of IRK and Abl indicated that the greatest differences are in the regions responsible for binding the Pϩ3 amino acid side chain. The PϪ1 region differs only slightly from that of IRK. The activation loop (amino acids 500 -521; see Fig. 1) is extremely flexible when examined using molecular mechanics (Sculpt, Interactive Simulations, Inc.), and the structure of the loop in our model is one of many possible conformations. Additionally, the structure of IRK used for our model is multiply phosphorylated on the activation loop (12). Abl contains a single tyrosine within the activation loop, Tyr-513, which is phosphorylated in vivo and in vitro and is believed to be involved in enzyme activation (25,26). Although we believe that our model represents the activated form of Abl, the activation loop tyrosine is modeled in its unphosphorylated state.
Design of Peptide Substrates-Previous studies of the in vitro substrate specificity of Abl have been carried out in this laboratory and by others (for review, see Ref. 27). We designed two groups of peptide substrates to examine any changes in specificity at the PϪ1 and Pϩ1/Pϩ3 positions for engineered mutant forms of Abl (Table I). (Amino acids in peptide substrates are designated by their position relative to the phosphorylated tyrosine. For example, in the sequence Ile-Tyr-Ala-Ser-Pro, Ile is at the PϪ1 position, Ala is at the Pϩ1 position, and Pro is at the Pϩ3 position.) Peptides used to examine Pϩ1 and Pϩ3 specificity share the sequence Ser-Arg-Gly-Asp-Tyr-Xaa 1 -Thr-Xaa 2 -Gln-Ile-Gly, where either Xaa 1 or Xaa 2 is varied. These peptides are based on a peptide sequence derived from a phosphorylation site of insulin receptor substrate-1 used previously in our laboratory to examine the substrate specificity of wildtype Abl at the Pϩ1, Pϩ2, and Pϩ3 positions (19). In these earlier studies, we found that amino acids at the Pϩ2 position do not strongly influence substrate recognition. Moreover, in the ternary structure of IRK, specificity in peptide binding is achieved through interactions with the Pϩ1 and Pϩ3 residues (12). For these reasons, we did not examine the effects of residue changes at the Pϩ2 position in this study.
We chose to use a different series of peptides to examine PϪ1 specificity (Table I). This is because, in the context of the insulin receptor substrate-1 peptides, we did not observe a strong dependence on the amino acid at the PϪ1 position for Abl phosphorylation. 2 Peptides designed to examine specificity at the PϪ1 position share the sequence Leu-Ile-Glu-Asp-Ala-Xaa-Tyr-Ala-Ala-Arg-Gly, where Xaa is varied. This sequence is based on the autophosphorylation site of Src and has been previously used in our laboratory to examine PϪ1 specificity in wild-type Abl (11). Amino acids for the substituted position were chosen to explore the effect of size, charge, and hydrophobic character. Because the two groups of peptides are dissimilar in sequence, we did not attempt to draw conclusions about the relative importance of PϪ1 versus Pϩ1/Pϩ3 recognition for each mutant.
Mutations That Affect Pϩ3 Peptide Recognition-Experiments with synthetic peptides and peptide libraries have demonstrated that Abl prefers proline at the Pϩ3 residue of a substrate. In the crystal structure of IRK, Leu-1219 is part of a binding pocket that surrounds the Pϩ3 methionine side chain of the peptide substrate (12). In the Abl homology model, the residue homologous to Leu-1219 is Tyr-569. Tyr-569 is partially solvent-exposed in our model and could adopt the same role as Leu-1219 in IRK (Fig. 2). Three separate mutant forms of Abl-CAT were engineered with amino acid substitutions at Tyr-569: Y569L, Y569A, and Y569W. Alanine was chosen as a substitute for Tyr-569 to examine the effects of minimizing the amino acid side chain length. In c-Src, the residue corresponding to Tyr-569 is a leucine (28), and the specificity of c-Src at the Pϩ3 position differs from that of Abl (10). For this reason, we chose to introduce leucine as a substitute for Tyr-569. We also mutated Tyr-569 to Trp because the Abl homology model suggests that a large side chain at this position might cause the putative Pϩ3 substrate-binding pocket to be too narrow to accommodate amino acids with large side chains such as proline.
Initial screens of Abl mutants were carried out using peptide concentrations of 1.5 mM and measuring [ 32 P]ATP incorporation after 20 min to assess any changes in substrate phosphorylation (Fig. 3). Experiments using low (50 M) peptide concentrations showed no changes in the rank order of specificity compared with those using higher peptide concentrations (data not shown). These initial activity measurements indicated that the Y569W and Y569L mutants had an altered specificity (Fig.  3); in particular, recognition of Pro at the Pϩ3 position was greatly reduced. These changes in specificity were characterized further by more detailed kinetic analyses (Table II). The kinetic measurements show that the Y569W mutant phosphorylated the Pϩ1 Ala /Pϩ3 Met peptide best, with a V max /K m value of 5.6. The Pϩ1 Met /Pϩ3 Pro peptide, the best for the wild type with a V max /K m value of 10.7, was phosphorylated less efficiently by this mutant, with a V max /K m value of 1.3 (Table II). The phosphorylation of other peptides by the Y569W mutant was similar to the wild type ( Fig. 3 and Table II). The difference in specificity observed in the Y569W mutant therefore arises from a decrease in V max /K m for the Pϩ3 Pro peptide specifically rather than an overall decrease in the phosphorylation of all peptides. Kinetic measurements also showed a decrease in the phosphorylation of the Pϩ1 Met /Pϩ3 Pro peptide by Y569L, with a V max /K m value of 4.4 (wild-type V max /K m ϭ 10.7) ( Table II). The Y569A mutant did not display any changes in phosphorylation of the peptides tested compared with wild-type Abl (Fig.  3), and we did not carry out kinetic analysis of this mutant.
Mutations That Affect Pϩ1 Peptide Recognition-Two residues in Abl-CAT, Phe-521 and Ile-523, were identified as residues that might contribute to Pϩ1 substrate specificity (Fig.  2). Ile-523 was chosen based on the homology model. A solventaccessible surface representation suggests that Ile-523 is partially solvent-exposed and forms the edge of a groove into which a substrate amino acid side chain may fit. Two substitutions were made for Ile-523: I523V and I523A. These amino acid substitutions were chosen to examine the effect of shortening the side chain on accommodating Pϩ1 residues. Phe-521 was selected to test the possibility that residues near the activation loop contribute to specificity; based on homology to other protein kinases, the activation loop in Abl is predicted to span residues 500 -521. We examined the effects of smaller (F521A) and larger (F521W) side chains at this position. Initial measurements of activity were used to assess changes in specificity (Fig. 4) using a panel of peptides that vary at the Pϩ1 position. Initial rate measurements were used to characterize the changes in specificity shown in the I523V and F521A mutants (Table III). The I523A and F521W mutants did not display any changes in specificity and were indistinguishable from wildtype enzyme in our initial comparisons (Fig. 4). For this reason, we did not pursue kinetic characterization of these mutants.
The specificity of the I523V mutant shows a change at the Pϩ1 position. Phosphorylation of the Pϩ1 Ala /Pϩ3 Met peptide by the wild type (V max /K m ϭ 6.3) and I523V (V max /K m ϭ 10.0) was comparable (Table III). However, the I523V mutant phosphorylated the Pϩ1 Ile /Pϩ3 Met peptide ϳ3.5 times better than the wild type (Table III). Initial rate measurements on the F521A mutant showed generally higher V max /K m values for the Pϩ1 Ala /Pϩ3 Met , Pϩ1 Nle /Pϩ3 Met , and Pϩ1 Ile /Pϩ3 Met peptides when compared with wild-type values (Table III). Like wildtype enzyme, the F521A mutant still phosphorylated the Pϩ1 Ala /Pϩ3 Met peptide best, with V max /K m ϭ 13.0, compared with the wild type, which has a V max /K m value of 6.3 for the same peptide.
Mutations That Affect PϪ1 Peptide Recognition-Isoleucine at the PϪ1 position is a strong determinant for substrate recognition in wild-type Abl (10,11). We produced Abl mutants with amino acid substitutions at residues predicted to be near the PϪ1-binding region of the enzyme. Abl-CAT residue Leu-444 was chosen as a target for mutagenesis based the role of the corresponding residue in IRK, Lys-1085. In IRK, Lys-1085 extends from the ␣D helix and makes a water-mediated hydrogen bond with the PϪ1 residue (Asp) of the substrate peptide (12). In our model, Leu-444 extends from the structurally homologous helix and is partially solvent-exposed. We chose to change Leu-444 to lysine to examine the possibility of a change in substrate specificity to that of IRK (which prefers Glu at PϪ1) (10). An L444E mutation was made to examine the possibility that introduction of an acidic residue at this position may favor the binding of a basic amino acid at the PϪ1 position of peptide substrate. The L444K and L444E mutants retained the overall preference for Ile at the PϪ1 position, although they differed from the wild type in phosphorylation of other substrates ( Fig. 5 and Table IV). The L444K mutant has V max /K m values of 8.0 for the PϪ1 His peptide and 21.0 for the PϪ1 Ile peptide, whereas wild-type Abl has V max /K m values of 6.8 for PϪ1 His and 12.3 for PϪ1 Ile (Table IV). Thus, L444K shows enhanced recognition of PϪ1 Ile relative to wild-type Abl. The L444E mutant demonstrated a specificity different from that of L444K. The V max /K m values for the L444E mutant are 14.5 for PϪ1 His and 18.9 for PϪ1 Ile (Table IV); thus, there is a selective increase in PϪ1 His recognition relative to the wild type. Both L444K and L444E have lower V max /K m values for PϪ1 Glu than wild-type Abl. Wild-type enzyme has a V max /K m value of 1.3 for PϪ1 Glu . The L444K mutant has a V max /K m value of 0.5 for PϪ1 Glu , and the L444E mutant has a V max /K m value of 0.07, a 19-fold reduction (Table IV).
Abl-CAT residues Gly-556 and Ser-558 were chosen because these residues are solvent-accessible in the homology model, in close proximity to Leu-444 (Fig. 2). We produced G556A, G556V, S558A, and S558N mutants. The G556A and S558A mutants were indistinguishable from the wild type in our initial screens (Fig. 5). The G556V mutant, however, has a reduced preference for Ile at the PϪ1 position and an increased preference for Leu at the PϪ1 position. For wild-type enzyme, the V max /K m value for PϪ1 Ile is 12.3. For the G556V mutant, the V max /K m value for PϪ1 Ile is 4.2 (Table IV). The S558N mutant showed subtle changes in specificity. The V max /K m value for PϪ1 Ile , 12.0, is similar to the value of 12.3 for wildtype Abl, and the V max /K m value of this mutant for PϪ1 His , 5.4, is close to the value of 6.8 for wild-type Abl (Table IV). The V max /K m value of this mutant for PϪ1 Glu , 0.08, is substantially lower than the corresponding value of 1.3 for wild-type Abl phosphorylating PϪ1 Glu (Table IV).
We also chose to mutate Trp-525 of Abl based on our modeling studies. To identify residues in Abl that may make contacts with the PϪ1 side chain, we used the activated IRK structure to model Ile (in place of Asp) at the PϪ1 position of the peptide substrate. The indole of Trp-1175 of IRK packs against the side chain of one energetically favorable rotamer of the modeled Ile. In our Abl model, the residue homologous to Trp-1175, Trp-525, lies at the bottom of the putative PϪ1-binding pocket (Fig. 2). Trp at this position is conserved in protein-tyrosine kinases (28). We substituted the bulky residues Phe and His for Trp-525 to minimize perturbations in the structure. The W525F and W525H mutants preferred the PϪ1 Ile peptide overall and showed modest changes toward other substrates. For example, the enzymes showed different abilities to phosphorylate the PϪ1 His and PϪ1 Glu peptides (Table IV). W525H had a de- Lys-Lys-Leu-Ile-Glu-Asp-Ala-Glu-Tyr-Ala-Ala-Arg-Gly PϪ1 Glu Lys-Lys-Leu-Ile-Glu-Asp-Ala-Ile-Tyr-Ala-Ala-Arg-Gly PϪ1 Ile Lys-Lys-Leu-Ile-Glu-Asp-Ala-Leu-Tyr-Ala-Ala-Arg-Gly PϪ1 Leu Lys-Lys-Leu-Ile-Glu-Asp-Ala-His-Tyr-Ala-Ala-Arg-Gly PϪ1 His Lys-Lys-Leu-Ile-Glu-Asp-Ala-Ala-Tyr-Ala-Ala-Arg-Gly PϪ1 Ala Lys-Lys-Leu-Ile-Glu-Asp-Ala-Lys-Tyr-Ala-Ala-Arg-Gly PϪ1 Lys Lys-Lys-Leu-Ile-Glu-Asp-Ala-Gln-Tyr-Ala-Ala-Arg-Gly PϪ1 Gln
We tested for indirect effects using the Pϩ1 and PϪ1 mutants as well. We carried out initial comparisons of the Pϩ1 mutants using four peptides that vary at the Pϩ3 position: Pϩ1 Met /Pϩ3 Met, Pϩ1 Met /Pϩ3 Thr , Pϩ1 Met /Pϩ3 Ala , and Pϩ1 Met / Pϩ3 Pro (Fig. 4). In these experiments, there was no observable difference between the substrate specificities of wild-type Abl and the Pϩ1 mutants toward the peptides varying at Pϩ3 (Fig.  4). Similarly, we observed no differences between the Pϩ1 mutants and the wild type in recognition at the PϪ1 position (data not shown).
We screened the following four PϪ1 mutants of Abl for indirect effects: L444K, L444E, S558N, and S558A. We tested the following Pϩ1/Pϩ3 peptides: Pϩ1 Met /Pϩ3 Met , Pϩ1 Ala /Pϩ3 Met , Pϩ1 Glu / Pϩ3 Met , Pϩ1 Met /Pϩ3 Ala , and Pϩ1 Met /Pϩ3 Thr . All four of the mutants displayed the same rank order of substrate preference as the wild type in this experiment (Pϩ1 Ala /Pϩ3 Met Ͼ Pϩ1 Met / Pϩ3 Met Ͼ Pϩ1 Glu /Pϩ3 Met Ͼ Pϩ1 Met /Pϩ3 Ala Ϸ Pϩ1 Met /Pϩ3 Thr ) (data not shown). We conclude from these studies on indirect effects that the sites on Abl for recognition of the PϪ1 and Pϩ1/Pϩ3 positions are distinct. On the other hand, the Pϩ1 and Pϩ3 sites may have some overlap, as at least one mutation (Y569L) had an effect on recognition of both positions.

DISCUSSION
Although Abl is capable of phosphorylating a wide range of peptide and protein substrates, the best peptide substrates for Abl contain the sequence Ile-Tyr-Ala-Xaa-Pro, as shown in peptide library studies (10,11). Ile at the PϪ1 position and Pro at the Pϩ3 position are the most important determinants of substrate specificity for Abl. Here, we have identified residues in the catalytic domain of Abl involved in peptide substrate binding and specificity. These residues are located primarily in the C-terminal lobe of the catalytic domain, which has been implicated previously in substrate binding for other protein kinases (12,24,27).
The mutant forms of Abl described here fall into three classes with respect to substrate specificity. 1) Two mutants (Y569W and Y569L) have altered substrate specificity. These mutants no longer prefer proline at the Pϩ3 position in peptide substrates. 2) Many of the mutants (e.g. I523V, W525H, and L444E) showed no change in the major determinants for substrate recognition, but differed from the wild type in their phosphorylation of other peptide substrates. For example, mutations aimed at altering recognition of the PϪ1 position resulted in enzymes that still preferred Ile at PϪ1, but that FIG. 3. Initial comparisons of wildtype and mutant forms of Abl (Y569W, Y569L, and Y569A) with P؉1/P؉3 peptide variants. Wild-type Abl and the three mutant forms of Abl were tested with a panel of eight peptides. The incorporation of [ 32 P]phosphate into peptides was determined after a 20-min reaction using the phosphocellulose paper assay.

TABLE II
Kinetic measurements for wild-type Abl and two mutant forms of Abl (Y569W and Y569L) Enzyme/peptide combinations were chosen based on initial comparisons of enzymatic activity toward Pϩ1/Pϩ3 peptides (Fig. 3). Kinetic constants for those enzyme/peptide combinations with significant changes from the wild type are shown. To compare phosphorylation of a particular peptide by mutant versus wild type, the V max /K m value for the mutant was divided by the value for the wild type. This ratio is given in the last column. diverged from the wild type when screened against peptides containing other amino acids at PϪ1. In these cases, these residues may not be involved in direct interactions with the PϪ1 residue of substrate. Instead, because of their vicinity to the PϪ1 position, they may act indirectly, stabilizing the local structure to interact favorably with Ile at the PϪ1 position. 3) Some mutants (e.g. Y569A, I523A, and G556A) showed no changes in specificity when assayed against a variety of peptide substrates. These mutants were not characterized by kinetic analysis. The Y569W mutation in Abl has the most dramatic effect on substrate specificity of the mutants we report here. Wild-type Abl phosphorylates a peptide substrate with Pro at the Pϩ3 position best. The Y569W mutant phosphorylates Pϩ1 Met / Pϩ3 Pro ϳ10 times less efficiently than does wild-type Abl (Table II). All other amino acid side chains tested at the Pϩ3 position of the substrate were phosphorylated at the same level as in the wild type (Fig. 3). Our structural model suggests that this change in specificity arises from a steric clash between the side chain of Trp-569 in the mutant and the side chain of proline in the substrate. This is not the case with the other peptide substrates tested that have smaller amino acid side chains at the Pϩ3 position. Proline at the Pϩ3 position plays a role in substrate recognition in vivo in at least one case: Abl phosphorylates c-Crk at Tyr-221 within the sequence Tyr-Ala-Gln-Pro (9). Phosphorylation by Abl is believed to modulate the protein binding and transforming activity of Crk (29). Preliminary experiments indicate that, in contrast to wild-type Abl, the Y569W mutant has no activity toward Crk in vitro. 3 Mutations predicted to affect substrate recognition at the Pϩ1 position (I523V) or at the PϪ1 position (L444K, L444E, G556V, S558N, W525F, and W525H) do so in a more subtle manner. These mutations do not change the overall preference for Ile at PϪ1 or Ala at Pϩ1; however, we observed effects on specificity when we screened these mutants against peptides containing other residues at PϪ1 or Pϩ1. For example, the I523V mutant still phosphorylated the Pϩ1 Ala /Pϩ3 Met peptide best (of the peptides tested), but the V max /K m value for the Pϩ1 Ile /Pϩ3 Met peptide was 3.5 times higher in this mutant than in the wild type (Table III). The L444E mutant, while still preferring Ile at the PϪ1 position, was 2.1 times more efficient at phosphorylating the PϪ1 His peptide than the wild type and 19 times less efficient at phosphorylating the PϪ1 Glu peptide than the wild type (Table IV). There are at least two explanations for these subtle effects on substrate specificity. (i) The residues may not make direct contact with bound substrate, but might instead be involved indirectly in maintaining the three-dimensional structure of Abl to favor certain amino acids in the substrate. (ii) Additionally or alternatively, substrate specificity at PϪ1 or Pϩ1 may be achieved by a combination of residues, such that single amino acid substitutions do not cause complete alterations in substrate recognition. Indirect effects could explain why, for example, the L444K mutant phosphorylates the PϪ1 Ile peptide better than the wild type and shows a decrease in the phosphorylation of PϪ1 Glu when compared with the wild type (Table IV). There are residues in the three-dimensional structures of Src family kinases that appear to correspond to residues identified in our study (30,31). The substrate specificity of Src differs from that of Abl at the Pϩ3 position (10). Src prefers a phenylalanine at the Pϩ3 position in a peptide substrate, whereas Abl prefers proline. The residue homologous to tyrosine 569, a FIG. 4. Initial comparisons of wildtype and mutant forms of Abl (F521A, F521W, I523V, and I523A) with P؉1 peptide variants. Each of the four mutant forms of Abl was tested with the panel of eight peptides. The incorporation of [ 32 P]phosphate into peptides was determined after a 20-min reaction using the phosphocellulose paper assay.

TABLE III
Kinetic measurements for wild-type Abl and two mutant forms of Abl (I523V and F521A) Enzyme/peptide combinations were chosen based on initial comparisons of enzymatic activity toward all Pϩ1 peptides (Fig. 4). Kinetic constants for those enzyme/peptide combinations with significant changes from the wild type are shown. To compare phosphorylation of a particular peptide by mutant versus wild type, the V max /K m value for the mutant was divided by the value for the wild type. This ratio is given in the last column.   L444E, S558N, G556V, W525H, and W525F) Enzyme/peptide combinations were chosen based on initial comparisons of enzymatic activity toward all PϪ1 peptides (Fig. 5). Kinetic constants for those enzyme/peptide combinations with significant changes from the wild type are shown. To compare phosphorylation of a particular peptide by mutant versus wild type, the V max /K m value for the mutant was divided by the value for the wild type. This ratio is given in the last column. residue involved in Pϩ3 specificity in Abl (Fig. 3), is a leucine (Leu-472) in Src. This sequence difference may account for the differences seen in substrate specificity; an L472Y mutant of Src might phosphorylate Pϩ3 Pro -containing peptides more efficiently. A tryptophan substitution at this position could prevent large side chain amino acids from binding in this region, as we observed for Abl. The substrate specificities of tyrosine kinase catalytic domains are important in maintaining the fidelity of cellular signal transduction pathways. This is best illustrated in the case of the RET receptor. A naturally occurring mutation in the kinase domain of this receptor changes a methionine residue to a threonine residue in a region homologous to the region of Abl shown here to be involved in substrate recognition of the Pϩ1 residue (32,33). This change affects the substrate specificity of the enzyme at the Pϩ1 position, changing the preference from methionine at that position in the substrate to alanine (10). This mutant form of the RET receptor is implicated in multiple endocrine neoplasia type 2A (32,33).
Mutations throughout the Abl protein have been reported previously (4). Many of these mutations affect the regulation of the enzyme in vivo. One such mutation, which is sufficient to activate c-Abl enzymatic activity in vivo, is found in the catalytic domain. This mutation changes a tyrosine to phenylalanine within the ATP-binding fold of the enzyme (34). Mutations that affect substrate recognition by the catalytic domain, however, have not been reported previously. Our studies on Abl have highlighted seven residues as playing important roles in peptide substrate recognition. We also show that a single amino acid change of Tyr-569 to tryptophan can affect the substrate specificity dramatically. The results raise the possibility of altering tyrosine kinase substrate specificity in vivo by protein engineering.