Crystallographic and Solution Studies of an Activation Loop Mutant of the Insulin Receptor Tyrosine Kinase INSIGHTS INTO KINASE MECHANISM*

The tyrosine kinase domain of the insulin receptor is subject to autoinhibition in the unphosphorylated basal state via steric interactions involving the activation loop. A mutation in the activation loop designed to re-lieve autoinhibition, Asp-1161 3 Ala, substantially in- creases the ability of the unphosphorylated kinase to bind ATP. The crystal structure of this mutant in complex with an ATP analog has been determined at 2.4-Å resolution. The structure shows that the active site is unobstructed, but the end of the activation loop is disordered and therefore the binding site for peptide substrates is not fully formed. In addition, Phe-1151 of the protein kinase-conserved DFG motif, at the beginning of the activation loop, hinders closure of the catalytic cleft and proper positioning of a -helix C for catalysis. These results, together with viscometric kinetic measurements, suggest that peptide substrate binding induces a reconfiguration of the unphosphorylated activation loop prior to the catalytic step. The crystallographic and solution studies provide new insights into the mechanism by which the activation loop controls phosphoryl transfer as catalyzed by the insulin receptor. The Ala mutant cytoplasmic domain of the insulin receptor, IRCD DA , generated and purified as described. 3 This form of the kinase was used for denaturation studies. To generate the 35-kDa form (residues 978–1283) of the mutant (IRKD DA ) used in the crystallo- graphic and viscometric studies, the 0.65-kb Xho I- Stu I fragment from pX-D1161A-IRCD was inserted into the Xho I- Stu I sites of pALTER- IRK vector, which includes the point mutations Cys-981 3 Ser and Tyr-984 3 Phe described The fragment swap was verified with Nhe I digestion; this restriction enzyme site was introduced previously to replace a Stu I site by a silent mutation at Ala-1048 and Ser-1049 The resulting pALTER-IRKD DA was digested with Hin dIII, filled in with the large Klenow fragment of DNA polymerase, and a 0.9-kb fragment was released with Bam HI. This was inserted into the Bam HI- Sma I sites of the baculovirus expression vector pVL1393 (PharMingen), producing pVL1393-IRKD DA , and the recombinant virus was generated using a Baculogold kit (PharMingen); the mutation was reconfirmed in pVL1393-IRKD DA by DNA sequencing. Proteins were expressed and purified as described 3 The absence of A-loop phos- phorylation in each form (46 kDa and 35 kDa) of the mutant was determined by endoproteinase Lys-C digestion and peptide mapping by reverse-phase high performance liquid chromatography.

The insulin receptor is an ␣ 2 ␤ 2 heterotetrameric glycoprotein possessing intrinsic protein-tyrosine kinase (PTK) 1 activity (1,2). Upon insulin binding to the ␣ subunits, the insulin receptor undergoes a poorly characterized conformational change that results in autophosphorylation of specific tyrosine residues in the cytoplasmic portion of the ␤ subunits. Three regions in the ␤ subunits are sites of autophosphorylation: the juxtamembrane region, the activation loop (A-loop) within the tyrosine kinase domain, and the C-terminal tail (3)(4)(5)(6). Autophosphorylation of tyrosine residues stimulates receptor catalytic activity (7,8) and creates recruitment sites for downstream signaling molecules such as the insulin receptor substrate (IRS) proteins (9) and Shc (10,11).
Previous crystallographic studies of the tyrosine kinase domain of the insulin receptor (IRKD) have demonstrated that upon autophosphorylation, the kinase A-loop undergoes a major change in conformation (12,13). In the crystal structure of the unphosphorylated, low activity form of IRKD (IRKD 0P ) without ATP (apo), Tyr-1162 in the A-loop is situated in the active site, blocking access to peptide substrates (12). In this A-loop configuration, the beginning (proximal end) of the Aloop interferes with ATP binding. The crystal structure of the tris-phosphorylated, activated form of IRKD (IRKD 3P ) reveals how autophosphorylation of Tyr-1158, Tyr-1162, and Tyr-1163 stabilizes a specific A-loop configuration in which the substrate binding sites (MgATP and peptide) are accessible and the important catalytic residues are properly positioned (13,14).
Solution studies of IRKD indicate, however, that in the presence of millimolar quantities of ATP (as are present in a cell), the A-loop of unphosphorylated IRKD is in equilibrium between inhibiting, "gate-closed" conformations, as represented by the apoIRKD 0P crystal structure, and "gate-open" conformations in which Tyr-1162 is displaced from the active site (15). When the A-loop adopts a gate-open conformation, the kinase is competent to serve as either enzyme or substrate in a transautophosphorylation reaction. Prior to A-loop autophosphorylation, gate-open conformations of the A-loop would exist in which the majority of the A-loop has no particular conformation (because of lack of phosphotyrosine-mediated interactions), but is nevertheless disengaged from the active site. After autophosphorylation, the A-loop is stabilized in the gate-open conformation observed in the IRKD 3P structure.
A detailed understanding of the mechanism by which insulin triggers the initial autophosphorylation event in the insulin receptor requires a structural description of the basal state (unphosphorylated) kinase with bound substrates (ATP and protein). Ideally, this would be provided by a crystal structure of IRKD 0P with bound ATP analog and peptide substrate. To date, attempts to obtain crystals of such a ternary (or binary) complex have been unsuccessful. Steady-state kinetic studies of IRKD provide a plausible explanation for this failure: the K m values for ATP and peptide substrate are elevated prior to autophosphorylation, 0.9 and 2 mM, respectively, decreasing to 0.04 and 0.05 mM upon autophosphorylation. 2 These data are consistent with the autoinhibitory mechanism suggested by the apoIRKD 0P structure and underscore the inherent difficulty of loading the kinase with substrates prior to A-loop autophosphorylation.
Recently, a substitution in the A-loop of IRKD, Asp-1161 3 Ala (IRKD DA ), has been introduced that dramatically alters the A-loop equilibrium in the unphosphorylated kinase. 3 This particular substitution was motivated by the apoIRKD 0P crystal structure in which the Asp-1161 side chain participates in several hydrogen bonds that stabilize the gate-closed configuration of the A-loop (Fig. 1). Steady-state kinetic experiments demonstrate that the K m(ATP) in the basal state is ϳ10-fold lower for this mutant than for wild-type IRKD, suggesting that the A-loop equilibrium is shifted toward gate-open conformations. 3 Interestingly, this mutation does not affect K m(peptide) in the unphosphorylated state, which remains high (several millimolar). The kinetic properties of IRKD DA after autophosphorylation are indistinguishable from those of the wild-type kinase.
The lower K m(ATP) for this mutant affords the possibility of structurally characterizing IRKD with ATP bound prior to insulin-stimulated A-loop autophosphorylation. Indeed, crystals of IRKD DA with a bound ATP analog (AMP-PCP) were readily obtained. Here we present the structure of the binary complex of IRKD DA with MgAMP-PCP at 2.4-Å resolution. This structure and the accompanying viscometric and denaturation data provide insights into the structural rearrangements that occur within the basal state kinase to promote catalysis.

EXPERIMENTAL PROCEDURES
Protein Production-The 46-kDa form (residues 953-1355) of the Asp-1161 3 Ala mutant cytoplasmic domain of the insulin receptor, IRCD DA , was generated and purified as described. 3 This form of the kinase was used for denaturation studies. To generate the 35-kDa form (residues 978 -1283) of the mutant (IRKD DA ) used in the crystallographic and viscometric studies, the 0.65-kb XhoI-StuI fragment from pX-D1161A-IRCD was inserted into the XhoI-StuI sites of pALTER-IRK vector, which includes the point mutations Cys-981 3 Ser and Tyr-984 3 Phe described previously (16). The fragment swap was verified with NheI digestion; this restriction enzyme site was introduced previously to replace a StuI site by a silent mutation at Ala-1048 and Ser-1049 (17). The resulting pALTER-IRKD DA was digested with HindIII, filled in with the large Klenow fragment of DNA polymerase, and a 0.9-kb fragment was released with BamHI. This was inserted into the BamHI-SmaI sites of the baculovirus expression vector pVL1393 (PharMingen), producing pVL1393-IRKD DA , and the recombinant virus was generated using a Baculogold kit (PharMingen); the mutation was reconfirmed in pVL1393-IRKD DA by DNA sequencing. Proteins were expressed and purified as described (18). 3 The absence of A-loop phosphorylation in each form (46 kDa and 35 kDa) of the mutant was determined by endoproteinase Lys-C digestion and peptide mapping by reverse-phase high performance liquid chromatography.
Crystallographic Studies-Crystals of the binary complex of IRKD DA and MgAMP-PCP were grown at 20°C by vapor diffusion in hanging drops containing 2.0 l of protein solution (9 mg/ml IRKD DA , 1.5 mM AMP-PCP (Sigma), and 4.5 mM MgCl 2 ) and 2.0 l of reservoir buffer (18% polyethylene glycol 8000, 100 mM Tris-HCl, pH 7.5, 100 mM NaCl, and 5 mM dithiothreitol). Crystals belong to the orthorhombic space group P2 1 2 1 2 1 with unit cell dimensions a ϭ 57.9 Å, b ϭ 69.6 Å, and c ϭ 89.3 Å when frozen. There is one molecule in the asymmetric unit, and the solvent content is 51% (assuming a protein partial specific volume of 0.74 cm 3 /g). Crystals were transferred into a cryosolvent consisting of 30% polyethylene glycol 8000, 100 mM Tris-HCl, pH 7.5, 100 mM NaCl, and 15% ethylene glycol. A data set was collected from a flash-cooled crystal on a Rigaku RU-200 rotating anode equipped with a Rigaku R-AXIS IIC image plate detector. Data were processed using DENZO and SCALEPACK (19). A molecular replacement solution was found with AMoRE (20), using the structure of IRKD 0P (PDB entry 1IRK) (12) as a search model. Rigid-body, positional, and B-factor refinement and simulated annealing were carried out using CNS (21) (Table I). Model building was performed using O (22).
Viscometric Measurements-The viscosity dependence for k cat and k cat /K m was determined using sucrose as the microviscogen following published procedures (23,24). The solution viscosities were determined using an Ostwald capillary viscometer maintained at 24 Ϯ 0.1°C in a temperature-controlled circulating water bath. The relative viscosity, rel , is the ratio of the viscosity in the presence versus the absence of viscogen and was determined as described by Adams and Taylor (25). The standard buffer was 50 mM Tris acetate, pH 7.0. Viscosity dependences of k cat were done using 0.1 M IRKD DA (unphosphorylated), 0.14 -2.7 mM Y-IRS939 peptide substrate (REETGSEYMNMDLG), and constant 1 mM ATP and then extrapolated to saturating substrate by fitting the observed k cat versus peptide concentration to a rectangular hyperbola. To determine the viscosity dependences of k cat /K m(peptide) , measurements were done at 1 mM ATP and 0.14 mM Y-IRS939. A control reaction measuring k cat was done using ATP␥S to show the absence of viscosity effects in the IRKD DA -catalyzed reaction because phosphoryltransferase reactions with ATP␥S are usually chemistrylimited rather than diffusion step-limited (26 -28). The control reactions were done at 1 mM ATP␥S, 0.1-2.0 mM Y-IRS939, and 0.4 M IRKD DA , and the data were extrapolated to saturating peptide substrate by fit to a hyperbolic equation, with K m(peptide) ϭ 2.8 mM. 3 All reactions were done twice in triplicate. The calculation of stepwise rate constants for IRKD DA was done according to Adams and co-workers (24,25), and for wild-type IRKD by Ablooglu and Kohanski. 2 Denaturation Experiments-IRCD DA (46-kDa form of the mutant) was denatured by increasing concentrations of guanidinium chloride at 24°C in 50 mM Tris acetate, pH 7.0, with 1 mM dithiothreitol. The protein concentration was 0.5 M. The excitation wavelength was 295 nm, and steady-state emission spectra were collected between 310 and 420 nm at 1-nm increments, using an SLM 4800 spectrofluorimeter operating in the single-photon counting mode. The centroid of the emission spectrum was determined after subtraction of a blank spectrum for each guanidinium chloride concentration, which was obtained from the refractive index measured with a Bausch and Lomb refractometer. Details regarding instrument settings and data handling are given in Bishop et al. (18).

RESULTS
Crystal Structure of IRKD DA in Complex with MgAMP-PCP-In the original structure of unphosphorylated (low activity) IRKD, the A-loop traverses the ATP-binding cleft between the N-and C-terminal lobes of the kinase, and Tyr-1162 in the A-loop is bound in the active site, hydrogen-bonded to Asp-1132 and Arg-1136 in the catalytic loop (12). Asp-1161 contributes to the stabilization of this inhibitory conformation of the A-loop by participating in four hydrogen bonds ( Fig. 1). In the crystal structure of the Asp-1161 3 Ala mutant IRKD in complex with MgAMP-PCP ( Fig. 2), the A-loop adopts a conformation in which the active site is unobstructed (gate-open), consistent with solution studies measuring the accessibility of the active site. 3 In the IRKD DA structure, the proximal end of the A-loop, containing the protein kinase-conserved 1150 DFG sequence, is positioned more similarly to that in the activated IRKD 3P structure than that in the apoIRKD 0P structure (Fig. 3A). Aloop residues 1155-1171, which include the three autophosphorylation sites Tyr-1158, Tyr-1162, and Tyr-1163, have no supporting electron density in the IRKD DA structure and are presumed to be disordered. The A-loop becomes ordered again at PTK-conserved Pro-1172, which adopts the same conformation in the three IRKD structures (IRKD DA , IRKD 0P , and IRKD 3P ). The absence of Tyr-1162 in the active site in the binary IRKD DA structure is consistent with biochemical studies (16) and modeling exercises, which indicate that MgATP and Tyr-1162 cannot bind in the active site simultaneously, i.e. that cis-autophosphorylation of Tyr-1162 is not sterically possible.
In protein kinases, residues in both the N-and C-terminal lobes bind and thus position ATP for phosphoryl transfer (29). The extent to which ATP is bound productively depends on the degree of lobe closure, the relative disposition of the two lobes. A superposition of C-terminal lobe residues for the three IRKD structures reveals that the degree of lobe closure for IRKD DA is intermediate between IRKD 0P and IRKD 3P (Fig. 3A). From its position in the IRKD 0P structure, the N-terminal lobe is rotated 11°toward the C-terminal lobe in the IRKD DA structure. An additional 8°is required for the N-terminal lobe to reach the position observed in the IRKD 3P structure. Analysis of the changes in backbone , torsion angles shows that the hinge points for the N-terminal lobe rotation are at Arg-1061, before ␤-strand 4 (␤4), and Met-1079, in the segment linking the Nand C-terminal lobes.
When the N-terminal ␤ sheet is superimposed for the three IRKD structures, ␣-helix C (␣C) in the IRKD DA structure is observed to be in essentially the same position with respect to the ␤ sheet as it is in the IRKD 0P structure. Thus, in the transition from the autoinhibited, gate-closed conformation of the A-loop (IRKD 0P ) to a gate-open conformation with bound ATP (IRKD DA ), the entire N-terminal lobe rotates as a rigid body. However in the transition to the activated state (IRKD 3P ), ␣C undergoes an independent (from the ␤ sheet) motion that entails a 12°rotation toward the C-terminal lobe and a 28°rotation about the helical axis (Fig. 3B). These movements of ␣C, which mainly occur through , changes at Phe-1054 and Thr-1055 at the base of the helix, are necessary to position protein kinase-conserved Glu-1047 (in ␣C) proximal to conserved Lys-1030 (in ␤3). Lys-1030 coordinates the ␣and ␤-phosphates of ATP in an active protein kinase configuration (30,31).
The IRKD DA structure suggests that the rotation of ␣C required for a properly configured active site relies on the precise positioning of Phe-1151 in the DFG motif. Although the position of Phe-1151 in the IRKD DA structure is roughly similar to that in the IRKD 3P structure, there are critical differences. In the IRKD 3P structure, Phe-1151 is buried deep in a hydrophobic pocket underneath ␣C (Fig. 4A). This pocket is composed of residues from ␣C (Glu-1047, Val-1050, Met-1051), from the ␣C-␤4 loop (Phe-1054, Val-1059), from ␣E (Leu-1123), and from ␤8 (Ile-1148). In contrast, the side chain of Phe-1151 in the IRKD DA structure points upward toward ␣C and is situated in a shallow hydrophobic pocket comprising the same residues as above (some with different side-chain rotamers) and additionally Phe-1128 in the segment preceding the catalytic loop (Fig.  4B). With Phe-1151 in this position, ␣C is sterically hindered from undergoing the movements that bring Glu-1047 into the active site. Moreover, conserved Asp-1150, which coordinates Mg 2ϩ , is pulled back from the active site vis à vis its position in the IRKD 3P structure (Fig. 3B).
The ATP analog (AMP-PCP) that was co-crystallized with IRKD DA is bound in the cleft between the two kinase lobes (Fig.  2) and is ordered throughout, including the ␥-phosphate. The conformation of AMP-PCP in the IRKD DA structure is different from the conformation of AMP-PNP observed in the ternary IRKD 3P structure and closely resembles the conformation of ATP and AMP-PNP in crystal structures of cyclic AMP-dependent protein kinase (30,31). In the ternary IRKD 3P structure, the ␥-phosphate of AMP-PNP is swung away from the hydroxyl group of the tyrosine substrate in the active site, presumably due to the imperfect fit with nitrogen rather than oxygen as the bridging atom between the ␤and ␥-phosphates.
Due to the incomplete rotation of the N-terminal lobe toward the C-terminal lobe in IRKD DA , AMP-PCP binds to the "roof" of the cleft (N-terminal residues) but not to the "floor" (C-terminal residues). Lys-1030 in ␤3 is hydrogen-bonded to the ␣-phosphate, but the ribose hydroxyl groups are not within hydrogenbonding distance to Asp-1083 in the C-terminal lobe, as in IRKD 3P . Moreover, only one Mg 2ϩ ion is evident in the IRKD DA structure, coordinated by Asp-1150 and the ␤and ␥-phosphates of AMP-PCP. Because of the retracted position of Asp-1150, the coordination of this Mg 2ϩ is weak: Mg-O distances Ն 2.4 Å. Due to the lack of lobe closure and the consequent positioning of AMP-PCP at the roof of the cleft, the second Mg 2ϩ ion present in the IRKD 3P structure, coordinated by Asn-1137 of the catalytic loop (C-terminal lobe), is absent in the binary IRKD DA structure.
Although the A-loop in the IRKD DA structure does not occlude the peptide binding site as in the IRKD 0P structure, K m(peptide) , unlike K m(ATP) , is not decreased in the Asp-1161 3 Ala mutant. 3 In the structure of ternary IRKD 3P , residues 1169 -1171 at the distal end of the A-loop are hydrogen-bonded via main-chain atoms to peptide substrate residues P ϩ1 through P ϩ3 (P 0 is the acceptor tyrosine), forming two short, antiparallel ␤ strands (13). In addition, the side chains of Leu-1170 and Leu-1171 are constituents of the binding pockets for the P 0 and P ϩ3 side chains, respectively. Thus, the disorder in the IRKD DA A-loop at the distal end results in a peptide binding site that is not fully formed, which is reflected in the high K m(peptide) . In contrast, in the gate-open conformation stabilized by A-loop autophosphorylation, the distal end of the A-loop is ordered even in the absence of peptide substrate. 4 Viscometric Analysis-The binding and chemical steps associated with an IRKD-catalyzed phosphorylation reaction are summarized in Scheme 1 for the experimental conditions where enzyme (E) is saturated with ATP (T), tyrosyl peptide (Y) binds with on-and off-rate constants k 2 and k Ϫ2 , the rate constant for the chemical step is given by k 3 , and the net rate constant for release of products ADP (D) and phosphotyrosyl peptide (pY) is given by k 4Ј .
The stepwise rate constants present in the steady-state kinetic parameters were derived using Cleland's method (32) and were established from the viscosity dependence of k cat and k cat / K m(peptide) (Fig. 5).
The principle behind viscometric analysis is that increased solution viscosity will affect the diffusion-dependent steps of substrate binding (k 2 and k Ϫ2 ) and product release (k 4Ј ) but not the chemical step (k 3 ), because the latter does not involve solute (substrate) exchange between the bulk phase and the enzyme's active site (23).
To identify whether chemistry is the rate-limiting step in steady-state phosphorylation of the peptide substrate Y-IRS939 by IRKD DA , the viscosity dependence of k cat and k cat /K m(peptide) was determined. The parameters (k cat ) rel and (k cat /K m ) rel are the ratio of k cat and k cat /K m , respectively, in the presence of viscogen versus the control reaction in aqueous buffer without viscogen. If the rate constant for a diffusion-dependent step is much smaller than for a diffusion-independent step, then the plot of (k cat ) rel or (k cat /K m ) rel versus rel will have zero slope. For IRKD DA , both global parameters were sensitive to changes in viscosity, increasing linearly with increasing viscogen (Fig. 5). The slope for (k cat ) rel versus rel was 0.8 Ϯ 0.1, and the slope for (k cat /K m ) rel versus rel was 1.2 Ϯ 0.2. For IRKD DA -catalyzed reactions without viscogen, k cat ϭ 9.6 Ϯ 0.4 s Ϫ1 and k cat /K m(peptide) ϭ 4.2 Ϯ 0.7 ϫ 10 3 M Ϫ1 s Ϫ1 . From these values, we calculate k 3 ϭ 59 s Ϫ1 and k 4Ј ϭ 14 s Ϫ1 . 5 Because these rate constants differ by only 4-fold, the steady-state rate constant of the reaction (k cat ) is partially limited by chemistry and partially by product release (Equation 1). These are approximately the same values of k 3 and k 4Ј determined for the activated kinase, IRKD 3P : 46 s Ϫ1 and 11 s Ϫ1 , respectively. 2 Denaturation Experiments-Denaturation of IRCD DA (46-kDa form of the mutant) in guanidinium chloride was monitored using fluorescence and is presented (Fig. 6) as the change in centroid of the emission spectrum (defined in Ref. 18). The data are compared with denaturation profiles from unphosphorylated and phosphorylated wild-type IRCD taken from previous work (18). As before, there are three transitions in the denaturation profile. Denaturation over transition I follows the same pattern observed for phosphorylated wild-type IRCD, and denaturation over transition III follows the same pattern as observed for unphosphorylated wild-type IRCD. This indicates that the Asp-1161 3 Ala substitution, while releasing the A-loop from the gate-closed conformation, does not otherwise alter the intrinsic conformational flexibilities within the A-loop or the two kinase lobes.
The conformational free energy in the unphosphorylated, gate-open mutant (IRCD DA ) was calculated from these results: ⌬G H 2 O ϭ Ϫ2.8 kcal/mol over transition I and Ϫ8.5 kcal/mol over transition III, for a net free energy of unfolding ⌬G H 2 O,net ϭ Ϫ11.3 kcal/mol. Compared with values published previously for the basal and activated state wild-type kinase (⌬G H 2 O,net ϭ Ϫ14.1 and Ϫ10.2 kcal/mol, respectively) (18), the overall differ- FIG. 5. Viscosity dependence of the IRKD DA -catalyzed reaction. Plots of (k cat ) rel (filled circles) and (k cat /K m ) rel (open circles) versus relative viscosity rel are shown using ATP and Y-IRS939 as substrates and sucrose as the viscogen. The control experiment for a chemically limited reaction was done using ATP␥S and Y-IRS939, and (k cat ) rel was determined (squares). Error bars show 1 S.D. Data were fit by linear regression (solid lines for (k cat ) rel and dotted lines for (k cat /K m ) rel ). ence in free energy (⌬⌬G H 2 O,net ) is 2.8 kcal/mol less inherent conformational stability in IRCD DA compared with the unphosphorylated wild-type kinase, mostly from transition I. The mutant showed 1.1 kcal/mol greater stability than the phosphorylated wild-type kinase, mostly from transition III. Therefore, the Asp-1161 3 Ala mutation in the A-loop results in a kinase that has a conformational stability intermediate between the basal state and the activated state.

DISCUSSION
Structural Studies-With few exceptions, phosphorylation of the A-loop in PTKs is one of the key regulatory mechanisms by which catalytic activity is increased (33). Crystallographic studies of PTKs and the related protein-serine/threonine kinases in their unphosphorylated (basal) and phosphorylated (activated) forms have provided a wealth of structural data for understanding this regulatory mechanism. The A-loop in the structures of basal state PTKs, including IRKD (12), the fibroblast growth factor receptor tyrosine kinase (34), Src (35)(36)(37), Hck (38,39), and Abl (40), exhibit a wide range of conformations. In the IRKD and Abl structures, one of the tyrosines in the A-loop (Tyr-1162 in IRKD, Tyr-393 in Abl) is bound in the active site, mimicking an exogenous tyrosine substrate. In the Src (37) and Hck (39) structures, the corresponding tyrosine in the A-loop, Tyr-416, does not mimic a tyrosine substrate but nevertheless contributes to the obstruction of the peptide substrate binding site. In contrast, in the structure of the unphosphorylated fibroblast growth factor receptor kinase (34), the A-loop tyrosines do not sterically occlude the peptide binding site, yet the configuration at the distal end of the A-loop hinders peptide binding.
The ATP binding sites in the crystal structures of basal state PTKs also display various degrees of accessibility. In the apoIRKD 0P structure (12), the proximal end of the A-loop passes through the ATP binding cleft. This is probably true for Abl as well, although the Abl structure contains a bound ATPcompetitive inhibitor (40), and therefore the course of the Aloop in the absence of inhibitor is not known. In the Src (37) and fibroblast growth factor receptor kinase (34) structures, in which an A-loop tyrosine is not mimicking a tyrosine substrate, an ATP analog is bound in the nucleotide binding site. However, in these two structures, the ATP analog is not bound productively by residues from both the N-and C-terminal lobes of the kinase, as in the ternary IRKD 3P structure (13).
Solution studies indicate that at millimolar concentrations of ATP present in cells, gate-open A-loop conformations are favored in basal state IRKD (15). Previous attempts to co-crystallize IRKD 0P with an ATP analog have been unsuccessful, probably because of the high K m(ATP) . The present crystal structure is of an A-loop mutant of IRKD (Asp-1161 3 Ala) that results in an approximate 10-fold reduction in K m(ATP) (the fold change in K m(ATP) also reflects the change in affinity). In this structure, an ATP analog is bound in the nucleotide-binding site, and the majority of the A-loop, including the three autophosphorylation sites, is disordered (i.e. assumes multiple conformational states). Even though the A-loop is largely disordered, the active site is clearly unobstructed. Thus, the IRKD DA structure and the apoIRKD 0P structure (12) are representative of gate-open and gate-closed conformations, respectively, of the unphosphorylated kinase. Prior to phosphoryl transfer, ATP is bound between the Nand C-terminal lobes of protein kinases, interacting with residues in both lobes (29). In the structure of IRKD DA with bound ATP analog, the nucleotide interacts with the N-terminal but not the C-terminal lobe, and the degree of lobe closure is intermediate between the IRKD 0P and IRKD 3P forms (Fig. 3A). In IRKD DA , lobe closure is hindered by Phe-1151 of the protein kinase-conserved DFG motif at the proximal end of the A-loop. Rather than tucking underneath ␣C into a deep hydrophobic pocket, as observed in the IRKD 3P structure, Phe-1151 is situated in a shallower pocket nearby (Fig. 4B). The distal end of the A-loop, which will serve as a platform for peptide substrate binding (13), is disordered in the IRKD DA structure, providing a rationale for the elevated K m(peptide) for this mutant (see below).
The IRKD DA structure indicates that proper positioning of conserved Phe-1151 is critical for the downward and inward rotation of ␣C and suggests that binding of ATP to the basal state kinase is not sufficient to induce the ␣C transition. In the IRKD 3P structure, conserved Glu-1047 in ␣C is proximal to conserved Lys-1030 in ␤3, orienting the lysine side chain for ATP (phosphate) binding (Fig. 3B). The optimal position (for catalysis) of conserved Asp-1150, which coordinates an essential Mg 2ϩ ion, is also dependent on the proper positioning of neighboring Phe-1151.
Upon autophosphorylation, the conformation of the A-loop is stabilized by short ␤-strand interactions between the A-loop and other C-terminal lobe residues and by electrostatic interactions involving the phosphoryl groups of Tyr(P)-1162 and Tyr(P)-1163 (13). In this configuration, the proximal and distal ends of the A-loop are "pulled taut," which favors the buried position of Phe-1151 beneath ␣C (Figs. 3A and 4A). Prior to autophosphorylation, the IRKD DA structure suggests that peptide substrate binding to the distal end of the A-loop is sufficient to reconfigure, at least transiently, the A-loop for catalysis.
Integration of Solution and Structural Studies-The IRK-D DA crystal structure is the third structure of the kinase domain of the insulin receptor, each of which represents a different conformational state. For each structure, we also have available kinetic parameters (Fig. 5) 2 and conformational free energies ( Fig. 6 and Ref. 18). Each type of analysis (structural, kinetic, and thermodynamic) shows that IRKD DA is intermediate between basal IRKD 0P and fully activated IRKD 3P and together provide further insights into the mechanism by which A-loop conformation regulates this kinase.
Every catalytic cycle comprises binding and chemical steps and conformational changes. These may be discrete or overlapping, depending on the relative rate constants for individual steps. For IRKD these steps were summarized in Scheme 1 (see "Results"). Comparing reactions catalyzed by the unphospho- rylated forms, IRKD DA and IRKD 0P , the dissociation constants for peptide substrate are nearly the same, but the rate constants for the chemical step (k 3 ) differ by almost 50-fold. Peptide binding equilibrates prior to the chemical step in the IRKD 0P -catalyzed reaction 2 but not in the IRKD DA reaction (see "Results"). The smaller rate constant for k 3 in the IRKD 0P versus IRKD DA reaction indicates a higher free energy barrier to phosphoryl transfer for IRKD 0P , making it 50 times less likely that a ternary complex will convert to a transition state complex. This barrier in IRKD 0P may arise structurally if the A-loop reconfiguration were less advanced than observed in the IRKD DA ⅐AMP-PCP binary complex. The conformational change in the A-loop needed to complete cleft closure and ␣C rotation in the IRKD 0P basal state would occur after peptide binding has equilibrated. If the conformational change is folded into the chemical step in this way, it could yield the 50-fold smaller rate constant in the basal state IRKD 0P catalytic cycle.
When IRKD DA and IRKD 3P are compared, k 3 is virtually the same for both. The same k 3 indicates that an equivalent free energy barrier exists between the respective ternary and transition state complexes. This suggests that the ternary IRKD DA complex should resemble the ternary IRKD 3P complex (13), with repositioning of the N-terminal ␤ sheet and ␣C accomplished in the ternary IRKD DA complex prior to the actual phosphoryl transfer event. A similar ternary complex for IRK-D DA and IRKD 3P would require a conformational change at the proximal end of the IRKD DA A-loop from the position observed in the binary complex (Fig. 3). The kinetics suggest that this might be accomplished through peptide binding, which is slower to IRKD DA than to IRKD 3P and reaches equilibrium before the chemical step for the latter but not the former (see "Results"). 2 These kinetic features of unphosphorylated IRK-D DA could be explained by a substrate-induced conformational change in the A-loop that occurs coincidentally with peptide substrate binding.
We envision that peptide binding begins at the P Ϫ1 and P 0 sites, which are fully open in the IRKD DA structure, and proceeds until the P ϩ1 and P ϩ3 residues become seated properly. The peptide substrate is captured by the faster chemical step of the catalytic cycle (k 3 Ͼ k Ϫ2 ), and thus binding does not equilibrate before the phosphoryl transfer occurs. Backbone hydrogen bonding and hydrophobic interactions involving the P ϩ1 and P ϩ3 side chains could impose order on Gly-1169 -Leu-1171 at the distal end of the A-loop, and peptide would be bound as observed in the ternary IRKD 3P structure (13). This peptideinduced ordering of residues at the distal end of the A-loop would trigger a concomitant structural rearrangement at the proximal end, burying Phe-1151 underneath ␣C and bringing Lys-1030, Glu-1047, and Asp-1150 into the proper alignment prior to the chemical step. We suggest that in the transition state of each IRKD-catalyzed reaction (including the basal state reaction), the A-loop conformation at the proximal and distal ends is essentially the same, with the final configuring of the A-loop occurring after substrate binding for IRKD 0P , during substrate binding for IRKD DA , and before substrate binding for IRKD 3P .
These and previous studies from our laboratories cited here show that a definable set of conformational changes are required for substrate binding and product formation on the insulin receptor kinase. Among these conformational changes, we have now identified a critical role for Phe-1151 of the conserved DFG motif in closure of the catalytic cleft and ␣C rotation.