The protein kinase C inhibitor bisindolyl maleimide 2 binds with reversed orientations to different conformations of protein kinase A.

As the key mediators of eukaryotic signal transduction, the protein kinases often cause disease, and in particular cancer, when disregulated. Appropriately selective protein kinase inhibitors are sought after as research tools and as therapeutic drugs; several have already proven valuable in clinical use. The AGC subfamily protein kinase C (PKC) was identified early as a cause of cancer, leading to the discovery of a variety of PKC inhibitors. Despite its importance and early discovery, no crystal structure for PKC has yet been reported. Therefore, we have co-crystallized PKC inhibitor bisindolyl maleimide 2 (BIM2) with PKA variants to study its binding interactions. BIM2 co-crystallized as an asymmetric pair of kinase-inhibitor complexes. In this asymmetric unit, the two kinase domains have different lobe configurations, and two different inhibitor conformers bind in different orientations. One kinase molecule (A) is partially open with respect to the catalytic conformation, the other (B) represents the most open conformation of PKA reported so far. In monomer A, the BIM2 inhibitor binds tightly via an induced fit in the ATP pocket. The indole moieties are rotated out of the plane with respect to the chemically related but planar inhibitor staurosporine. In molecule B a different conformer of BIM2 binds in a reversed orientation relative to the equivalent maleimide atoms in molecule A. Also, a critical active site salt bridge is disrupted, usually indicating the induction of an inactive conformation. Molecular modeling of the clinical phase III PKC inhibitor LY333531 into the electron density of BIM2 reveals the probable binding mechanism and explains selectivity properties of the inhibitor.

As the key mediators of eukaryotic signal transduction, the protein kinases often cause disease, and in particular cancer, when disregulated. Appropriately selective protein kinase inhibitors are sought after as research tools and as therapeutic drugs; several have already proven valuable in clinical use. The AGC subfamily protein kinase C (PKC) was identified early as a cause of cancer, leading to the discovery of a variety of PKC inhibitors. Despite its importance and early discovery, no crystal structure for PKC has yet been reported. Therefore, we have co-crystallized PKC inhibitor bisindolyl maleimide 2 (BIM2) with PKA variants to study its binding interactions. BIM2 co-crystallized as an asymmetric pair of kinase-inhibitor complexes. In this asymmetric unit, the two kinase domains have different lobe configurations, and two different inhibitor conformers bind in different orientations. One kinase molecule (A) is partially open with respect to the catalytic conformation, the other (B) represents the most open conformation of PKA reported so far. In monomer A, the BIM2 inhibitor binds tightly via an induced fit in the ATP pocket. The indole moieties are rotated out of the plane with respect to the chemically related but planar inhibitor staurosporine. In molecule B a different conformer of BIM2 binds in a reversed orientation relative to the equivalent maleimide atoms in molecule A. Also, a critical active site salt bridge is disrupted, usually indicating the induction of an inactive conformation. Molecular modeling of the clinical phase III PKC inhibitor LY333531 into the electron density of BIM2 reveals the probable binding mechanism and explains selectivity properties of the inhibitor.
Deregulated protein kinase activity causes a wide variety of human diseases, usually by producing an overactive kinase. This is consistent with the fact that most protein kinases in the cell are inactivated most of the time to ensure the integrity of signal transduction. Thus, the many diseases that are correlated with protein kinase deregulation, including the majority of all cancers, usually arise from mutations or other events that activate kinases, cause their overexpression, or disable their intracellular inhibition. The prevalence of kinase deregulation in disease clearly demonstrates the need for therapeutic protein kinase inhibitors, whereas the ubiquity and variety of protein kinases (collectively, the "kinome") necessitate precise target selectivity. Despite this seeming difficulty, several protein kinase inhibitors have been approved for human treatment or are in advanced clinical trials.
Crystal structure analyses of protein kinase inhibitor complexes reveal the intermolecular interactions responsible for ligand binding, and have thereby enabled structure-based rational design and optimization of kinase inhibitors. To date, crystal structures have been determined for some 30 protein kinases, representing some 6% of the 518 protein kinases in the human genome (1). Many of these structures have been complexes with protein kinase inhibitors, but most have shown an inactivated state often incompatible with inhibitor binding. Inactivity is associated most often with displacements of helix C, the major ␣ helix of the kinase N-lobe, with concomitant disruption of the active site salt bridge in the active site between a conserved lysine residue and a conserved glutamate located in the middle of helix C. Similarly often, the activation loop shows unproductive conformations, either blocking the active site, or locking helix C into an inactive conformation, or causing other structural states of the kinase incompatible with kinase activity. Often, these structures are "open" with respect to the "closed" conformation of kinases in catalytic conformations. In more than half of all protein kinase structures inactivity is associated with a steric block in the ATP-binding site (2)(3)(4).
Despite the variations in sequence, the fold of the active protein kinase catalytic domain is well conserved. Inactivated protein kinase structures differ more, but cluster into protein kinase subfamilies that reflect different inactivation mechanisms. As a consequence of the conservation of the active structure, many properties can be analyzed with respect to relatively few sequence positions that define that property. A centrally important example of such a property is the selectivity of binding at the ATP binding pocket. Protein kinases share a common bi-lobal catalytic domain structure that forms the ATP-binding site at the lobal interface. ATP binds at this interface via interactions with some 15 residues of the protein, including about 10 side chain interactions that therefore are especially important as potential determinants of ATP site * This work was supported in part by the Bayerische Wirtschaftsministerium. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The inhibitor selectivity. Thus, the essential binding properties of ATP and other ATP site ligands can in many cases be simulated for a particular protein kinase target by a limited set of point mutations of a closely related protein kinase. The construction of such hybrids has been demonstrated for kinases (5)(6)(7). Even though subtle differences in structure or flexibility can lead to quite different overall reaction kinetics, for practical ligand design purposes, most or all protein-ligand interactions will be revealed or can be modeled using the surrogate kinase approach. That a single residue can be identified as the principal selectivity determinant for an inhibitor type by mutation of a series of kinases verifies this approach (8).
Along these lines, the cAMP-dependent protein kinase (PKA) 1 has been used as a surrogate kinase for co-crystallization with several protein kinase inhibitors, such as H7, H8, and H89 (9), staurosporine (10), balanol (11), and recently the Rhokinase inhibitors Y27632, H1152P, and Fasudil (12). The complexes with staurosporine, balanol, and Fasudil are in partially open conformations, in contrast to the other inhibitor/substrate-PKA complexes that are in closed conformations. In these three structures the hydrogen bond between the activation loop Thr 197 phosphoryl group and His 87 from helix C of the N-lobe is not formed as a result of the partial opening of the cleft via rotation of the N-lobe with respect to the C-lobe.
Mutants of PKA have also been designed to improve its value as surrogate kinase for PKB inhibitors (7). Because of its close relationship to PKC and its well established crystallization conditions, PKA is currently also the best model system for studying PKC inhibitors. So far, co-crystallization attempts with PKA and the flexible bisindolyl maleimide (BIM) cognates of staurosporine, described as PKC inhibitors (13, 14 -17), have failed to produce crystal structures. However, the triple mutant V123A,L173M,Q181K of PKA␣ (PKAB3), originally designed as a model for PKB (7), has formed high quality crystals of a BIM inhibitor complex. A sequence alignment of PKA␣ and PKB␣ with PKC isoforms (Table I) shows how the PKAB3 triple mutant is similar to PKA as a surrogate for PKC with the additional ability to model inhibitor-methionine interactions for three conventional PKC isoforms.
Three distinct subfamilies of PKC isoforms can be defined according to their essential activators: conventional PKCs (␣, ␤I/II, and ␥) require phosphatidylserine, diacylglycerol, and Ca 2ϩ ; novel PKCs (␦, ⑀, , and ) need phosphatidylserine and diacylglycerol but not Ca 2ϩ ; atypical PKCs ( and ) are insensitive to both diacylglycerol and Ca 2ϩ although phosphatidylserine regulates activity (for reviews, see Refs. 18 and 19 and citations therein). Furthermore, additional lipid mediators, like fatty acids and lysophospholipids, have been shown to influence the catalytic activity of PKCs (reviewed in Ref. 20). In general, interaction of PKCs with the activators leads to phosphorylation of a threonine residue on the activation loop of all PKC isoforms and additionally of a serine or threonine residue in the hydrophobic motif of the conventional and novel PKCs. The atypical PKCs possess a glutamate at the hydrophobic motif phosphorylation position that intrinsically performs the activation function (for reviews, see Refs. 18 and 21). PKC isoforms are involved in nearly all essential cell processes, PKC␣, for example, is involved in regulation of proliferation, apoptosis, differentiation, cell migration, adhesion, among other cellular and pathogenic processes (for a review, see Ref. 22).
Despite its early identification and importance in cancer research, no PKC crystal structure has been reported to date. The only available co-crystal structures of PKC inhibitors with a kinase target are those of the relatively unselective staurosporine (IC 50 , PKC 5 nM (23)) and its closely related derivative UCN01 (IC 50 , PKC␣ 29 nM (24)). Besides co-crystallization with PKA (10) (Protein Data Bank code 1STC), the extended and rigid planar staurosporine has been crystallized with Cdk2 (25) (Protein Data Bank code 1AQ1), CSK (26) (Protein Data Bank code 1BYG), and others. UCN01 (7-hydroxystaurosporine) has been co-crystallized with Cdk2 (27) (Protein Data Bank code 1PKD) and Chk1 (28) (Protein Data Bank code 1NVQ). The bisindolyl maleimide class of PKC inhibitors is derived from staurosporine by elimination of a single bond that converts the extended planar aromatic group into the three aromats of the compound names with corresponding additional degrees of flexibility. One PKC inhibitor of this class, LY333531, shows PKC isoform specificity (e.g. 80-and 60-fold selectivity for PKC␤ I and PKC␤ II over PKC␣ (29) and is in phase III clinical trials for diabetic retinopathy and diabetic macular edema (Ref. 30 and citations therein). The effects of the additional flexibility of BIM inhibitors on the structural binding modes, and the means by which this alteration can introduce selectivity to the inhibitor have not been explained.
Here we present the crystal structure of bisindolyl maleimide 2 (BIM2) in a complex with the triple mutant V123A,L173M,Q181K (PKAB3) of PKA. By means of this surrogate kinase approach, we identify the key binding modes and can evaluate their significance for PKC. The asymmetric unit of the crystal structure consists of two protein-inhibitor complexes, with different conformations of the two kinase molecules, bound with opposite orientations of different conformers of BIM2.  does not interact with BIM2 but has been exchanged to prevent a rotation of the glutamine towards the ATP-binding site (7). Phe 327 does not interact with Bim2, but probably with LY333531. PKC␤I and PKC␤II are merged because they do not differ within the kinase domain.
site salt bridge between Lys 72 and Glu 91 is disrupted, a characteristic of inactive kinase conformations. The two different inhibitor binding modes are enabled by the inherent symmetry of the maleimide moiety and the rotational freedom available to the indole moieties. The similarity of BIM2 and LY333531 ( Fig.  1) allows modeling of LY333531-PKC interactions and suggests an explanation for the selectivity properties of the inhibitors.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Recombinant mutated bovine C␣ catalytic subunit of the cAMP-dependent protein kinase (PKAB3 (7)) was solubly expressed in Escherichia coli BL21(DE3) cells and then purified via affinity chromatography and ion exchange chromatography as described earlier (9). Two positions distinguish bovine (Asn 32 and Met 63 ) from human PKA (Ser 32 and Lys 63 ). 4-Fold phosphorylated protein was used for crystallization of BIM2.
Activity Tests-The determination of enzyme activity was accomplished by an ATP regenerative NADH consuming assay according to Ref. 31. After the addition of 0.42 mM MEGA 8 (ensures solubility of the inhibitor) to the assay mixture (100 mM Mops, pH 6.8, 100 mM KCl, 10 mM MgCl 2 , 1 mM phosphoenolpyruvate, 0.1 mM Kemptide, 1 mM ␤-mercaptoethanol, 15 units/ml lactate dehydrogenase (Sigma), 8 units/ml pyruvate kinase (Sigma), 0.21 mM NADH) we have added the successive Me 2 SO/inhibitor solution and the enzyme and started the reaction with ATP. The decrease of NADH was measured as time-dependent at ϭ 340 nm with three independent measurements per data point.
Data Collection and Structure Determination-Diffraction data were measured at the Deutsches Elektronen Synchrotron (DESY, Hamburg) from frozen crystals on a CCD detector (Mar research) at 1.05 Å wavelength. The data were processed with the programs MOSFLM and SCALA. The crystals have orthorhombic symmetry (P2 1 2 1 2 1 ) with cell constants 82.06, 89.00, and 116.38 in a crystal packing arrangement not previously reported (Table II). The structure was determined by molecular replacement using MOLREP from the CCP4 program suite. 2 As starting model we chose a PKA-PKI-(5-24)-staurosporine complex (1STC (10). Calculation of Matthews coefficient and solvent content suggested two molecules in the asymmetric unit with 52.3% solvent. Indeed, very good monomer rotation and translation function solution was found, enabling determination of a second molecule with an Rfactor of 47.5. A first map calculated after rigid body refinement (Rfactor 45%) showed that PKI was not present in the complex, despite its presence in crystallization solutions as usual. Furthermore, difference densities in the B molecule showed that large segments of the molecule needed to be remodeled. We deleted the segments (the entire N-terminal lobe up to 123 and the C-terminal residues from 315 onwards) of about 150 amino acids together and calculated new maps. After several circles of model building and refinement the R-factor fell to 28.8% (R-free 33.9%) and most parts of molecule B were replaced. In molecule A only residues 315-350 were omitted and rebuilt. The segment 317-332 (330 for molecule B) remained undefined. Phosphorylation sites were found at Ser 139 , Thr 197 , and Ser 338 . Ser 10 is not resolved. Water molecules were automatically inserted using CCP4 programs PEAK-MAX and WATPEAK and visually inspected. Finally, the inhibitor molecules were built and the whole complex was further refined. Refmac 5.1.24 was used for refinement, and MOLOC 3 was used for model building and graphical modeling. For data and refinement statistics, see Table II.
Superimposition, Calculation of Secondary Elements, and Angle Determination-Superimpositions were performed using the programs Insight II or MOLOC. In the case of MOLOC, amino acid residues 150 -300 were used for superimposition of structures 1STC, 1CTP, 1CMK, and 1J3H and both molecules of the here presented structure on structure 1CDK as basis. The secondary elements were calculated with the program InsightII using the Kabsch-Sander algorithm. The angle determination was performed in the following way. The backbone of each first and last two amino acids of the helices were taken to calculate the center of masses using the gromacs package. 4 These center of masses defined the top and bottom of the helix axes. The coordinates of the top and bottom were used to create vectors in the R3 and with cos␤ ϭ x⅐y/͉x͉⅐͉y͉ (x and y are the vectors defining the helix axes). The angle between these vectors of two different structures was calculated.
Sequence Alignment-Sequence alignments were performed using ClustalW. 5  Thus, the clearest determinant of the selectivity of protein kinase inhibitors is the amino acid composition of the catalytic ATP-binding site. It follows that amino acid exchanges in the catalytic site of one kinase can be useful as a surrogate for another kinase, reported, for example, for PKA mutants that mimic PKB (7). For this study we used the triple mutant 2 www.ccp4.ac.uk/main/html. 3 www.moloc.ch. 4 www.gromacs.org. 5 www.ebi.ac.uk/clustalw. PKAB3 (V123A,L173M,Q181K), which was originally constructed to model the adenine-binding site of PKB in PKA. Sequence alignment of PKA, PKB, and PKC shows that one of these mutations, namely L173M (numbering of PKA), introduces a residue that is conserved in PKB and the three classical PKC isoforms ␣, ␤I/II, and ␥. The other PKC isoforms have, like PKA, a leucine residue in the corresponding position. The second exchange, V123A, does not occur among PKC isoforms, but the important contacts of residue 123 to inhibitors are backbone contacts and presumably relatively unaffected by the side chain. The third exchange, Q181K, was chosen only after observation that the double mutant (V123A,L173M) lead to a new rotamer conformation of the glutamine of PKA that placed the amide group into the ATP binding pocket (7). This mutation introduces a residue conserved as a lysine among all PKC isoforms.

PKAB3 as Model for the PKC-ATP-binding
Overall Structure-The triple mutated recombinant catalytic subunit of cyclic AMP-dependent PKA and the PKC inhibitor BIM2 formed crystals notably lacking the pseudo-substrate peptide PKI(5-24), although the peptide was present under the standard crystallization conditions. The 4-fold phosphorylated PKA is disordered at the N terminus including the phosphorylation position of (p)Ser 10 . The region between residues 317 and 332 is also not visible. The corresponding dimer formed the asymmetric unit in an orthorhombic space group P2 1 2 1 2 1 (Table II) with cell constants and a packing arrangement previously unobserved for PKA (a ϭ 82.1, b ϭ 89.0, c ϭ 116.4). This arrangement is apparently induced by BIM2 binding and arises from the two new N-and C-domain conformations of PKA in the crystal. One of the PKA complexes of the asymmetric unit (molecule A) is in an intermediate open conformation, similar to that observed for the PKA complex with staurosporine (1STC (10)) and also HA-1077 (1Q8W (12)), whereas the other, molecule B, is in the most open conformation described so far for PKA (Fig. 2). The inhibitors in both molecules A and B occupy the ATP-binding site, but different conformations of the inhibitor bind with different orientations in the two molecules (Fig. 3).
Many different relative N-and C-lobe orientations have been observed for PKA. One measure to identify the closed conformations is the existence of an H-bond between the imidazole of His 87 from helix C and the phosphoryl group of Thr 197 from the activation loop. Structures for which this His 87 -(p)Thr 197 contact is broken because of the opening movement of the N-lobe exhibit a variety of open conformations (Fig. 2). The hinging movements of PKA that cause the opening occur primarily within the dipeptide segment of Gly 125 and Gly 126 positioned between the ATP binding residues and helix D, as defined originally by Olah and co-workers (34). Rotations of helix C can be taken as a measure of the opening of PKA structures, and ranks structures from the most closed in structure 1CDK through other open conformations of PKA such as staurosporine bound (1STC (10), or the apoenzyme structures (1CTP (35,36) and 1J3H (37)) to the most open conformation of BIM2MolB from this study (Table III and  ith respect to 1CDK. The greatest rotations occur in the structures 1J3H-molecule A (14.2°) and BIM2MolB (14.6°) (Table III). In addition to the hinge movement shared by all structures, BIM2MolB shows a rotation of its N-lobe by 15°when compared with molecule A of 1J3H. This rotation is clockwise for the N-lobe from a viewpoint above the N-lobe toward the C-lobe. The axis of this rotation is parallel to the viewing axis. The C terminus of the helix is fixed to the C-domain, which results in a rotation of helix C of one-half the total rotation relative to both domains, similar to the motion of a piston. The rotation of helix C relative to the N-domain contributes to the disruption of the salt bridge between Lys 72 and Glu 91 , a characteristic of inactive kinase structures (2)(3)(4). Another striking conformational difference of BIM2MolB to any other PKA structure is the displacement of the hinge region that connects the N-and C-lobes (amino acids Glu 121 -Pro 125 ) and forms the most important inhibitor and ATP binding interactions. This displacement amounts to 3.38 Å at the C␣ of Ala/Val 123 and  2. Root mean square differences of the C␣ positions of the four PKA structures relative to closed conformation PKA-PKI-AMP-PNP structure 1CDK. The C-terminal lobes of the molecules were superimposed to assess the extent of motion of the N-terminal lobe (and hydrophobic motif binding C-terminal strand) relative to the Cterminal lobe. 5.05 Å at the carbonyl of Pro/Ala 124 in comparison to molecule A of mouse PKA structure 1J3H (37) (Fig. 4).
The Bisindolyl Maleimides, Cognates of Staurosporine-Many cognates of staurosporine have been identified as potential therapeutic kinase inhibitors; some are in clinical trials. Target complex structures available for them include the rigid planar inhibitor staurosporine itself (e.g. Refs. 10 and 38) and the closely related UCN01 (28,39). Here we report a kinase complex structure for a non-planar, flexible staurosporine cognate of the bisindolyl maleimide class (15)(16)(17). The structure of PKAB3 and bisindolyl maleimide 2 demonstrates the similarities and differences of binding modi between the planar staurosporine and the flexible bisindolyl maleimide 2, indicating the factors that govern selectivity in this inhibitor class. It is unique in that two inhibitor conformations bind the same site of the same enzyme in two different orientations, and furthermore, in a single crystal. Enzyme inhibitors that bind with different orientations have been observed previously, for example, serine proteinase inhibitors that as a function of pH bind with opposite orientations (46), or kinase inhibitors that bind different kinases with unequal orientations (40).
The chemical structures of the 3 inhibitors discussed in this article are shown in Fig. 1. They originate from a single chemical class and share the symmetric basic architecture of a maleimide flanked by two indole rings. In the case of the indolocarbazole staurosporine, the entire structure is rigidified by an additional C-C bond (C7-C8) that links the indoles and maleimide into an extended aromatic and planar system, and additionally via cyclic bonding of the sugar moiety from the two indole nitrogen atoms (Fig. 1C). This extended and rigid planarity of staurosporine is shown in many crystal structures (with PKA and PKI (10), LCK (41), and CDK2 (27)) to induce an ideal planar fit in its many kinase hosts, which may cause its relatively low specificity. Lacking the C7-C8 bond, and with a single bond connecting a pyrrolidine tail, BIM2 possesses degrees of freedom that allow rotation of the indole rings out of a single plane and that allow a large number of conformations for the pyrrolidine tail (Fig. 1A). The crystal structure shows how these degrees of freedom are utilized (Fig. 5, a and b). The small molecule crystal structure of BIM4, a derivative consisting only of the maleimide head group and the two indoles, shows a comparable rotation of one indole as found in our structures, indicating an energetically preferred conformation (15). LY333531, also known as ruboxystaurin, seems to represent an intermediate between staurosporine and BIM2 with respect to flexibility of the indole moieties. Although the indoles are not directly linked as in staurosporine, the 6-atom ether linker connecting the N-atoms of both indoles forms a cycle that restricts the orientations available to the indole rings (Fig. 1B).
Binding of BIM2, Molecule A-In the partially open structure of PKAB3 in molecule A, BIM2 forms 4 H-bonds with the enzyme in molecule A. The symmetrical maleimide makes two hydrogen bonds to the hinge region between N:19 and Glu 121 :O and O32 and Val 123 :N (Fig. 6a). A third contact of the maleimide is formed from O33 to Thr 183 :OG1, whereas the N30 of the pyrrolidine group connects to Glu 170 :OE2 (Table V). The contact to the side chain of Glu 170 is possible, because Glu 170 is in a different rotamer conformation. Another residue, Glu 127 , also shows a new rotamer, because its original position is occupied by the pyrrolidine group. Glu 170 , together with Glu 127 , are essential residues for the recognition of the substrate consensus arginine residues, including those of PKI. The interaction of BIM2 with both residues in different rotamer conformations might explain the absence of PKI (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24) in the structures.
The total number of van der Waals contacts between inhibitor and enzyme is 127. In detail, the maleimide is coordinated by the hinge region (Glu 121 -Ala 123 ), Ala 70 , Met 120 , Tyr 122 , and Thr 183 . The main interacting amino acid residues to indole I are Lys 72 and Thr 183 , whereas the peripheral glycine loop residues Leu 49 and Val 57 coordinate indole II. The pyrrolidine moiety interacts mainly with Phe 54 , leading to a distorted conformation of the glycine loop, and makes side chain contacts to Glu 127 and Glu 170 . Met 173 is centered directly under the BIM2 and has contacts to all moieties of the inhibitor (Table IV).
Binding of BIM2, Molecule B-In the open enzyme structure of molecule B, BIM2 binds with an opposite orientation. Despite the displacement of the hinge region in this structure, it forms 2 H-bonds to the same hinge region atoms as in molecule A, between N19 and Glu 121 :O, respectively, and O33 and Val 123 :N (Fig. 6b). This is possible because of the symmetrical nature of the maleimide head group, and because the BIM2 inhibitor occupies a different location with respect to molecule A. First, BIM2 follows the outwards directed hinge region displacement to keep the contact for these hydrogen bonds. Second, BIM2(B) is rotated in the plane of its maleimide group so that in an overlay of molecules A and B, the indole II group of BIM2(B) comes close to the indole II moiety of BIM2(A). Apart from this, no further hydrogen bonds to the enzyme are established. The total number of inhibitor to protein van der Waals contacts are 91 in this conformation. The interactions of BIM2 in molecule B are limited to contacts to the N-terminal lobe and the hinge region. No contacts to the C-terminal lobe are established, a consequence of the open conformation of the kinase, the displaced position of the hinge region, and the shifting rotation of the inhibitor. In detail, the maleimide is coordinated by the hinge region (Glu 121 -Gly 125 ), Ala 70 and Met 120 . Indole I is tightly packed by Leu 49 , Tyr 122 , and Pro 124 , whereas indole II interacts with Val 57 and Lys 72 . Most striking is the extensive contact of indole I to Pro 124 , which makes 16 van der Waals contacts, one-sixth of the total number of contacts of this inhibitor with the enzyme. We are not aware that any other protein kinase inhibitor interacts with the side chain of a residue in the homologous position. This unusual contact appears to be a consequence of the inwardly rotated conformation of indole II, which can only be accommodated when the BIM2(B) molecule and the N-terminal lobe are rotated, which wedges the inhibitor between the side chains of Pro 124 and Val 57 .
Comparison of the Two Binding Modes-A comparison of the binding modes of BIM2 in molecule A and molecule B highlights the similarities but especially the significant differences.

FIG. 4. Superposition of the structures for 1J3H molecule A and
BIM2MolB. Superposition of 1J3H molecule A (orange) and BIM2MolB (blue) are shown.

FIG. 5. Stereo view of the difference density maps prior to inhibitor modeling, contoured at 2 for BIM2 bound to molecule A (a) and molecule B (b).
In molecule A, the entire inhibitor geometry is unambiguously defined, whereas in molecule B, there is evidence of some disorder especially at the pyrrolidine moiety. Therefore, a second contouring at 1.5 is added. Building the unambiguously defined portions of the inhibitor identified the major binding mode with no model bias.
The principal similarities involve the maleimide groups, with their identical contacts to the hinge (Table V) and extensive contacts to Ala 70 , a conserved residue from ␤-strand 3. Assuming that these identical interactions represent the essential binding interactions, the differences in binding modes observed in molecules A and B result from the different lobe configurations of the two kinase monomers. A particularly notable difference between the two binding modes is the reversal of the overall inhibitor orientation (Fig. 7). Although the bisindolyl maleimide scaffold itself is symmetric, the pyrrolidine substitution on indole I breaks the symmetry and defines an overall orientation. Thus, whereas the hydrogen bonding pattern of the maleimide and hinge loop remains the same between the two binding modes, the hinge hydrogen bond acceptor of the inhibitor is alternately the oxygen distal or proximal to the substituted indole I in molecules A and B, respectively (Fig. 6,  a and b). The electron densities of the inhibitors (Fig. 5, a and  b) confirm these two orientations, although there is evidence for some disorder in molecule B, most likely associated with different pyrrolidine binding geometries of low occupancy. The reversed orientation is accompanied by a different arrangement of the indole moieties (Figs. 5, a and b, and 7a). Viewed from the N-lobe (as "up"), the indole moieties all rotate upwards out of the plane of the maleimide and maintain intramolecular contact. In molecule A, however, they are oriented to the "right" or toward the bulk of the kinase domain, whereas in molecule B, they are oriented to the "left" or away from the kinase domain. This leads to generally different sets of interactions between the inhibitor and kinase residues. An exception is Val 57 , which is in both orientations of inhibitor one of the residues that contacts indole II. Both orientations of the pyrrolidine form a face to face set of van der Waals interactions with indole I. The pyrrolidine orientation in molecule B seems to be stabilized only by these intramolecular interactions, because no contacts to the kinase are established. In summary, BIM2 adopts two very different conformations, with different rotamers for pyrrolidine-tethered indole I (␦ 90°) and the untethered indole II (␦ 30°) (Fig. 7a).
Because of its symmetry, the maleimides bind similarly in both molecule A and molecule B; the indoles, pyrrolidines, and linkers bind very differently. The maleimides form functionally identical contacts to the same hinge region atoms and six and seven van der Waals contacts to Ala 70 , a conserved residue from ␤-strand 3. The reversal of the BIM2 orientation, as defined and constrained by the asymmetric pyrrolidine linker, leads to very different conformational adaptations to the binding site. The conservation of interactions between indole II and both indole I and pyrrolidine leads to an inwardly rotated conformation of indole II and an outwardly rotated indole I. The reversal of the inhibitor orientation replaces the outward conformation of indole I with an inward conformation of indole II in the same binding site, and vice versa at the other indole binding site. In molecule A, indole II is rotated out of the maleimide plane toward the glycine flap, interacting with Leu 49 and Val 57 from the adjacent N-and C-terminal anchoring ends of the glycine flap, respectively. These close interactions of indole II with Val 57 replace the more distributed interactions that are observed in planar structures such as staurosporine. In molecule B, indole II also interacts with Val 57 , but in this case, indole II is positioned at the inner or solvent shielded site of the ATP pocket. Thus, the interactions between indole II and Val 57 require rotations of both the inhibitor and the N-lobe ␤-sheet relative to their positions in molecule A, in opposite directions, to bring the interacting partners together. Also associated with the effective rotation of the inhibitor toward solution in molecule B is a reorientation of Lys 72 that leads to occupancy of the volume occupied by the inhibitor in molecule A. The effective rotation of BIM2 in molecule B also involves new contacts with Pro 124 from the hinge region, which also undergoes a conformational change apparently in response to BIM2 binding in this orientation. In summary, the reversed orientation of BIM2 in molecule B is associated with a rotation of inhibitor and opposite rotations of glycine loop and hinge region atoms that, together with a conformer change of Lys 72 , embedded the inhibitor aromats in a new set of interactions.
Comparison of the BIM2-Molecule A Complex and 1STC-The kinase lobe configuration of the PKAB3 molecule A-BIM2 complex resembles that of PKA-staurosporine (Fig. 2), facilitating the comparison of the two inhibitors. The most striking difference between staurosporine and BIM2 is the flexibility of the latter. Staurosporine, entirely rigid ( Figs. 1 and 7b), would be expected to induce any observed conformational changes of the enzyme. Indeed, almost all side chains in the vicinity of staurosporine move to expand the binding site, as do the peptide backbones at Thr 183 , and Phe 327 . In addition, Phe 54 from the glycine flap moves toward the inhibitor molecule (Fig. 7a  (10)). One consequence of the staurosporine architecture is that a large number of atoms are in close proximity available to form van der Waals interactions with suitable protein residues. This is notable especially for inhibitor interactions with residues Leu 49 , Met 120 , and Glu 127 (Table IV). However, the corollary expectation that a more flexible molecule might induce less change in the protein because of flexibility and greater propensity to be changed is clearly not confirmed for BIM2. In molecule A, and more dramatically in molecule B, several conformational changes are observed that are more pronounced and far reaching than in the PKA/staurosporine structure. This is vivid confirmation of an understanding that the protein kinase adopts a variety of conformations that is not readily apparent by a limited set of crystal structures (with low ligand variation and/or few crystal packing arrangements), at least in response to various small molecule ligands, and thus presumably also in response to a much greater variety of influences in a physiological environment.
Induced fit displacement of the glycine loop has been observed since the first PKA-inhibitor complexes (9). A striking feature of the molecule A structure in comparison to other PKA structures is a tilting of the glycine loop toward BIM2 (Fig. 8b). Especially the ␤ turn of residue Phe 54 at the tip of the glycine loop forms a number of hydrophobic contacts to the methyl group (C31, 5 of 7 contacts) substituent of the pyrrolidine ring. The interactions between staurosporine and Phe 54 occurs via the methyl group (C35, 2 of 2 contacts) substituent of the pyran ring (Fig. 8a). Superposition of the complex structures of 1STC and molecule A reveals that these two inhibitor contact atoms are displaced by some 3 Å with respect to each other. Similarly, the Phe 54 :CZ atoms of the superimposed structures are displaced by 2.8 Å. Thus, as in 1STC, BIM2 binding induces a motion of Phe 54 toward the inhibitor. However, unlike staurosporine, this causes the entire glycine loop ␤ turn to tilt downwards, thereby displacing residues 49 -59 by up to 5.5 Å (Phe 54 : O). PKA binds elongated ligands such as AMP-PNP (1CDK) or The PKAB3 structure has an alanine at this position. For a better comparison of the van der Waals (vdW) contacts the number in parentheses count the interaction of the CG1 and CG2 of the valine of PKA, whereas the first number corresponds to the interaction up to the CB.
c The PKAB3 has, like classical PKC isoforms, a methionine at this position and is therefore not directly comparable. d NR, no resolution. e The number in parentheses represent the number of different water molecules. f Includes the contacts mentioned in b. balanol (1BX6) also via direct interactions with Phe 54 , but these ligand geometries do not induce the glycine loop motions. Thus, this highly conserved phenylalanine seems to play a key role in inhibitor binding; mutant studies are required to evaluate the energetic consequences although glycine loop flexibilities will complicate their interpretation. The many interactions (24) of Val 57 from the glycine loop with the planar surface of staurosporine (Table IV) are restricted to five contacts with indole II in BIM2(A) ( Table IV).
Other notable differences in the binding mode of BIM2 and staurosporine concern the number of van der Waals contacts between BIM2 and Tyr 122 (Table IV). A comparison with the staurosporine structure reveals a loss of 6 contacts because of the inwardly rotated position of the indole II of BIM2 (Fig. 6). In addition, an interaction between Phe 327 and the inhibitor is not established, whereas staurosporine exhibits seven contacts to this amino acid. In the PKA-BIM2 structure described here, Phe 327 is flexible (with no apparent electron density, but a comparison with the PKA/staurosporine structure suggests that the distance between Phe 327 and indole II of BIM2(A) would be too far for a contact). The contact between the inhibitor and Phe 327 may be the major reason why this stretch is ordered in the partially open structures such as 1STC (10), 1BX6 (11), and 1Q8W (12). In the open apoenzyme structures 1CTP (42,43) and 1J3H (both molecules (37)), the region around Phe 327 is also unresolved. In the closed apoenzyme structures (e.g. 1Q61 und 1Q62 (7)), the whole enzyme archi-tecture is more compact and the structures show this entire polypeptide to be ordered.
PKAB3-BIM2 Binding Versus Hypothetical PKA-BIM2 Binding-An evaluation of the likely inhibitor binding interactions in PKA requires an assessment of the effects of the three point mutations of PKAB3 relative to PKA. As described above, the exchange of Q181K (PKA position PKAB3) was introduced to prevent an artificial rotamer conformation of the Gln 181 that apparently resulted from the V123A exchange (7). As such, neither Gln 181 nor Lys 181 is expected to directly affect inhibitor binding. The mutation V123A expands the volume of the adenine binding cavity, an alteration that potentially could alter the character of the hinge hydrogen bonding interactions at this position, although none has been evident from crystal structures so far. BIM2 in molecules A and B has 8 to 7 van der Waals contacts to Ala 123 (out of totals of 127 or 91, respectively). In silico exchange to valine reveals the formation of up to three van der Waals contacts in molecule A and no additional van der Waals contact in molecule B. In comparison, staurosporine has 13 hydrophobic contacts to the corresponding valine (Table IV); in silico exchange to alanine removes only three of them. The most interesting exchange is L173M. In molecule A this methionine is surrounded by the aromatic maleimide and indole core of the inhibitor, forming 13 van der Waals contacts mostly via the terminal methyl group. In silico mutation back to leucine basically preserves the side chain volume, but shortens its extension, thereby removing most FIG. 7. a, stereo view to compare the conformations of the two BIM2 molecules (yellow, BIM2 molecule A; green, BIM2 molecule B). The compounds are superimposed using the chemically (not spatially) equivalent atoms of the maleimide moiety (e.g. mapping O33 molecule A to O33 of molecule B). The two orientations of BIM2 are related by a 2-fold symmetry axis through the maleimide moiety. The indole orientations are not related by this symmetry transformation, however, as they are rotated in each case toward the N-terminal lobe. b, comparison of the binding conformers of the two BIM2 molecules (yellow, BIM2 molecule A; green, BIM2 molecule B) and the staurosporine (color by atom). The compounds are fitted using spatially (not chemically) equivalent atoms of the maleimide moiety (e.g. O33 molecule A on O32 of molecule B). Compared with staurosporine, the rotation of the indole rings out of the plane is striking, as is the overall equivalence of the spatial volumes occupied by the inhibitors.
hydrophobic contacts under the assumption that the BIM2 in PKA shows the same indole rotamers. Determination of the IC 50 of BIM2 to PKA and PKAB3 reveals that PKAB3 (IC 50 , 6.35 M) is less sensitive to the inhibitor than PKA (IC 50 , 2.94 M) (Table VI). Thus, in PKA, BIM2 might be shifted somewhat to optimize contacts here.
However, previous efforts to co-crystallize wild type PKA together with BIM2 have failed, attempts with PKAB3 have succeeded. The three residues that differ between PKA and PKAB3 seem at first glance unlikely to cause a difference in crystallization properties. The exchange Q181K is at the surface of the protein and could theoretically alter crystal contacts, but, there are no crystal contacts to either Gln 181 or Lys 181 or adjacent residues in the relevant crystals.
The exchanges of V123A and L173M are clearly shielded within the ATP pocket and should not directly affect crystallization. Because the crystal form described here involves different kinase configurations and lacks PKI binding, and presumably competes with the standard crystal form during crystallization, even rather subtle differences between PKAB3 and PKA might lead to the differences in crystallization.
Proposed Binding Mode of BIM2 in PKC-Based on the complex structure of molecule A, and given the similarities among AGC group protein kinases, the key binding mechanisms of BIM2 in PKC can be proposed. An alignment of PKA and PKC isoforms reveals that PKC isoforms ␣ and ␤ show the highest degree of variability of amino acid residues involved in BIM2 interaction (Tables I and IV, in comparison to PKA). Deduced from the BIM2-molecule A complex and an analysis of neighboring amino acids (defined by proximity to within 6.5 Å) five residues need to be considered to predict the binding mechanism of BIM2 in PKC isoforms. The first of these residues is Met 173 , which has many van der Waals interactions with all moieties of BIM2 (Table IV). An equivalent methionine is present in the three classic isoforms PKC␣, PKC␤I/II, and PKC␥; the other PKC isoforms have a leucine, like PKA (Table I). The next two residues are the PKA (and PKB) glutamic acid residues of Glu 127 and Glu 170 , replaced by aspartic acid residues in all PKC isoforms. Glu 127 and Glu 170 form contacts with the pyrrolidine ring, and in silico exchanges to aspartate show that minor adjustments can re-establish a comparable number of van der Waals contacts (Asp 127 , 5; Glu 127 , 6; Asp 170 , 10; Glu 170 , 13) to the pyrrolidine ring.
The final two variable residues in contact with BIM2 (A) are Val 104 and Thr 183 , either likewise conserved in PKC or present as Thr or Ala, respectively (Table I). The exchange V104T occurs for all isoforms except PKC⑀ and PKC. T183A is present in six of the nine isoforms, whereas PKC␥, PKC, and PKC have a threonine as in PKA. In PKAB3, a hydrogen bond exists between BIM2:O33 and Thr 183 (Fig. 6a); this interaction is of course lost in the six PKC isoforms with alanine at this position. The van der Waals contacts to Val 104 can be equivalently established to a threonine at this position, possibly supplemented by a new hydrogen bond. This potential new hydrogen bond could compensate for the loss of the hydrogen bond at position 183 in several PKC isoforms. Only PKC⑀ lacks a hydrogen bonding possibility at both positions.
The combination of the side chains at the three positions 104, 173, and 183 are predicted to be the principal determinants of BIM2 selectivity among PKC isoforms. (Because residues 127 and 170 are conserved across PKC isoforms they cannot be relevant for isoform specificity.) Seven different combinations of residues at positions 104, 173, and 183 occur among the nine human PKC isoforms (PKC␤1 and PKC␤2 do not differ in the kinase core domain). This leads to the prediction of similar specificities for isoform pairs PKC␣ and PKC␤I/II, and PKC␦ and PKC, as corroborated by the reported activity of the related bisindolyl maleimide 1 toward different PKC isoforms. BIM1 is a potent inhibitor of PKC␣ (8.4 nM) and PKC␤1 (18 nM), a medium inhibitor of PKC␦ (210 nM) and PKC⑀ (132 nM) and shows low inhibition toward PKC (5.8 M) (19,44).
Proposed Binding of LY333531 in PKAB3/PKC-LY333531 (Fig. 1B) is a PKC␤I/II isoform-specific bisindolyl maleimide inhibitor (29). It inhibits PKC␤I/II with IC 50 values of 5-6 nM and is 80 -60-fold selective for these isoforms over PKC␣ (29). Other kinases like PKA and SRC are not inhibited by LY333531 (IC 50 Ͼ 100 M). LY333531 is in phase III clinical trials and has been shown to be effective against diabetic retinopathy and diabetic macular edema (Ref. 30 and citations therein). Because of the cyclic linkage of both indoles, LY333531 has fewer conformations available to it compared with BIM2; it lacks, however, the strict planarity of staurosporine. Superposition of LY333531 onto the BIM2 inhibitor of molecule A shows that LY333531 can occupy the same volumes, with, however, a different indole configuration (Fig. 9). The maleimide can remain unchanged with respect to BIM2. The indole arrangements of LY333531 are restricted by the cyclic linkage, however, so that the indole rings are oriented outwards, in contrast to the inwards rotation of indole II of both BIM2 conformations. In this respect, LY333531 is more similar to staurosporine. One consequence of this is that van der Waals contacts to Phe 327 , which is conserved in all PKC isoforms, may become established as in 1STC, and the number of contacts to Tyr 122 will increase. The cyclic linker can be modeled in alternate conformations (Fig. 9), in each conformation the dimethylamine remains near the N30 position of the pyrrolidine of BIM2 and van der Waals contacts to Phe 54 of the glycine loop and Glu/Asp 127 and Glu/Asp 170 can be established.
To evaluate these features for LY333531/PKC inhibition, we modeled the inhibitor into the molecule A structure after in silico exchanges V104T, A123V, P124N, E127D, E170D, and T183A to mimic PKC␤I/II. Furthermore, missing amino acid residues 317-332 were modeled based on the structure of PKA/ staurosporine. The greatest uncertainties in the model lie in contacts with Val 57 and with the glycine loop in general, because of its flexibility. The isoform selectivity profile for the bisindolyl maleimide LY333531 includes a significant preference for PKC␤I/II over PKC␣, which cannot be explained by this model, and no significant differences in amino acid composition exist within 6.5 Å of the inhibitor. The explanation probably involves either a new rotamer conformation to bring a new side chain into inhibitor proximity (as previously observed by PKAB2, the precursor of PKAB3 (38), Table VI) or alternatively involves factors more subtle than the side chain compo-sition of the binding pocket, such as protein backbone shifts. This question can be clarified by crystal structures of different PKC isoforms and LY333531.
Conclusions-Crystallographic studies of enzyme-inhibitor complex structures provide information directly relevant for drug design, particularly with respect to potency and selectivity. For protein kinases, and their typically ATP competitive inhibitors, the mechanisms by which kinases can be selectively inhibited lie mainly in variations among the residues that line the ATP-binding site, or by kinase specific flexibilities. The latter may be related to functionally relevant activity modulation mechanisms, or may be sequence-specific ligand inducible structural modifications.
The structures we describe here show aspects of both phenomena, uniquely showing extreme variability within a single crystal. Using a mutant form of PKA as a surrogate for PKC we succeeded in the co-crystallization of the inhibitor BIM2. The inhibitor binds in two different conformations to two different conformations of the same kinase. Each binding mode shows a range of apparent ligand-induced binding modes. Simultaneously, the different kinase lobe orientations, including the most open form of PKA observed so far, may conversely determine the overall inhibitor binding mode.
Although one of the structures (molecule A) resembles in many aspects the structure of PKA with staurosporine, the glycine loop is distorted in a new way, there are no contacts with the AGC-specific residue Phe 327 , and there are new contacts to the carboxyl group of Glu 170 that could affect the binding of substrate peptides. Even more unusual is the other molecule (B) in the asymmetric unit, with its extreme open conformation and additional rotations of inhibitor N-lobe and displacement of the hinge region peptide. Here, the glycine loop loses its antiparallel ␤-sheet character (45) and as in molecule A no contacts are found to Phe 327 or to the C-terminal lobe of the kinase. The structural similarities of BIM2 and LY333531, an inhibitor against diabetic retinopathy currently in phase III clinical trials, enable structure-based modeling of the binding mode of the drug. As a kind of intermediate between staurosporine and BIM2, LY333531 would be able to bind in the same volume and hinge region interactions as BIM2, but probably would additionally make van der Waals contacts to Phe 327 . Similar to both BIM2 and staurosporine, one methyl group of the free rotatable dimethylamine group is positioned appropriately to establish a contact to Phe 54 on the tip of the glycine loop. Also, it is likely that LY333531 is able to make at least one hydrogen contact to either the Val 104 or Thr 183 homologues in the various PKC isoforms. The combination of a valine with an alanine in PKC⑀ might explain the lowest affinity of LY333531 for PKC⑀ from all conventional and novel PKC isoforms tested (29). Because of the cyclic linkage of the two indole rings, the ring configurations are restricted to relatively symmetric conformations that place the six membered rings of each indole at the periphery of the inhibitor, a configuration not seen for the BIM2 geometries.