INTRODUCTION

Isoquinolinesulfonamide protein kinase inhibitors of the H series are among the most widely used inhibitors of Ser/Thr kinases and are indispensable in cellular and signal transduction research. Protein phosphorylation (central to cellular regulation) is mediated by the individual action of several hundred different protein kinases. Recent progress in the understanding of the highly complex cellular signaling networks depends largely on the availability and quality of specific agents that interfere with the pathways under investigation. The usefulness of a protein kinase inhibitor is defined by its ability to permeate cell membranes, its solubility, and its relative degree of specificity. Each year hundreds of publications describe work using isoquinolinesulfonamide inhibitors, despite the fact that the mode of specific inhibition by these compounds is not completely understood.

We have used the cAMP-dependent protein kinase (cAPK),¹ which binds many H series inhibitors, as a model system to investigate the factors governing inhibitor binding and specificity. Three of the most frequently used members of this class are H7, H8 (1), and H89 (2) (Fig. 1). They all act in competition to ATP but not to substrate. The highest selectivity and affinity is found with H89, which has a K_i of 48 nM for cAPK, whereas H8 has a moderate affinity for both cGPK and cAPK, and H7 inhibits protein kinase C in addition to cGPK and cAPK (Table I). The inhibitors have enabled assignment of specific roles of these protein kinases in numerous regulatory interrelations (3, 4, 5, 6, 7, 8). They are, however, not only valuable for the cell biologist. Pharmacologists are increasingly interested in the possibility of interfering with cellular signaling, especially in the area of anticancer drug discovery. There is evidence that isoquinolinesulfonyl inhibitors may be promising in this field (9, 10, 11, 12, 13, 14).

Table I. Inhibitory constants of H series protein kinase inhibitors cGPK, cGMP-dependent protein kinase; MLCK, myosin light chain kinase; PKC, protein kinase C; CK-1, protein kinase CK-1, CK-2, protein kinase CK-2. Compound Ki (µM) Ref. cGPK cAPK MLCK PKC CK-1 CK-2 1-(3-Isoquinollnesulfonyl)-2-methylpiperazine (H7) 5.8 3.0 97 6.0 100 780 1 N-[2-(Methylamino)ethyl]-5-isoquinolinesulfonamide (H8) 0.48 1.2 68 15 133 950 1 N-[2-(p-Bromocinnamylamino)ethyl]-5-isoquinolinesulfonamide (H89) 0.48 ± 0.13 0.048 ±0.008 28.3 ± 17.5 31.7 ± 15.9 38.3 ± 6.0 136.7 ± 17.0 2

Recently, crystal structures from several different protein kinases have been solved (for review see Refs. 15, 16, 17, 18, 19). The structures confirmed not only the high degree of structural conservation of the highly homologous catalytic kinase core (20) but simultaneously showed how subtle differences are responsible for individual properties of protein kinases. Most of the residues that are highly conserved or invariant in the protein kinase family line the active site, and many of them interact with ATP (21, 22). Despite high similarity of K_m values for ATP binding among protein kinases, the H inhibitors have remarkably different K_i values. The crystal structures of kinase bound inhibitor molecules described here now show the mode of inhibitory action and the factors governing selectivity and will provide a firm basis for the design of new protein kinase inhibitors.

EXPERIMENTAL PROCEDURES

Expression vector pT7-7 was used to express the bovine C catalytic subunit (23) in strain Escherichia coli BL21(DE3). Cells grown in LB medium (50 mg/liter ampicillin) were induced at 0.6 OD₆₆₀ by addition of 0.4 mM isopropyl--D-galactopyranoside for 16 h at 23 °C. The cells were collected and sonified in 30 mM MES, 50 mM KCl, 1 mM EDTA, 1 mM dithiothreitol. For affinity purification (24), the supernatant of a 45 min, 30,000 × g centrifugation was made 2 mM MgATP, 0.1% Chaps and applied to a column with agarose bound protein kinase inhibitor peptide PKI-(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24). After washing with 50 mM and 200 mM NaCl in 20 mM Tris-HCl, 2 mM MgCl₂, 0.1% Chaps, 400 µM ATP, pH 7.4, about 10 mg of crude catalytic subunit per liter of cell culture eluted in 200 mM arginine, 50 mM NaCl, 1 mM EDTA, 20 mM Tris-HCl, pH 7.4. The recombinant kinase was fully active compared with enzyme from bovine heart. The enzyme was diluted 5-fold in H₂O and applied to a Mono-S 10/10 FPLC column for separation of differently phosphorylated fractions (25) using a LiCl gradient from 0 to 300 mM in 25 mM BisTris-propane, pH 8.5. The identity of the samples was confirmed by electrospray mass spectrometry.

Hanging droplets containing 20 mg/ml protein, 25 mM Mes-BisTris, 75 mM LiCl, 0.1 mM EDTA, 1 mM dithiothreitol, 1.5 mM octanoyl-N-methylglucamide (26), 1.5 mM H inhibitor, and 1 mM PKI-(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24), pH 6.3 to 6.6, were equilibrated at 5 °C against 15% methanol. After 1-3 days crystals were harvested into similar buffer, containing 1 µM inhibitor, 0.1 µM PKI-(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24), and 25% methanol. In each case one crystal was sufficient to obtain a complete data set.

X-ray diffraction were collected with a Siemens X1000 area detector using a copper target Rigaku Rotaflex x-ray generator and graphite crystal K monochromator. Data collection and evaluation were done using the SIEMENS suite of programs, including FRAMBO, ASTRO, and SAINT. All crystals had P2₁2₁2₁ symmetry with cell constants near (73.4, 76.2, 80.5); H89-cAPK·PKI deviated somewhat with cell constants (73.6, 77.2, 80.3). Nominally complete data sets to 2.2 Å (R_sym ~6%) ranged from 63 (H7-cAPK·PKI) to 77% (H8-cAPK·PKI-(peak II)) completeness after applying a 3/f_obs cutoff during refinement. Model refinement was done using XPLOR (27) with kinase coordinates 1cdk (22) after fitting to coordinates 2cpk (29) of the Brookhaven Data Bank (29) as initial model. Model building and refinement concentrated on the inhibitor and binding region; R_free was not used, with a cutoff of 6 Å to minimize bulk solvent distortion. Only those solvent molecules in and around the binding region were modeled. Model building and graphical modeling was done with the program O (31). The final R factors for the models with inhibitor and active site water molecules (18-21 molecules) ranged from 18.8 to 19.9% (Table II).²

Table II. Crystallographic refinement Crystal H7-cAPK · PKI H8-cAPK · PKI(I) H8-cAPK · PKI(II) H89-cAPK · PKI Cell dimensions 73.57, 76.32, 80.56 73.4, 76.2, 80.5 73.38, 76.20, 80.52 73.63, 77.21, 80.27  Final R-factor 18.8% 20.9% 19.9% 19.4% Total no. of unique reflections in refinement (3) 14,158 15,590 17,182 12,157 Resolution range, completeness: 6.0-2.2 63% 70% 77% 62% Highest resolution shell, completeness: 2.20-2.29 (*2.3-2.39) 32% 39% 51% 26%a Bond, angle deviations 0.016 Å, 2.2 ° 0.014 Å, 1.9 ° 0.013 Å, 1.8 ° 0.013 Å, 1.9 ° Rsym 6.0% 6.5% 5.7% 5.8% Total, atoms: water 2973  (18) 2953  (0) 2974  (21) 2983  (21)

RESULTS AND DISCUSSION

Several crystal structures of recombinant mouse or native porcine catalytic subunit of the cAMP-dependent protein kinase (cAPK) have been described, in binary complex with peptide inhibitor PKI-(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24) (28), in ternary complexes with additional MgATP or MnAMP-PNP, and in the unbound apo form (21, 22, 29, 32). The structures showed nucleotide and substrate binding and the mechanism of phosphotransfer. We have determined four crystal structures of the bovine recombinant catalytic subunit of cAPK in complex with the protein kinase inhibitors H7, H8, or H89 and the pseudosubstrate inhibitor peptide PKI-(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24) (H7-cAPK·PKI, H8-cAPK·PKI, H89-cAPK·PKI). The peptide inhibitor will be referred to as ``pseudosubstrate'' or ``PKI'' in the following to avoid confusion with the H series inhibitors. Mass spectrometry of differently phosphorylated enzyme populations confirmed 4 mol (peak I) and 3 mol of phosphate (peak II) per mol of enzyme (25). H8 was co-crystallized with both peak I and peak II enzyme and H7 and H89 with peak II enzyme only. The phosphorylation of Ser-139 in peak I enzyme, in addition to phosphorylation of Thr-197 and Ser-338 in both peaks I and II, agrees with previous measurements (33).

The overall structures of all three H inhibitor complexes of the bovine enzyme conform to the known ternary complex with PKI and ATP or AMP-PNP (AMPPNP-cAPK·PKI); the differing degrees of autophosphorylation in the two H8 co-crystals had no apparent other structural consequences. In the nucleotide bound form, the ATP molecule occupies the cleft between the two lobes that contains the active site, with the adenine base well buried and the -phosphate facing the peptide (pseudo)-phosphorylation site. The inhibition kinetics of H inhibitors with respect to MgATP (1) suggest competition at the ATP-binding site; this is confirmed in all crystal structures as the isoquinoline sulfonyl moieties mimic adenosine binding.

Unbound cAPK can adopt an open conformation, resulting from a rotation of the upper lobe relative to the lower lobe (32) (orientation of Fig. 2). In all H inhibitor co-crystals the enzyme is in an essentially closed conformation (Fig. 2) with two differences: a very small rotation of the N lobe by 2° toward a slightly more open conformation compared with the AMPPNP-cAPK·PKI structure, and a large variability of the conformation of the first two -strands in the region of the glycine flap (Gly-50 to Val-57). This glycine flap has sequence similarity to glycine loop motifs of other nucleotide binding proteins but is structurally unique (34). In the H7-cAPK·PKI and H8-cAPK·PKI co-crystals, it partially occupies at least two positions.

The conformation of the pseudosubstrate PKI-(5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24) in the inhibitor co-crystals differs somewhat from the ternary complexes with nucleotide. Arg-I18, part of the substrate recognition consensus sequence, has a new side chain orientation. In the AMPPNP-cAPK·PKI complex it interacts with the ribose 3OH, Tyr-330, Glu-170, and with the carbonyl of Thr-51 in the glycine flap. In the H inhibitor co-crystal structures, the side chain of Arg-I18 points away from the cleft and forms hydrogen bonds with Glu-170 and Asp-328 (Fig. 3). Partial occupancy of the Arg-I18 side chain has been reported from a binary complex (26). Apparently, the presence of a nucleotide promotes formation of attractive contacts in this region. In contrast to the ternary AMPPNP-cAPK·PKI complex, every contact of the pseudosubstrate peptide to the N-lobe is lost, due to the new conformation of the glycine flap and a 0.9-Å displacement of the PKI pseudophosphorylation site away from the cleft. This is in support of the known synergism of PKI and nucleotide binding by this enzyme (3, 22, 35, 36). The absence of a direct interaction of the PKI with any of the inhibitor molecules excludes an active role of PKI in inhibitor binding. This is supported by identical K_D (for free enzyme) and K_i (in the phosphorylation reaction) values of H8. PKI apparently contributes to crystal formation by providing crystal contacts (not shown).

One of the intriguing aspects of H inhibitor binding to the protein kinase is the spatial congruence of the isoquinoline sulfonyl and sulfonamide groups with a bound adenosine (Fig. 4, A and B). A hydrogen bond between the Val-123 amide and the isoquinoline ring N-13 mimics one of the three H bonds to adenine (22). The isoquinolines provide many, and in the case of H7 and H8 most, of the Van der Waals contacts with enzyme residues (Table III), Val-57 and Ala-70, both almost invariant, Leu-49, Met-120, Glu-121, Tyr-122, Leu-173, Thr-183, and Phe-327. Val-104 is unique in having contacts with ATP in the corresponding complex but not with the inhibitor molecules. Inhibitor binding alters few side chain conformations; Met-120 is shifted 0.8 Å at the  sulfur, accommodating the larger isoquinoline rings, whereas Thr-183, which forms an H bond to the N-7 of adenine, adopts several conformations. In the H8-cAPK·PKI structures, Thr-183 shares a hydrogen bond with the carboxylate of Asp-184, which also is in a new conformation (Figs. 5 and 6).

Table III. Van der Waals contacts of adenosine and isoquinoline ring atoms to enzyme residues p*, polar contact. Numbers include the number of polar contacts in brackets. Residue H7 H8 H89 ATP Leu-49 1 2 2 1 Val-57 2 5 3 6 Ala-70 5 4 4 4 Val-104 1 Met-120 1 1 1 2 Glu-121 1 1 2 1p* Tyr-122 5 3 3 1 Val-123 3  (1p*) 4  (1p*) 6  (1p*) 1p* Leu-173 7 6 5 5 Thr-183 3 3 6 5  (1p*) Phe-327 4 4 3 2 Total 31  (1p*) 32  (1p*) 34  (1p*) 26  (3p*)

The adenosine binding site has been studied with lin-benzoadenine, where a C-6 ring separates the adenine five- and six-membered rings, and 1,N⁶-ethenoadenine nucleotides (37). The large Mg-lin-benzoadenine derivatives bind with affinities similar to MgADP or MgATP. Thus, it is not surprising that the isoquinolines are accepted in the adenine position. However, changes in the N-6 position, as for guanine nucleotides or 1,N⁶-ethenoadenine, appear to affect binding strongly. Bulky substitutions at N-6, as in the purine analogs olomoucine or isopentenyladenine, can cause completely different binding modes compared with adenosine, as observed in the cyclin-dependent protein kinase (40).

In the molecular replacement solution of the structures of the H7-cAPK·PKI and H8-cAPK·PKI complexes, the glycine flap initially was modeled in a position very similar to the ternary AMPPNP-cAPK·PKI complex. Difference density, however, suggested a much more open conformation of the glycine flap. Remodeling indicated partial occupancy for two positions of the glycine flap with relatively high B values (Figs. 6 and 7). Both conformations are interrelated by a 14° rotation of the glycine flap about an axis through the Gly-50 and Val-57 C atoms. This was observed in all crystals except H89-cAPK·PKI. Consequently, the side chain of Lys-78 also occupies two positions, one is compatible with an H bond to the Ser-53 carbonyl in the lower flap position, but would clash with the flap in its upper position. Very weak electron density in both H8-cAPK·PKI and in the H7-cAPK·PKI maps may suggest a third, much lower position of the glycine flap; however, artifacts cannot be excluded. While this low loop was not refined, there are no obvious steric restrictions for such a position of the glycine flap, which could contribute to inhibitor selectivity by enhancing the fit of the binding site. In the H89-cAPK·PKI crystal, in contrast, the glycine flap has clearly defined electron density and normal B values (Fig. 7), indicating dominance of a single upper position of the flap (Fig. 8).

H7 has the lowest affinity and selectivity for cAPK of the three inhibitors (Table I). The structural position of the H7 isoquinoline rings resembles that of H8 and H89. In contrast to H8 and H89, however, the dihedral angle of the H7 sulfonyl group differs by an 86° rotation clockwise around the C-5-SO₂ axis (Figs. 4A and 9). The electron density of the H7 inhibitor without any model bias is shown as an omit map in Fig. 10. The piperazine ring is in a chair conformation with an axial N-H bond at the N-17. The axial methyl group and the equatorial sulfonyl group are in cis, and the asymmetric carbon atom C-22 has the S enantiomer configuration. The N-17 in the piperazine ring shares a hydrogen with the backbone carbonyl of Glu-170 (Fig. 9). Together with the hydrogen bond of the isoquinoline nitrogen to the carbonyl of Val-123, H7 has two polar contacts and a total of 50 van der Waals contacts to enzyme atoms; two H bonds are formed with water molecules. One of them, Wat406, may contribute to H-7 binding by bridging the N-17 to the 1 oxygen of Asn-171.

Stereo image of inhibitor H7 (H7-cAPK·PKI) with electron density maps: the light blue maps show the final 2F_o-F_c map contoured at 1 ; superimposed on the inhibitor is the green difference density F_o-F_c after the first cycle of protein refinement before modeling any inhibitor structure.

The smallest inhibitor, H8, has intermediate affinity for cAPK (Table I); Fig. 4B shows its superposition with ATP. One of the SO₂ oxygens (O-2) of the sulfonamide group superimposes roughly with the O-4 ring oxygen. O-1 is close to the amide of Gly-50, unfavorably oriented for a weak interaction (Fig. 5), which is impossible for H7 because of the differing dihedral angles at the sulfonyl groups. A water molecule (Wat-412) bridges one H8 sulfonyl oxygen to the carboxylate of Asp-184. Only one water molecule (Wat-424) makes contacts with N-4 amide. The N-17 of H8 makes H bond contacts to the backbone carbonyl of Glu-170 and to the carboxylate of Asp-184, an invariant residue and ligand of the high affinity metal ion in the AMPPNP-cAPK·PKI structure; in complex with H8, however, it is in an altered conformation. The N-17 of H8 is near (3.6 Å) the O-1 of invariant Asn-171, the ligand of the second metal ion in the ternary AMPPNP-cAPK·PKI complex. The N-17, like the N-17 of H7, is bridged to the O-1 oxygen of Asn-171 via water Wat-406. H8 shares three H bonds and 46 nonpolar Van der Waals contacts with enzyme atoms; in addition it has four polar and seven nonpolar contacts to water molecules. The larger number of polar contacts to the enzyme is consistent with its higher affinity for H8 than for H7.

In agreement with the structures, Hagiwara et al. (41) showed metal ion-independent binding of 1 mol/mol H8 to cAPK. They analyzed the effect of active site reagents upon binding of H8. Like MgATP, H8 efficiently protects the enzyme against modification by the nucleotide analog p-fluorosulfonylbenzoyl-5-adenosine. In contrast to ATP, but similar to ADP, H8 is rather ineffective against the action of non-nucleotidic reagents such as 5,5-dithiobis-(2-nitrobenzoic acid) and 7-chloro-4-nitro-2,1,3-benzoxadiazole or o-phthalaldehyde, which are known to inactivate the kinase by modifications on Cys-199 or Lys-72 (38). In the structures, H8 is 16 Å from Cys-199 and 8 Å from Lys-72. Bhatnagar et al. (42) used lin-benzoadenosine derivatives to analyze the nucleotide binding site of cAPK. Only lin-benzoadenosine, but not lin-benzo-ADP or lin-benzo-ATP binds to a catalytic subunit modified by p-fluorosulfonylbenzoyl-5-adenosine, 7-chloro-4-nitro-2,1,3-benzoxadiazole, or 5,5-dithiobis-(2-nitrobenzoic acid) (38). On the other hand, only AMP, but not MgADP or MgATP, is able to replace lin-benzoadenosine from the modified kinase, indicating similar binding sites for adenosine, lin-benzoadenosine, and for H8, distinct from the - and -phosphate subsite, corresponding to the crystal structures. Slightly different behavior of lin-benzo-nucleotides has been observed in cGPK (39).

The common parts of the H8 and the H89 molecules have very similar spatial structures and binding modes. Some differences are apparent, however (Fig. 8). The side chain of Thr-183 partially occupies two orientations, one in a position as in the ternary complex and the other as in the H8-PKI·cAPK co-crystal. H89 binding, in contrast to H8, did not cause a conformational change of Asp-184. The N-17 interacts with the carbonyl of Glu-170 as in both other inhibitors and, like N-17 of H8, is close to the side chain of Asn-171 (3.45 Å). A number of contacts are made via bridging water molecules. A central position is held by water Wat-424, which bridges both O-1 and N-4 of the sulfonamide to the side chain of Glu-127 and to the main chain carbonyl of Leu-49. Wat-413 bridges the N-17 to the side chains of Asn-171, Thr-183, and almost to Asp-184 (3.47 Å). The large bromocinnamoyl group extends H89 to roughly the size of an ATP molecule but cannot mimic any of the properties of a triphosphate. While the triphosphate group of ATP is tightly integrated into an extensive network of polar interactions with invariant and catalytic active site residues, including Lys-72, Asp-166, Lys-168, Asn-171, and Asp-184, none of these residues interacts directly with the bromocinnamoyl group of H89. Instead, the H89 side chain points away from the catalytic loop toward the glycine flap, fixing it into its upper position. The large bromine contributes to Van der Waals contacts. H89 has a total of 66 nonpolar Van der Waals contacts and two H bonds to enzyme atoms, in addition to five nonpolar and three polar interactions with water molecules. The high affinity of H89 compared with H8 is explained by the much larger number of Van der Waals contacts to the enzyme.

The binding of the H inhibitors in the adenosine binding site raises the question why nucleotide binding proteins are not inhibited by the isoquinoline sulfonamides in general. Few data are available; there is little indication, however, for significant side effect inhibition of other enzymes. We visually inspected the adenosine binding sites of non-protein kinases in the Brookhaven Protein Data Bank (30). Most proteins seem to have no or incomplete and open binding pockets for the adenosine moiety; these enzymes primarily interact with the phosphoryl group or part of the adenosine derivative, making the binding of isoquinolinesulfonamide inhibitors unlikely. Among these proteins are actin (1atn), DNA-polymerase  (1bpe), fructose-1,6-bisphosphatase (1fbp), or phosphofructokinase (4pfk). Enzymes possessing an adenosine pocket may have little affinity for the isoquinoline derivatives if their binding site is more polar or sterically different from cAPK. Because of the flexible nature of biomolecules, steric effects are the most unreliable to predict. Enzymes with different X angles of their bound ribose may fall in this group as well as enzymes with atoms of the peptide backbone in spatial proximity to the purine, which could interfere with the somewhat larger isoquinoline. Examples are adenosine deaminase (1add), aspartyl-tRNA synthetase (1asz), adenylate kinase (ake), or glutaminyl-tRNA synthetase (1gtr). However, the actual binding mode may be difficult to predict.

The crystal structures of two other kinases in an active conformation have been solved recently, protein kinase CK-1 (CKi-1) from Schizosaccharomyces pombe (43) and phosphorylase kinase -subunit (PhK) from rabbit (44), allowing an evaluation of the potential for H inhibitor binding. The affinity of CK-1 for H7, H8, or H89 is lower than for cAPK (Table I); it is inhibited, however, by another isoquinolinesulfonyl inhibitor CKI-7. The ATP-binding sites of CK-1 and cAPK differ in several residues. The side chain of Met-120, which moves upon inhibitor binding in cAPK, is conservatively replaced by isoleucine in CKi-1. CKi-1 has a one-residue deletion. The residues Val-123-Ala-Gly-Gly are replaced by {Leu-88-Gly-89-Pro90}.³ This might narrow the adenine pocket, bringing {Leu-88} too close to the isoquinoline rings. The conservative replacements of Leu-49 and almost invariant Val-57 by isoleucine may be sterically unfavorable for the sulfonyl groups, while the extra methyl group of {Ile-26} is likely to clash with the aromatic ring in the H89 side chain and account for the low affinity for H89. CK-7 has a chloride in the 5-position and the sulfonamide in the 8-position. Assuming that the ring nitrogen makes a similar contact to the amide of {Leu-88}, CK-7 has to bind differently to CK-1.⁴

The K_i of H7 for PhK is 47 µM, which is in the range of the K_m for ATP in this enzyme (45, 46). The structural similarity between cAPK and PhK is high. Superposition indicates very limited differences in the ATP-binding pocket. {Ile-87} in the position of Val-104 and {Phe-103} as a conservative replacement of Met-120 do not visibly interfere with the isoquinolines. However, the large number of Van der Waals interactions of Tyr-122 with the isoquinolines presumably cannot be achieved by {Leu-105}. This also applies to Phe-327, which has no substitute in PhK. Both residues interact especially with the isoquinolines, as they have only very few contacts to ATP (Table III). PhK has a serine in the position of Gly-55. Sterical conflicts with the bromocinnamoyl moiety of H89 appear possible.

The popularity of the inhibitor H7 results mainly from its ability to interfere with protein kinase C pathways, with an affinity for H7 half that of cAPK. H8 inhibits cAPK more strongly by a factor of 10. Looking for nonconserved residues in the ATP-binding site of PKC from rat (20) compared with cAPK revealed very few differences. Thr-183 is alanine, and Leu-173 is methionine in some isoenzymes of PKC, which might result in a less tight fit of the pocket. Glu-127, responsible for substrate recognition in cAPK and almost in contact to the sulfonylamides, is an aspartate in PKC. Unless the backbone conformation of the enzyme changes, a preference of PKC for H7 is not apparent from these differences. A contribution to selectivity from outside the catalytic core appears possible. Phe-327 interacts with the isoquinoline rings in all three inhibitors; however, because of the rotated sulfonyl group of H7, the Phe-327 ring is in Van der Waals contact to it. Here is a region of potential interaction with PKC-specific residues.

While the triphosphate subsite is highly conserved throughout the protein kinase family, a larger variability exists in the residues lining the adenosine pocket, the site which is used by the isoquinolinesulfonamide inhibitors. The finding that some residues are involved in isoquinoline binding to a larger extent than in purine binding (Table III) may help to explain that the binding constants vary between protein kinases much more for certain classes of inhibitors than for ATP. Residues from outside the conserved catalytic core, such as Phe-327, can also be of significance. In other protein kinases here is a potential site for selective interaction. As binding of the H inhibitors is remarkably similar in all three cases, the side chain of the inhibitors appears to be little responsible for the general mode of binding. This should facilitate the construction of inhibitor variants. Clearly, more crystal structures of other inhibitor complexes are needed to complete our understanding of the binding parameters in the ATP-binding sites of protein kinases.

In contrast to many other enzymes, the protein kinase has a well defined and relatively hydrophobic binding pocket for the purine. During evolution protein kinases have optimized their active site to accommodate many different substrate proteins. This has led to the development of an extremely buried nucleotide binding site and reduced sterical restrictions at the peptide site (34). The drawback of such an optimized peptide binding site is that the buried ADP product can only slowly be exchanged for ATP (22, 47). The relative hydrophobicity of the adenine pocket might reduce the binding forces for the purine to fasten nucleotide exchange.