Structural basis for p38 MAP kinase quinazolinone and pyridol-pyrimidine inhibitor specificity
The quinazolinone and pyridol-pyrimidine classes of p38 MAP kinase inhibitors have a previously unseen degree of specificity for p38 over other MAP kinases. Comparison of the crystal structures of p38 bound to four different compounds shows that binding of the more specific molecules is characterized by a peptide flip between Met109 and Gly110. Gly110 is a residue specific to the α, β and v isoforms of p38. The ð isoform and the other MAP kinases have bulkier residues in this position. These residues would likely make the peptide flip energetically unfavorable, thus explaining the selectivity of binding. To test this hypothesis, we constructed G110A and G110D mutants of p38 and measured the potency of several compounds against them. The results confirm that the selectivity of quinazolinones and pyridol-pyrimidines results from the presence of a glycine in position 110. This unique mode of binding may be exploited in the design of new p38 inhibitors.
Mitogen-activated protein kinases (MAPKs) are constituents of numer- ous signal transduction pathways that control complex processes such as differentiation, proliferation and cell death. They are also involved in sig- naling pathways that respond rapidly to environmental changes, such as those required for homeostasis and acute hormonal responses1–4. Members of the MAP kinase family share sequence similarity and con- served structural motifs, and are all activated by dual phosphorylation of conserved threonine and tyrosine residues in the activation loop. However, the various MAP kinases are responsive to different extra- cellular stimuli, and each activates a unique, although overlapping, spectrum of cellular targets. Whereas the extracellular signal-regulated kinases (ERKs) are activated in response to hormones and growth factors, the c-Jun N-terminal kinases (JNKs) and p38 are activated by lipopolysaccharide (LPS) and by environmental stresses such as heat shock, hyperosmolarity and radiation5–9.
The MAP kinase p38 has been implicated in both the transcriptional and translational control of cytokine gene expression10, including that of tumor necrosis factor- TNF- and of interleukins IL-1, IL-8 and IL-6. It also has a role in the regulation of COX-2 gene expression11. Among its substrates are transcription factors such as ATF-2, Elk-1, CHOP-1 and MEF2C12–14, kinases such as MAPKAP kinases 2 and 3 and other enzymes such as phospholipase 2 (ref. 15) and NADPH oxidase16. Selective modulation of p38 MAP kinase activity could therefore provide therapeutic intervention for a wide range of inflammatory and auto- immune diseases. The anti-inflammatory properties of pyridinyl imida- zole compounds such as VK19911 and SB203580 (Fig. 1a) were recognized as early as 1972 (ref. 17). The high selectivity of these compounds for p38 has been attributed to the presence of Thr106 in the p38 ATP-binding site. Other MAP kinases, except p38, have in this position either a methio- nine or a glutamine residue, and the larger side chains of these residues pre- vent the binding of the fluorophenyl ring of these inhibitors18–21.
Recently, two new classes of inhibitors (3,4-dihydropyridol[3,2-d] pyrimidines and 2(1H)-quinazolinones) have been reported22–24. These compounds are even more specific for p38 than for JNKs and ERKs, and this specificity could not be explained by modeling of the compounds in the p38 ATP-binding site.
We report here the structures of p38 bound to compound 1 (a pyrimidine imidazole compound), to compound 3 (a quinazoli- none) and to compound 4 (a pyridol-pyrimidine). The structure of p38 bound to compound 2 (quinazolinone), has previously been reported24. Comparison of these structures shows that the binding mode of the more selective inhibitors is characterized by a compound- induced peptide flip between Met109 and Gly110. Gly110 is a residue specific to the , and isoforms of p38; bulkier residues present in other MAP kinases would make the peptide flip much more energeti- cally unfavorable, thus explaining the high selectivity of this class of compounds. To test this hypothesis, we constructed G110A and G110D mutants of p38, and assayed them with a panel of inhibitors. The results support the hypothesis that the increased selectivity is due to glycine in position 110, and suggest that this selectivity determinant could be used in the design of new therapeutic agents.
RESULTS
Structures of p38α–inhibitor complexes
Compounds 1–4 (Fig. 1a) are competitive with ATP and bind to the ATP-binding site of p38 (Figs. 1b and 2)]. Compound 1 (ref. 25) is a tetra-substituted imidazole with a Ki for p38 between 20 pM and 50 pM. Under the conditions of the kinase assay, the IC50 of 0.13 nM for compound 1 approximates the concentration of the enzyme.
Structures of the compounds bound to p38 (Fig. 2a–d) were obtained by soaking the inhibitors into pre-existing apo enzyme crys- tals. Although this technique may sometimes introduce biases (because of crystal lattice contacts or pre-organized structures), soak- ing experiments produce meaningful structures for p38 (refs. 18,20). We have also produced cocrystals of p38 with other inhibitors (data not shown), which show that the p38 structure does not change markedly upon inhibitor binding.
Although these compounds are ATP-competitive inhibitors, they are much more selective for p38 and than for the p38 and iso- forms
(data not shown) and for the other members of the MAP kinase family (Table 1). In addition, they have almost no activity against sev- eral other kinases, including tyrosine kinases22,25,26. Most of the inter- actions of the compounds with p38 are with regions of the protein not directly involved in ATP binding27, and thus not highly conserved among kinases, accounting for the observed selectivity. Compound 1 (Fig. 2a) forms two hydrogen bonds with atoms of the residues that contribute to ATP binding, in the so-called linker region27 (His107–Ala111, red in Fig. 1b): the 2-aminopyrimidine ring interacts with the main chain nitrogen and main chain oxygen atoms of Met109. The conformation of this peptide region in the bound p38 is identical to that in the apo enzyme. In contrast to SB203580 and simi- lar compounds20, there is no interaction between the side chain nitro- gen of the conserved Lys53 and the N3 of the imidazole ring of compound
1. Compounds 2 (ref. 24) and 4 make three interactions with the linker region through the urea moiety (to the main chain oxypounds results from the presence of a threonine in position 106 of p3818–20: the analogous position in other MAP kinases is occupied by residues with bulkier side chains that would not accommodate the large substituents in specific p38 inhibitors.
The N-ethylphenyl ring of compound 1 and the dichlorobenzene of compounds 2 (ref. 24), 3 and 4 lie in the hydrophobic region II27 (magenta, Fig. 1b), within van der Waals distance of the side chain atoms of Val30, Ile108, Asp112 and the main chain atoms of Ala111. This pocket, which is quite wide in inactive p38, could be another determinant of the greater selectivity of compound 1 for p38 than for other kinases. The third substituent of the central moiety (the piperi- dine ring in compound 1) is located in the phosphate-binding area, under the glycine-rich loop (spanning residues Ala30–Val40, orange, Fig. 1b). This group generally forms a stacking interaction with the side chain of Tyr35, which assumes different conformations in the dif- ferent complexes and with respect to the apo p38 structure. The pres- ence of a piperidinyl moiety has been associated with increased aqueous solubility and increased potency22–25. Ordered water mole- cules are located in the inhibitor-binding site in all complexes (Fig. 2). They participate in an extensive hydrogen-bonding network with main chain and side chain atoms of residues that make up the binding site, but do not directly interact with the bound ligand.
Comparison of the structures of p38 bound to compounds 1 and 2 (ref. 24) (Fig. 3a) shows that N6 (as defined in Fig. 1a) in compound 1, a hydrogen-bond donor, is replaced by O1 in compound 2, a hydrorobenzyl group in compounds 2 (ref. 24), 3 and 4 occupy the hydrogen-bond acceptor. This change in polarity is accommodated by a peptide flip at the Met109–Gly110 bond, so that in the complex con- taining compound 2, the carbonyl is interacting with two hydrogen- bond donors (the main chain nitrogens of Met109 and Gly110). Similarly stabilized peptide flips (caused by a switching of the hydrogen-bond donor and acceptor distribution around the peptide plane) have been observed in the structure of Pseudomonas aeruginosa azurin28. In p38, the peptide flip changes the (,) angles of Met109 (i residue) and Gly110 (i + 1 residue) from a (,R) conformation to a (R,L) conformation, without affecting the orientation of adjacent peptide planes or side chains (Fig. 3b). This flip belongs to the group 2 flips, which are characterized by having (i) values that are in the R or L regions when (i + 1) is positive29. As (i + 1) must take on pos- itive values if peptide-plane flipping is to occur, the residue at position i + 1 is often a glycine; the X-Gly motif seems to be quite common among the cases analyzed29. Using semi-empirical molecular orbital calculation, Gunasekaran et al.30 have shown that in the low-energy path for the rotation occurring during the peptide-plane flip, the residue i + 1 crosses = –180,180 boundary. If the residue at posi- tion i + 1 is a glycine, this route would be of a lower energy than that for all other residues. Consequently, although the peptide flip is still possible if the i + 1 residue is not a glycine, the presence of a residue other than glycine would probably reduce the likelihood that the pep- tide flip would occur.
G110A and G110D mutants
To test the hypothesis that the glycine residue at position 110 is a major determinant of the increased selectivity of the quinazolinones and the pyridol-pyrimidines, we constructed G110A and G110D mutants of mouse p38. Mouse and human p38 are 99.4% identical in sequence
(the only two differences are at positions 48 and 263), and their struc- tures are virtually identical: the r.m.s. deviation after aligning the unli- ganded mouse structure (PDB entry 1P38) to the unliganded human structure (PDB entry 1WFC) using CE (http://cl.sdsc.edu/ce.html) is 0.7 Å. Thus, the results obtained with the mutant mouse enzymes should be a valid model for the human enzyme.
The two mutants are catalytically active and have the same kinetic pro- file as the wild-type enzyme (Table 2). IC50 values were determined for the four compounds whose structures are reported (Table 1) and for two other well known p38 inhibitors, SB203580 and VX745. As expected from the crystal structure, all three enzyme forms are inhibited by com- pound 1 or SB203580 with comparable IC50 values, indicating that bind- ing is unaffected by the mutations (no peptide flipping is necessary for binding of this class of compounds). On the other hand, all quinazoli- none and pyridol-pyrimidine compounds show a much lower affinity for the two mutants. In all cases the effect on the G110D mutant is more pronounced than on the G110A mutant, but the differences did not reach statistical significance for most. However, the trend is apparent.
Analysis of the crystal structures shows that the dichlorobenzene ring is bound quite close to the linker region: the distance between the ring potent compound for which a crystal struc- ture has been determined until now. This compound binds in part in a hydrophobic pocket that is structurally distinct from the ATP-binding pocket31, and there is only lim- ited structural overlap between BIRB796 and the pyridinyl-imidazole, quinazolinone and pyridol-pyrimidine inhibitors. Similarly to those of compounds 1–4, the high affinity of BIRB796 results from an interaction with the linker region (His107–Ala111) and from an optimization of interactions with the hydrophobic regions of the binding site31. BIRB796 has a pattern of selectivity compara- ble to those of the pyridinyl-imidazole com- pounds and compound 1. The selectivity has been explained by the fact that these diaryl urea inhibitors also utilize hydrophobic pocket I, which contains Thr106, a residue unique to p38 (ref. 31). Compounds 2–4 show a greater selectivity ratio than do the diaryl urea inhibitors, for reasons discussed below.
DISCUSSION
Comparison with other p38 kinase structures
Several crystal structures of p38 bound to chemically different inhibitors are available in the Protein Data Bank. PDB entries 1A9U, 1BL6, 1BL7, 1BMK20 and 1IAN21 contain pyridinyl imidazole–type compounds. They all have IC50 values between 15 nM and 48 nM20,21 and in general are selective for p38 over ERK and JNKs. The greater potency of our compounds (IC50 values between 0.1 nM and 4 nM; Table 1) as compared with these pyridinyl imidazole compounds prob- ably results from the greater number of hydrophobic interactions that compounds 1–4 make with the enzyme. Compounds 1–4 have bulkier substituents to the core moiety than do SB203580 and VK19911, and can thus optimize the interactions with hydrophobic pockets I and II (Figs. 1a and 2). As indicated by structural analysis18,20,21, the pyridinyl imidazole compounds derive most of their selectivity from their occu- pation of hydrophobic pocket I, where Thr106 is located. The increased selectivity of compounds 2–4 results from the conformational change in the enzyme described above and discussed below.
A structurally different class of inhibitors has also been recently described31 (PDB codes 1KV1 and 1KV2). One of these compounds, BIRB796, has a reported Kd for p38 of 0.1 nM, which makes it the most and the C atom of Gly110 is 4.3 Å, 4.1 Å and 4.7 Å in the complexes with compounds 2 (ref. 24), 3 and 4, respectively. In the case of the G110A or G110D mutants, after the peptide flip, some adjustment of the Met109 and Gly110 residues and the nearby peptide planes must occur to reposition residue 110 in an energetically favorable region of the Ramachandran plot. This would also have some effect on the position of the side chain and of the carbonyl oxygen of residue 110, which would then probably give rise to steric repulsions between the mutant protein and the inhibitor, thus contributing to the observed lower affinity. In addition, because the side chain of aspartate is longer than that of ala- nine, the effect with the G110D mutant would be expected to be more pronounced, and this is indeed reflected in the observed IC50 values.
Structural basis for inhibitor selectivity Although the interactions of all the compounds with the nonconserved regions of the ATP-
binding site are very similar, the selectivity of the quinazolinone and pyridol-pyrimidine inhibitors is unprecedented. This unexpected selectivity results from a peptide flip at Met109–Gly110 that is required by these classes of compounds to optimize interactions with the enzyme. This flip occurs relatively easily in p38 (where there is a glycine in position i + 1). However, the presence of a glutamic or aspartic acid residue (Fig. 3c), as in ERK or JNK, would make the peptide flip more energetically unfavorable, thus markedly lowering the affinity of these compounds for the other MAP kinases.
Differences in inhibitor binding for the mutants are also class dependent (Table 1). Analysis of crystal structures may help explain the differential effects of the mutations among different inhibitor classes. The inhibitor-binding region (Fig. 4) is an almost rectangular area, limited on three sides by the polypeptides residues 30–40 (the glycine-rich loop, blue in Fig. 4), 50–55, 67–78, 82–86, 104–115 (which includes the linker region, red in Fig. 4), 154–158 and 166–172 (the beginning of the activation loop). The fourth side is partially exposed to solvent. The inhibitors occupy almost the entire binding region, and this optimized set of van der Waals interactions is one of the reasons for the inhibitors’ high affinity. However, very little room is left to accommodate changes in inhibitor position that may be required, for example in the G110A and G110D mutants, when the conformation at the linker region is changed. This is especially true for the pyridol-pyrimidine inhibitors: the positions of the urea-contain- ing ring and of the dichlorobenzene ring are very close for compounds 2 and 3; in compound 3, the difluorobenzyl group is linked to the core bicyclic system by a sulfide linker to position 6 (instead of a direct link- age to position 5; Fig. 1). This causes a twist of the molecule in the binding site, so that the dehydropiperidine is pushed toward the glycine-rich loop. In addition, the difluorobenzene is oriented differ- ently in the pocket, and the entire inhibitor is kept in a more rigid con- formation. Modeling studies with the G110A and G110D mutants show that it is possible to achieve a peptide flip and retain the required pattern of hydrogen bonds to the inhibitor, while keeping residue 110 in an acceptable region of the Ramachandran plot (left-handed helix region). In these models, however, the position of the C of residue 110 is shifted 1 Å from the position observed in the crystal structures of the inhibitors reported here. This shift is also accompanied by a repositioning of the amino acid side chain. Residue 110 is much closer to the dichlorobenzene of the inhibitor (the average distance is 3.4 Å, as compared with the 4.1–4.3 Å observed in the X-ray structures), which would suggest that the ligand is partially rearranged to prevent steric clashes with the protein. Given the differences in orientation and in steric constraints between inhibitor classes discussed above, quina- zolinone inhibitors may, in general, accommodate the changes in pro- tein structure better than the pyridol-pyrimidine inhibitors. Larger compounds would also be more conformationally restrained and thus less likely to bind to the mutants. This is confirmed by the observed changes in IC50 values (Table 1).
Figure 4 Binding orientation of 3,4-dihydropyridol[3,2-d]pyrimidine and 2(1H)-quinazolinone to p38. Surface representation of the inhibitor- binding site: the linker region is red, the glycine-rich loop blue. Compounds 2 (green)24 and 4 (magenta) are shown with their molecular surface outlined.
Conclusions
We have confirmed the structural basis for the unusual selectivity of the 3,4-dihydropyridol[3,2-d]pyrimidine and 2(1H)-quinazolinone inhibitors of p38 by constructing mutants that retained wild-type cat- alytic properties but, as predicted, lost affinity for the compounds. The selectivity of these classes of compounds is in part dependent on a small residue (a glycine) belonging to a portion of the enzyme (the linker region) that has never been linked to selectivity. Notably, the subtle change in the protein structure that accounts for the selectivity could not have been predicted by modeling studies alone. This new binding feature, together with previous knowledge of p38 inhibitors,VX-745 may allow for the design of other new p38 inhibitors.