- Email: [email protected]
Kinase hinge binding scaffolds and their hydrogen bond patterns
Bioorganic & Medicinal Chemistry xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Bioorganic & Medicinal Chemistry journal homepage: www.elsevier.com/locate/bmc
Kinase hinge binding scaffolds and their hydrogen bond patterns Li Xing a,⇑, Jacquelyn Klug-Mcleod b, Brajesh Rai b, Elizabeth A. Lunney c a
Pfizer Worldwide Research and Development, 200 Cambridge Park Drive, Cambridge, MA 02140, United States Pfizer Worldwide Research and Development, 1 Eastern Point Road, Groton, CT 06340, United States c Pfizer Worldwide Research and Development, 10777 Science Center Drive, San Diego, CA 92121, United States b
a r t i c l e
i n f o
Article history: Received 21 May 2015 Revised 24 July 2015 Accepted 8 August 2015 Available online xxxx Keywords: Protein kinase Kinase hinge Hydrogen bonds Hinge scaffolds Potency and selectivity Crystal structure database
a b s t r a c t Protein kinases constitute a major class of intracellular signaling molecules, and describe some of the most prominent drug targets. Kinase inhibitors commonly employ small chemical scaffolds that form hydrogen bonds with the kinase hinge residues connecting the N- and C-terminal lobes of the catalytic domain. In general the satisfied hydrogen bonds are required for potent inhibition, therefore constituting a conserved feature in the majority of inhibitor-kinase interactions. From systematically analyzing the kinase scaffolds extracted from Pfizer crystal structure database (CSDb) we recognize that large number of kinase inhibitors of diverse chemical structures are derived from a relatively small number of common scaffolds. Depending on specific substitution patterns, scaffolds may demonstrate versatile binding capacities to interact with kinase hinge. Afforded by thousands of ligand–protein binary complexes, the hinge hydrogen bond patterns were analyzed with a focus on their three-dimensional configurations. Most of the compounds engage H6 NH for hinge recognition. Dual hydrogen bonds are commonly observed with additional recruitment of H4 CO upstream and/or H6 CO downstream. Triple hydrogen bonds accounts for small number of binary complexes. An unusual hydrogen bond with a non-canonical H5 conformation is observed, requiring a peptide bond flip by a glycine residue at the H6 position. Additional hydrogen bonds to kinase hinge do not necessarily correlate with an increase in potency; conversely they appear to compromise kinase selectivity. Such learnings could enhance the prospect of successful therapy design. Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction The protein kinases encoded in the human genome are involved in regulating essentially all aspects of cellular processes, from metabolism and cell cycle progression to differentiation and apoptosis. Small molecule kinase inhibitors play promising roles in treating a number of diseases including inflammation, neurodegenerative disorders, cardiovascular diseases and cancer.1 Hence they constitute a major component of the pipelines of drug discovery research and development in pharmaceutical industry. At the moment most of the approved kinase inhibitors for clinical use are in oncology, with Tofacitinib being the only marketed therapy for rheumatoid arthritis. Protein kinase enzymes phosphorylate the hydroxyl-containing amino acids (e.g. serine, threonine or tyrosine) under sequencespecific contexts. Despite the large variety of cellular functions Abbreviations: KD, kinase domain; CSDb, crystal structure database; HB, hydrogen bond; RMSD, root mean squared distance; P-loop, phosphate-binding loop; 3D, three-dimensional. ⇑ Corresponding author. Tel.: +1 (617) 665 5369; fax: +1 (617) 665 5575. E-mail address: [email protected] (L. Xing).
associated with protein kinases, their catalytic domains are structurally conserved. The approximately 300 amino acids of the kinase domain are structurally segregated into two distinct lobes. The smaller N-terminal lobe consists of mainly beta sheets and one conserved alpha helix (C-helix). The larger C-terminal lobe is mostly helical and contains the activation segment, which includes residues that in many kinases are phosphorylated upon kinase activation. The hinge region connects the two lobes. ATP binds in the folding cleft located between the two lobes forming conserved hydrogen bond interactions with the hinge. Protein kinases catalyze the transfer of the gamma phosphate group from ATP to the recipient substrate. Targeting the ATP binding site in drug design has led to discovery of numerous inhibitor templates that can bind to members of the kinase family. Protein kinases themselves can be activated upon phosphorylation, often on the activation loop. X-ray crystal structures have elucidated that catalytically active kinases share almost identical conformations, whereas unactivated kinases have diverse structures. Two frequently observed inactive conformations expose additional pockets that are accessible to kinase inhibitors, commonly termed DFG-out and C-helix-out.2,3 Partially occupying
http://dx.doi.org/10.1016/j.bmc.2015.08.006 0968-0896/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
2
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
the active site, the DFG-out conformation directly precludes binding of ATP. The C-helix-out state, on the other hand, can spatially accommodate ATP, but the catalytic machinery is disrupted. With the broken salt bridge between the conserved glutamate and lysine residues, the kinase is rendered inactive, incapable of turning over substrates. Based on surveys of published and Pfizer internal crystallographic data, the structures of over 200 distinct protein kinases have been resolved. Kinases with displaced C-helices or DFG-out conformation of the activation segment each account for approximately 30% of the total structures. Typically compounds containing certain chemical moieties help stabilize the inactive kinase conformation, for example, an urea and/or amide group extending toward the activation loop to facilitate DFG-out, or a hydrophobic aryl in the back pocket to secure the C-helix-out. Some kinases have been shown to naturally rest in the inactive states. For example, the C-helix rotates out in the crystal structure of the apo unphosphorylated BTK-KD.4 In complex with a number of compact ligands well confined in the adenine pocket, the DFG motif and activation loop of cFMS are in the canonical inactive conformation.5 Furthermore, a few kinases have shown unlimited versatility of adopting both DFG-out and C-helix-out capabilities, including SRC, ABL and CDK6.6–13 Kinase structures with simultaneous C-helix-out and DFG-out is possibly a rare event given the redundant mechanism of kinase inactivation, but evidently not mutually exclusive as demonstrated by X-ray structures including EGFR, FAK, BMX and PDK1 kinases.14–17 For ATP competitive inhibitors, hydrogen bond interaction with the kinase hinge is usually indispensable for potent inhibition, regardless of the kinase conformational state that is targeted. In order to make better decisions to select, design and prioritize multiple chemotypes of choice for lead optimization, understanding the interaction patterns of the primary scaffolds that anchor the compounds via hinge hydrogen bonds is of paramount importance. Recently putative hinge-binding fragments were compiled from decomposing known kinase inhibitors mapping to kinase pharmacophore models.18 We have previously reported the extraction of kinase hinge binding scaffolds from large body of X-ray structural data.19 Herein we focus on the statistical analysis of the signature hydrogen bonds in terms of their three-dimensional (3D) properties. With an attempt to understand the relationship between the propensity of the key molecular interactions and their implication for biological activity as well as kinome specificity, the learnings presented here could provide guidance for novel compound design. 2. Methods 2.1. Crystal structure database (CSDb) Kinase protein/ligand crystal complexes were retrieved from the in-house crystal structure database, a large collection of X-ray structure repository that contains both internally solved structures and those selected and imported from the Protein Data Bank.20 A total of 3980 crystal structures of kinases in complex with small, drug-like ligands exemplifying a variety of ligand binding modes were pulled from the database. All of them are better than 3.5 Å in resolution, of which 85% are higher resolution structures of 62.5 Å. These data were then utilized to mine and extract the hinge interacting fragments. An important aspect of this database was that all structures from a given family were aligned onto a common frame of reference based on superposition of preselected residues within the ATP site, making the subsequent analysis of the data derived from this database particularly efficient. The structural diversity of the compounds in the data set was analyzed. When subjected to a nearest-neighbor analysis, the compounds were found to cover different structural series of 670
clusters and 369 singletons, according to Ward’s similarities (distance cutoff of 0.1) using CDK-Daylight fingerprints. The diverse representation of the chemical classes attests the data set is not biased by any specific structural template, and the trends identified are generally applicable. 2.2. Extraction of hinge scaffolds The amino acids on kinase hinges are fully annotated from H1 to H13, with H3 corresponding to the gatekeeping residue.19 A set of criteria was defined to guide hinge extraction as described previously.19 Using the kinase annotation table, the ligand atoms that make hydrogen bonds with the hinge residues are identified. The rest of the molecules were then traversed along the connectivity map to complete the extraction of the core structures, including all ring fragments, the hetero-atoms and functional groups directly bonded to the rings containing cyano, nitro, nitroso, carbonyl, amide, urea, carboxylate, sulfone, sulfonamide, sulfoxide, sulfinic and sulfonic acids.19 The rest of the structural substituents comprising atoms and bonds that were farther away from the hinge were then removed. Hydrogen atoms were added to fill the open valences. The geometric criteria of a hydrogen bond interaction X–H Y–Z are: X Y distance <3.4 Å, and both XHY and HYZ angles >90°. In the end, a total of 595 unique hinge scaffolds were identified from the nearly 4000 kinase inhibitors in CSDb. 2.3. Binding modes of hinge scaffolds Each hinge binding scaffold, defined by a unique structural representation, could orient itself in many different ways in the ATP pocket. Frequently, each altered positioning presents distinct interactions with the kinase hinge. To account for such orientational and positional flexibility, each hinge scaffold is clustered into discrete representative conformations based on their in-place root-mean-squared-distance (RMSD) involving all heavy atoms. If the root-mean-squared-distance (RMSD) is less than or equal to 1 Å the positions are considered degenerative, otherwise a different binding conformation is recorded for the same hinge scaffold. 2.4. Kinase activity and selectivity data Kinase activity data against targets denoted by X-ray experiments were obtained by querying against Pfizer screening database for the corresponding screens, tagged with IC50 as the endpoint type. Multiple IC50 measurements for the same target were retained only if they fell within the same order of magnitude and the values were averaged. The final dataset consisted of 1502 ligands with ranging activity values from 0.01 nM to 100 lM. Kinase selectivity data were extracted from multiple screening panels measured both in-house as well as by commercial vendors, for compounds that appear in CSDb. For each selectivity panel, compounds were tested at either 1 lM and/or 10 lM concentrations at ATP Km for the corresponding kinase. The Gini coefficient at each screening concentration were computed using the percent inhibition data. With a range between zero and one, the Gini coefficient has been commonly used to describe selectivity of kinase inhibitors since its introduction by Graczyk.21 3. Results and discussion 3.1. Hinge binding scaffolds The protein kinases bind a common endogenous ligand. The ATP binding pocket is situated at the folding cleft of the N- and C-lobes, which are connected by the hinge region. Interactions are formed
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
3
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
H N
O
NH2
NH2 N
N
N
N
N
NH2
N N H
N O
N
N
N
H N
HO NH2
Core4 (42) Adenine
Core2 (24) Staurosporine
Core88 (22) 1H-pyrazole-3-amine
Core3 (20) 2-aminopyridine
NH2
H N
N
N
N
N
Core138(18) 1H-indazole
N H
Core64 (18) 7H-pyrrolo[2,3-d]pyrimidin-4-amine
H2N
Core124 (18) 1H-pyrrolo[2,3-b]pyridine
N
N
N
H N
N
NH2
Core13 (17) pyrimidine-4,6-diamine
Core0 (15) pyridine
Core285 (10) 7H-pyrrolo[2,3-d]pyrimidine
Figure 1. Ten most common hinge scaffolds. In parenthesis are the number of different kinases found in binary complexes of CSDb.
with the hinge backbone through two hydrogen bonds: the adenine N1 is an acceptor with the backbone NH (H6) and the 6-amino group is a donor to the backbone CO (H4). Furthermore, the carbonyl group of the same H6 residue is oriented toward the pocket for additional hydrogen bond formation with an exogenous ligand. Effective interaction in the ATP pocket that complements the topological and the electrostatic characteristics of the kinase hinge is essential for potent inhibition of protein kinases.22 The ten most common scaffolds found in the CSDb are illustrated in Figure 1, along with the number of kinases that they complex in. The adenine (Core4) is notably the most nondiscriminative core, crossing over to 42 different kinases. Most of the ensuing compounds are either ATP analogs or its derivatives, due to the fact that adenine has very few valences available for chemistry derivatization. The second most common scaffold is exemplified by staurosporine-type of compounds (Core2). Based on this potent yet non-selective template, many staurosporine analogs have been developed for kinase research. Most of the common scaffolds employ two HB features, specifically a donor– acceptor pair. Depending on the specific binding conformation, they either strictly replicate the hinge contacts of ATP, or alternatively form HBs with H6 NH and CO simultaneously. The amino pyrazole (Core88) scaffold has the potential of forming three hydrogen bonds. In the most productive binding mode, its protons and lone pair engages H4 CO, H6 NH and H6 CO in sequence as exemplified by multiple crystal structures including SLK (PDB code: 2JA0, 2UV2, CAMK1d (PDB code: 2JC6, 2VN9), Jak2 (PDB code: 2XA4, 3Q32), ITK (PDB code: 3MJ1), etc. On the other hand, the pyridine (Core0) can only form one HB by accepting a proton from H6 NH. More discussion on conformational variation in ATP pocket for a given scaffold follows in next section. It is noted from the analysis that the majority of the scaffolds (72%) are in unique complex with one specific kinase. They could be either under-explored by medicinal chemistry and/or lack of crystallographic execution. Additionally they may present valuable opportunities for improving kinome selectivity and/or generating chemical novelty in the exceedingly crowded kinase field.19 3.2. Varying binding modes of hinge scaffolds A single hinge scaffold could present itself in many different orientations in the ATP binding site. As depicted in Figure 2, in its canonical conformation of binding kinase, the adenine of ATP donates a proton to the backbone carbonyl of H4, and accepts a proton from the backbone NH of H6 via N1 of the purine. In
H4 N
H5
N
O
O N
H6 N
O
Figure 2. A single hinge scaffold could present multiple binding orientations as exemplified by adenine in CSDb. The canonical binding conformation corresponding to ATP binding is colored in green in ball-and-stick model.
addition, alternative orientations are identified from CSDb mining, of which two representations are illustrated in Figure 2. In the magenta orientation, the imidazole of the adenine donates its proton to H4, and the N3 of the purine accepts the NH proton from H6. Alternatively, the third mode colored in cyan is the result of a 120° counter-clock-wise rotation coupled with a translational move toward downstream residues of the hinge from the canonical state. In this mode, the adenine forms hydrogen bond with H6 but no longer with H4 residue. Specifically, the imidazole nitrogen accepts the proton from H6 backbone NH, and the amino moiety donates its proton to the backbone carbonyl of H6. In addition to these three representations, other variations are introduced by overall shifts along the hinge relative to the gatekeeper residue, movements toward or away from the hinge, as well as translation in the vertical direction relative to the P-loop. Out of the nearly 600 hinge scaffolds we identified in total 812 differential binding modes. For the top ten most common scaffolds, the number of binding modes is plotted in Figure 3, along with the percentage on the minor y-axis of the total number of binary complexes for the respective scaffolds. All of them exhibit some capacity of repositioning and/or re-orienting in the ATP site, albeit at different degrees of versatility. Adenine presents the most variations in the binding site, owning to its abundant composition
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
4
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
NH2 N
N
N
N
N
NH2 N
NH2
N
N N H
Core285
Core0
Core13
Core64
Core138
Core124
Core3
Core88
Core2
Core4
% Of Binding Modes
Number Of Binding Modes
N
Figure 3. Number of different binding modes observed for the top ten most common kinase scaffolds as plotted on the major y-axis. The minor y-axis depicts the different modes as a percentage of the total number of kinase complexes observed for each scaffold.
of heteroatoms affording multiple alternatives of hydrogen bond donating and accepting patterns. The large number of exemplary structures in CSDb also enriches its observed rotational as well as translational versatility. In resemblance, the aminopyrimidine and aminopyrrolopyrimidine are amongst the ones that afford varying modes of interactions with kinases. Different substitution patterns off the available valences would define different binding modes in order to present the overall molecule in its highest 3D complementarity in the protein pocket. In comparison, the binding mode of the staurosporine core is much more conserved. Except for the translational fluctuations occasionally observed in the binding site, no rotational motion was found. Similarly for the pyridine core various translational movements were observed given its small size, but no orientational changes due to its limited HB potential. 3.3. Hinge hydrogen bonds Hydrogen bond is among the most important specific interactions in biological recognition processes. In particular for kinases, hydrogen bonds with the hinge backbone residues are considered the anchor of ligand binding, a generally indispensable interaction for potent enzyme inhibition. We examine the intermolecular hydrogen bond patterns in detail in the inhibitor-kinase complexes. As illustrated in Figure 4a, the majority of the hydrogen bonds were made with the NH of H6, which accounts for twothirds of the crystal structures. In such context it is worth noting that the PIM kinases have a proline at position H6 that is conserved for the family (PIM1, 2, and 3), and thus are devoid of H6 NH donor. In CSDb a total of 12 PIM structures were available, of which the ligands make primary contacts with H4 CO for hydrogen bond interaction. In total, approximately one third of the ligands donate a proton to H4 CO, and slightly less than one-third donate to H6 CO. The predominant interactions with H6 NH is about twice the number of interactions with H4 CO. A former report based on PDB survey drew similar results with regard to hinge hydrogen bonds.23 It appears to be consistent with the theoretical projection
that unsatisfying a hydrogen bond donor in a buried protein environment is more energetically costly than burying an acceptor.24 The high propensity of the H6 NH pairing could also be facilitated by its spacial accessibility to the ligand molecule, situated between H4 CO and H6 CO. In contrast, the H4 is sterically more hindered given its close proximity to the gatekeeping residue H3, the side chain of which, depending on its character, typically forms important contacts with the kinase inhibitor. Large body of research has established that complementary interactions with the gatekeeper is critical to drive potency as well as selectivity, by targeting differences in amino acids at the gatekeeping position of related kinases.25 It is conceivable that intermolecular contact with the gatekeeper side chain would sterically block hydrogen bond engagement with H4 NH. As a matter of fact no HB is observed with H4 NH in our current analysis. Indicated by Figure 4a, some scaffolds extend hydrogen bond interactions toward downstream H7, by either accepting from H7 backbone amine, or less commonly donating to its backbone carbonyl. Hydrogen bond with H5 is extremely rare. The reason lies in the ensuing conformation of the peptide bonds along the kinase hinge. In the canonical kinase conformation denoted in Figure 2, both the H5 NH and CO point away from the binding pocket, geometrically precluding them from partnering with the ligand. However, there are exceptions. Through mining of thousands of crystal structures, we did observe the unusual hydrogen bond with H5 CO for a small number of structures, and it is made possible by a peptide bond flip connecting H5 and H6. Detailed discussion will be provided in the following section. The propensity of multiple hydrogen bond formation is of interest in comparison with the single point hinge contact. The pairwise hydrogen bond patterns with any two of the three hinge elements (H4 CO, H6 NH and H6 CO) are depicted in Figure 4b, along with the propensity of satisfying all three hydrogen bonds. Amongst them the dual hydrogen bonds with H4 CO and H6 NH is the most prominent. Almost a quarter of the CSDb structures present such hinge interactions, strictly replicating the hydrogen bond pattern
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
5
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
867 2623
577 413 1330 1042
(b)
H4_O & H6_N & H6_O
(a)
H6_N & H6_O
H7_N
H6_O
H6_N
H5_O
H5_N
H4_O
H4_N
H7_O
27
H4_O & H6_O
185 7
H4_O & H6_N
8
Figure 4. Hinge hydrogen bond distributions in CSDb. (a) Number of hydrogen bond occurrences for each bearing backbone of hinge residues. (b) Number of structures engaging multiple hydrogen bonds with kinase hinge.
of ATP. In comparison, the dual hydrogen bonds with H6 NH and H6 CO are less frequent, accounting for 15% of the total hinge interactions. Very rarely the dual hydrogen bond pattern with H4 CO and H6 CO is observed. In general it is challenging to install two donors in the conforming geometry to satisfy such 3D requirement. Furthermore, in the absence of water-mediated interaction, leaving H6 NH unsatisfied is thermodynamically disfavored, burying a hydrogen bond donor in an enclosed protein environment. Once the two donors are approximately placed in a scaffold, an accepting atom in-between could naturally add a third HB with H6 NH, depotting the scaffold into the class of three-point contacts. As depicted in Figure 4b, a number of structures utilize all three HB points from H4 CO, H6 NH and H6 CO. The frequency of three-point contacts is appreciably notable, accounting for more than 10% of the total kinase structures in CSDb. Hydrogen bond distances, a reflection of the strength of the interaction, are analyzed in Figure 5 by box plots. Both the longer distances and the broader distributions of the observed distances could be indicators of weaker HB interactions. The median
Hydrogen Bond Distance (Å)
3.4
3.2
3.0
2.8
2.6
2.4
H4_O
H6_N
distances for the backbone CO’s are almost identical: 2.88 Å for H4 CO and 2.85 Å for H6 CO. This suggests similar HB interactions with the two carbonyls of the kinase hinge, with H6 CO displaying a slightly broader distribution. On the other hand, the median distance with H6 NH (2.95 Å) is noticeably longer than the backbone carbonyls. In agreement, a previous survey of PDB reported a median distance of 2.9 Å for NH donors, approximately 0.15 Å larger than the OH donors.26 Possibly due to the larger van de Waals radius of nitrogen than oxygen atom, the hydrogen bonds of NH donors and N acceptors could exhibit slightly longer distance in distribution. We applied the general criteria of angular definition of hydrogen bonds in the detection process, but did not analyze the detailed angle distributions. The angular preferences of hydrogen bonds are important, but could be less profound in protein structures. Within boundary, the protein structure could relax to adapt to a specific ligand, in many more ways than the small molecule crystal structures. Additionally, when macromolecular X-ray structures are fitted to electron densities the small molecule conformation often play subordinate role. Given that the overall resolution of protein structure is much lower than that of small molecules, the angular terms could bear much imprecision and was used as a filtering criterion. Weak interactions involving CF H, C(Cl/Br) O@C, CH O and NH p have been observed to contribute to ligand–protein binding in a subtly directional manner. Sometimes they are termed pseudohydrogen bonds. Calculations suggested that CH O@C interactions are about one-half the strength of an NH O@C hydrogen bond.27 Similarly, halogen bonds are significantly weaker than hydrogen bonds. Ab initio calculations on formaldehydehalobenzene dimers estimated gas phase interaction energies below 2.5 kcal/mol, corresponding to the strength of CH O pseudohydrogen bonds.28 These categorically weaker interactions are second tier to hydrogen bonds. Intermolecular interactions involving pseudohydrogen bonds and/or halogen bonds await further analysis in a future communication.
H6_O 3.4. Unusual hydrogen bond with H5
Figure 5. Distribution of hydrogen bond distance between ligand and kinase hinge. Average of 2.9 Å is similar to survey from small molecule X-ray with broader distributions. Box regions correspond to 50% of the distribution, lines extend to max 1.5 times of this interval, and dots are outliers.
As evidenced by previous analysis, the most common HBs are formed with H4 CO, H6 NH and H6 CO, with varying numbers of
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
6
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
kinase inhibitors satisfying one, two, or three of the hydrogen bond donating and accepting partnerships. This pattern is largely attributed to the typical backbone orientations along the kinase hinge, which directs CO and NH groups in alternation between pointing toward and averting away from the ATP pocket in a sequential manner (as depicted in Fig. 2). As a result, whereas CO of H4, NH and CO of H6 are naturally oriented toward the ligand binding pocket, the NH and CO of H5 in their inherent orientations point away from the active site. Consequently, intermolecular hydrogen bonds are generally not observed with H5 residue. It is intriguing that an unusual hydrogen bond with the H5 CO emerged from our analysis (Fig. 6). This interaction is embodied by the PDB structure (code: 3HMP) for the dual-specificity protein kinase Mps1 in complex with Cpd 4, a low affinity compound (IC50: 38 lM),29 the diaminopyridine-based compounds,30,31 as well as several in-house crystal structures. Comparing the holo structure with the apo Mps1 protein we found an interesting conformational switch along the kinase hinge. In the apo Mps1 structure (PDB code: 3DBQ) the CO of H5 points in the opposite direction of ligand binding, as is generally observed with the canonical hinge conformations. Upon binding the ligand, an HB donor is presented by the 4-amino group attached to the quinazoline moiety of Cpd 4, which effectively shifts the thermodynamic equilibrium of protein conformations. As a result, the peptide bond linking H5 and H6 undergoes an almost 180° rotation to satisfy a non-conventional hinge hydrogen bond between the H5 CO and the amine of Cpd 4. Toward confirming this is not a crystallographic artifact, we identified five additional in-house X-ray structures of Mps1 with compounds of similar HB donating capacities. The different inhibitor-kinase complexes exhibit the consistent HB interaction with H5 CO, afforded by the backbone flip of the peptide bond. On the other hand, in the other Mps1 binary complexes (e.g. PDB code: 4JS8, 4JT3 etc.) where the ligands do not employ a proper donor in the vicinity of H5 CO, the observed backbone conformations are consistent with the apo structure. From mining the crystal structure database, the reversed conformation of H5–H6 peptide bond is only observed for Mps1 kinase. We think this is an uncommon event where certain sequences along the kinase hinge may be required. Specifically the glycine reside immediately downstream of H5, Gly605 (H6)
in Mps1, could be instrumental in providing the necessary flexibility to accommodate the backbone flip. Devoid of a side chain, glycine residue is unique in its superfluous flexibility that is otherwise restricted in other amino acids. In analogy, the H6–H7 peptide bond flip has been reported for p38 kinase, in response to binding of a pyridinone chemotype where the second hydrogen bond is engaged with the re-arranged hinge conformation.25 The H7 residue, Gly110 of p38, was found critical to allow the flipped conformation. We mined the human kinase genome for propensity of glycine at the H6 position, and found it is conserved in five additional kinases: SNRK and TSSK2 from CAMK family, COT from STE family, MOS and TOPK from others. Given the rare occurrence of a glycine at the H6 position, the probability of the H5–H6 amide flip is predicted to be extremely low. Crystal structures of the aforementioned kinases are in need to elucidate additional conformational flexibilities. 3.5. More hinge hydrogen bonds do not yield higher potency In a buried environment it is thermodynamically favorable for the NH and CO groups to form hydrogen bonds, and the number of observed hydrogen bonds increases with decreasing solvent accessibility.32,33 Since the hinge section of the ATP pocket is less solvent accessible, additional HB contacts could potentially contribute to binding affinity. In Figure 7 the inhibitory pIC50 values for the ligands bound to their respective kinases were presented in box plots. The x-axis denotes the hydrogen bond partners from the hinge. As discussed previously, very few examples present dual hydrogen bonds with H4 CO and H6 CO, and only two of them have activity data, thus they were taken out of the statistical analysis. The number of data points for H4 CO and H6 CO single HB patterns were also small, 22 and 20 respectively, thus they were excluded due to insufficient sample size. The focus was on the singular HB with H6 NH containing 432 data points, double HB patterns of H4_CO/H6_NH (491 data points) and H6_NH/H6_CO (265 data points), and triple hydrogen bonds with H4 CO, H6 NH and H6 CO (208 data points).
10 9
pIC 50
8 7 6 Cys604 (H5)
5
N N H
N
Cpd 4
Figure 6. Unusual hydrogen bond with H5 carbonyl of Mps1/TTK, afforded by protein backbone flip of H5–H6 upon ligand binding. Apo Mps1 kinase is in yellow (PDB code: 3DBQ), and the binary complex with Cpd 4 (in gold) is colored in cyan (PDB code: 3HMP).
Two hydrogen bonds
H4_O & H6_N & H6_O
Cl
H6_N & H6_O
Gly605 (H6)
H4_O & H6_N
H6_N
4
Figure 7. More hydrogen bonds doesn’t translate to higher potency. Box regions correspond to 50% of the distribution, lines extend to max 1.5 times of this interval, and dots are outliers.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
7
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
The distributions show no significant trend of activity values as a function of the number of HBs. The median IC50’s are within three fold for compounds making one, two or three hydrogen bonds with the kinase hinge, which is within experimental uncertainty. In agreement, the analysis on nine approved kinase drugs revealed no correlation between the number of hinge hydrogen bonds and ligand affinity.23 This suggests the contribution of hydrogen bond to binding affinity is highly context-dependent. Usually, leaving a buried donor or acceptor unsatisfied results in a free energy penalty. On the other hand, desolvation of the donor and acceptor must occur for the hydrogen bond to form, such that the effects of hydration and hydrogen bond formation nearly cancel out. Toward that end, it has been shown that removing a hydrogen bond by an NH through a methyl replacement hardly reduced the binding affinity of peptidomimetic thermolysin inhibitors.34 In the context of kinase hinge interaction, whereas one or more of the neighboring CO and/or NH may not be satisfied, they are in general shielded by a non-specific interaction with a heavy and/or hydrogen atom. Beyond hinge, complementarity of attractive and repulsive interactions in other parts of the ATP pocket is also essential in defining the ultimate potency of a kinase inhibitor. We also examine the relationship between potency and HB distance, which represents a geometric measure of the strength of the interaction. The hydrogen bonds with H6 NH were analyzed since it has the highest propensity, totaling 1396 data points. As shown in Figure S1, most of the hydrogen bond distances are within the optimal range of 2.8–3.2 Å (1093, 78%). The potency distributions are essentially comparable, except for the HB distances that fall remarkably out of the optimal range, for example, <2.6 Å and/or >3.4 Å, which suggests interactions in a forced environment could be detrimental to binding affinity. 3.6. More hydrogen bonds could compromise selectivity Since hinge hydrogen bonds are provided by backbone atoms in the segment of kinase structures that are largely conserved by conformation, they are generally regarded as potency anchors.
It is of interest to investigate the implication of the hydrogen bond patterns in defining kinome specificity. Using kinome screening panel results at 1 lM and 10 lM, the distributions of Gini coefficients as a selectivity measure were plotted against the number of hydrogen bonds (Fig. 8). The sample sizes are sufficient to render statistical significance: there are in total 811 Gini coefficients computed from 1 lM data and 782 from 10 lM, and each box distribution contains between 123 and 272 data points. Not surprisingly, the 1 lM percent inhibition data consistently suggests higher specificity than the 10 lM measurements. The median Gini coefficients range from 0.47 to 0.57 for 1 lM, and 0.32 to 0.53 for 10 lM respectively. The overall distribution profiles are similar for the two measured concentrations. Distribution of Gini coefficients is also examined with respect to hydrogen bond distance of H6 NH for 1 lM and 10 lM concentrations separately (Fig. S2). For the full range between 2.6 Å and 3.4 Å, the median Gini coefficients fluctuate between 0.3 and 0.6. It appears that compounds of one hydrogen bond have higher specificity, and those with three hydrogen bonds are among the ones of lowest Gini coefficients. This supports the notion that enhancing interactions with the conserved regions of protein structures could compromise specificity in general. At a minimum, kinome selectivity does not seem to benefit from additional hydrogen bonds with the kinase hinge. Not only the scaffolds alone, but also substitution patterns beyond the core region have large impact on the selectivity profiles. The practice of tailoring design to complement specific interactions employing other structural elements of the ATP site, for example, the P-loop, the activation loop, and/or the C-helix, has been successfully employed by medicinal chemists. Built upon even the most common hinge element, prominent selectivity can be achieved for the kinase target of interest by exploiting subtle differences in amino acids that compose the ATP binding site. Further investigation reveals that due to the stringent 3D configurations of the donor–acceptor-donor pattern, many three-point contacts display one or more poor hydrogen bonds, for example, heavy-to-heavy atom distance of less than 2.8 Å and/or greater
1µM
Two hydrogen bonds
Three hydrogen bonds
H6_N & H6_O
H4_O & H6_N
H6_N
Gini Coefficient
10µM
Figure 8. Gini coefficients computed from 1 lM (blue) and 10 lM (yellow) kinase panel screen. More hydrogen bonds could compromise selectivity. Box regions corresponds to 50% of the distribution, lines extend to max 1.5 times of this interval, and dots are outliers.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
8
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
than 3.2 Å (330 out of 410, 80%). Thus the additional hydrogen bonds to the kinase hinge may not add to binding free energy as favorably as expected due to their sub-optimal geometry. This could impact both potency toward the primary target as well as selectivity against the off-targets. In medicinal chemistry design, we are usually faced with choices of hinge templates that are of one-, two-, or three-point contacts. While it is tempting to regard the scaffolds of more hinge contacts as superior, our data suggest the decision should be made in a case-specific manner. There is no general advantage for either potency or selectivity to fully satisfy all three hinge hydrogen bonds. Toward that end, the simple pyridine has emerged from our analysis as one of the most common scaffolds. The popularity of this single-point scaffold speaks to its utility and value, from which a number of late stage kinase inhibitors have been developed for oncology indication.35–37 4. Conclusion Kinase inhibitors can exert their modulation of target function by elaborate mechanisms. Type I inhibitors form H-bonds with the kinase hinge and occupy the adenine binding pocket. Type II inhibitors occupy the adenosine pocket, but also capture conformational changes that is typically either DFG-out or C-helix out, or both. The scope of the current investigation covers both type I and type II kinase inhibitors. Nearly 600 different hinge scaffolds were extracted from CSDb composed of Pfizer internal and public crystal structures of kinase-inhibitor complexes. Coordinates of crystal structures made possible the analysis of 3D properties characteristic of protein–ligand binding, in particular the hinge hydrogen bonds that are generally critical for potent kinase inhibition. The majority of hydrogen bonds are formed with H6 NH, a spatially accessible feature located in the middle of the kinase hinge. Dual hydrogen bonds are commonly observed with additional recruitment of H4 CO upstream and/or H6 CO downstream. Triple hydrogen bonds accounts for 10% of the total binary complexes, engaging all three prevailing features via the donor–acceptor-donor pattern of the inhibitor. Combining with the biological assay data, we demonstrate that more hydrogen bonds in the hinge region do not necessarily afford higher potency. Conversely, they may have deleterious effects on kinome selectivity. A single scaffold can adopt different orientations to interact with kinase hinge, a versatility that would be constrained by explicit substitution patterns beyond the hinge binding scaffold. Furthermore, an unusual hydrogen bond with a canonically unavailable conformation is observed, requiring a peptide bond flip conceivably made possible by a glycine residue. The compiled collection of hinge scaffolds along with their binding coordinates could serve as basis set for hinge hopping, a practice frequently employed to generate novel invention as well as to optimize existing leads in medicinal chemistry. Acknowledgement This study was sponsored by Pfizer Inc. Supplementary data Supplementary data (plot of activity (pIC50) versus hydrogen bond distance, plot of Gini coefficient versus hydrogen bond distance computed using 1 lM and 10 lM percent inhibition data, kinase scaffolds along with their 3D coordinates and activity data) associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.bmc.2015.08.006.
References and notes 1. Cohen, P. Nat. Rev. Drug Disc. 2002, 1, 309. 2. Jacobs, M. D.; Caron, P. R.; Hare, B. J. Proteins 2008, 70, 1451. 3. Brooijmans, N.; Chang, Y. W.; Mobilio, D.; Denny, R. A.; Humblet, C. Protein Sci.: Publ. Protein Soc. 2010, 19, 763. 4. Mao, C.; Zhou, M.; Uckun, F. M. J. Biol. Chem. 2001, 276, 41435. 5. Schubert, C.; Schalk-Hihi, C.; Struble, G. T.; Ma, H. C.; Petrounia, I. P.; Brandt, B.; Deckman, I. C.; Patch, R. J.; Player, M. R.; Spurlino, J. C.; Springer, B. A. J. Biol. Chem. 2007, 282, 4094. 6. Dar, A. C.; Lopez, M. S.; Shokat, K. M. Chem. Biol. 2008, 15, 1015. 7. Georghiou, G.; Kleiner, R. E.; Pulkoski-Gross, M.; Liu, D. R.; Seeliger, M. A. Nat. Chem. Biol. 2012, 8, 366. 8. Seeliger, M. A.; Nagar, B.; Frank, F.; Cao, X.; Henderson, M. N.; Kuriyan, J. Structure 2007, 15, 299. 9. Hennequin, L. F.; Allen, J.; Breed, J.; Curwen, J.; Fennell, M.; Green, T. P.; Lambert-van der Brempt, C.; Morgentin, R.; Norman, R. A.; Olivier, A.; Otterbein, L.; Ple, P. A.; Warin, N.; Costello, G. J. Med. Chem. 2006, 49, 6465. 10. Nagar, B.; Hantschel, O.; Seeliger, M.; Davies, J. M.; Weis, W. I.; Superti-Furga, G.; Kuriyan, J. Mol. Cell 2006, 21, 787. 11. Levinson, N. M.; Kuchment, O.; Shen, K.; Young, M. A.; Koldobskiy, M.; Karplus, M.; Cole, P. A.; Kuriyan, J. PLoS Biol. 2006, 4, e144. 12. Russo, A. A.; Tong, L.; Lee, J. O.; Jeffrey, P. D.; Pavletich, N. P. Nature 1998, 395, 237. 13. Cho, Y. S.; Borland, M.; Brain, C.; Chen, C. H.; Cheng, H.; Chopra, R.; Chung, K.; Groarke, J.; He, G.; Hou, Y.; Kim, S.; Kovats, S.; Lu, Y.; O’Reilly, M.; Shen, J.; Smith, T.; Trakshel, G.; Vogtle, M.; Xu, M.; Xu, M.; Sung, M. J. J. Med. Chem. 2010, 53, 7938. 14. Muckelbauer, J.; Sack, J. S.; Ahmed, N.; Burke, J.; Chang, C. Y.; Gao, M.; Tino, J.; Xie, D.; Tebben, A. J. Chem. Biol. Drug Des. 2011, 78, 739. 15. Michalczyk, A.; Kluter, S.; Rode, H. B.; Simard, J. R.; Grutter, C.; Rabiller, M.; Rauh, D. Bioorg. Med. Chem. 2008, 16, 3482. 16. Iwatani, M.; Iwata, H.; Okabe, A.; Skene, R. J.; Tomita, N.; Hayashi, Y.; Aramaki, Y.; Hosfield, D. J.; Hori, A.; Baba, A.; Miki, H. Eur. J. Med. Chem. 2012. 17. Nagashima, K.; Shumway, S. D.; Sathyanarayanan, S.; Chen, A. H.; Dolinski, B.; Xu, Y.; Keilhack, H.; Nguyen, T.; Wiznerowicz, M.; Li, L.; Lutterbach, B. A.; Chi, A.; Paweletz, C.; Allison, T.; Yan, Y.; Munshi, S. K.; Klippel, A.; Kraus, M.; Bobkova, E. V.; Deshmukh, S.; Xu, Z.; Mueller, U.; Szewczak, A. A.; Pan, B. S.; Richon, V.; Pollock, R.; Blume-Jensen, P.; Northrup, A.; Andersen, J. N. J. Biol. Chem. 2011, 286, 6433. 18. Zhao, H.; Caflisch, A. Bioorg. Med. Chem. Lett. 2015, 25, 2372. 19. Xing, L.; Rai, B.; Lunney, E. A. J. Comput. Aided Mol. Des. 2014, 28, 13. 20. Berman, H. M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T. N.; Weissig, H.; Shindyalov, I. N.; Bourne, P. E. Nucleic Acids Res. 2000, 28, 235. 21. Graczyk, P. P. J. Med. Chem. 2007, 50, 5773. 22. Noble, M. E.; Endicott, J. A.; Johnson, L. N. Science 2004, 303, 1800. 23. Ghose, A. K.; Herbertz, T.; Pippin, D. A.; Salvino, J. M.; Mallamo, J. P. J. Med. Chem. 2008, 51, 5149. 24. Barratt, E.; Bronowska, A.; Vondrasek, J.; Cerny, J.; Bingham, R.; Phillips, S.; Homans, S. W. J. Mol. Biol. 2006, 362, 994. 25. Xing, L.; Shieh, H. S.; Selness, S. R.; Devraj, R. V.; Walker, J. K.; Devadas, B.; Hope, H. R.; Compton, R. P.; Schindler, J. F.; Hirsch, J. L.; Benson, A. G.; Kurumbail, R. G.; Stegeman, R. A.; Williams, J. M.; Broadus, R. M.; Walden, Z.; Monahan, J. B. Biochemistry 2009, 48, 6402. 26. Bissantz, C.; Kuhn, B.; Stahl, M. J. Med. Chem. 2010, 53, 5061. 27. Vargas, R.; Garza, J.; Dixon, D. A.; Hay, B. P. J. Am. Chem. Soc. 2000, 122, 4750. 28. Riley, K. E.; Merz, K. M., Jr. J. Phys. Chem. A 2007, 111, 1688. 29. Chu, M. L.; Lang, Z.; Chavas, L. M.; Neres, J.; Fedorova, O. S.; Tabernero, L.; Cherry, M.; Williams, D. H.; Douglas, K. T.; Eyers, P. A. Biochemistry 2010, 49, 1689. 30. Kusakabe, K.; Ide, N.; Daigo, Y.; Itoh, T.; Higashino, K.; Okano, Y.; Tadano, G.; Tachibana, Y.; Sato, Y.; Inoue, M.; Wada, T.; Iguchi, M.; Kanazawa, T.; Ishioka, Y.; Dohi, K.; Tagashira, S.; Kido, Y.; Sakamoto, S.; Yasuo, K.; Maeda, M.; Yamamoto, T.; Higaki, M.; Endoh, T.; Ueda, K.; Shiota, T.; Murai, H.; Nakamura, Y. ACS Med. Chem. Lett. 2012, 3, 560. 31. Kusakabe, K.; Ide, N.; Daigo, Y.; Itoh, T.; Yamamoto, T.; Kojima, E.; Mitsuoka, Y.; Tadano, G.; Tagashira, S.; Higashino, K.; Okano, Y.; Sato, Y.; Inoue, M.; Iguchi, M.; Kanazawa, T.; Ishioka, Y.; Dohi, K.; Kido, Y.; Sakamoto, S.; Ando, S.; Maeda, M.; Higaki, M.; Yoshizawa, H.; Murai, H.; Nakamura, Y. Bioorg. Med. Chem. 2015, 23, 2247. 32. McDonald, I. K.; Thornton, J. M. J. Mol. Biol. 1994, 238, 777. 33. Efimov, A. V.; Brazhnikov, E. V. FEBS Lett. 2003, 554, 389. 34. Morgan, B. P.; Scholtz, J. M.; Ballinger, M. D.; Zipkin, I. D.; Bartlett, P. A. J. Am. Chem. Soc. 1991, 113, 297. 35. Wood, J. M.; Bold, G.; Buchdunger, E.; Cozens, R.; Ferrari, S.; Frei, J.; Hofmann, F.; Mestan, J.; Mett, H.; O’Reilly, T.; Persohn, E.; Rosel, J.; Schnell, C.; Stover, D.; Theuer, A.; Towbin, H.; Wenger, F.; Woods-Cook, K.; Menrad, A.; Siemeister, G.; Schirner, M.; Thierauch, K. H.; Schneider, M. R.; Drevs, J.; Martiny-Baron, G.; Totzke, F. Cancer Res. 2000, 60, 2178. 36. Langenberg, M. H.; Witteveen, P. O.; Roodhart, J.; Lolkema, M. P.; Verheul, H. M.; Mergui-Roelvink, M.; Brendel, E.; Kratzschmar, J.; Loembe, B.; Nol-Boekel, A.; Christensen, O.; Schellens, J. H.; Voest, E. E. Ann. Oncol.: Off. J. Eur. Soc. Med. Oncol./ESMO 2011, 22, 2508. 37. Raghav, K. P.; Blumenschein, G. R. Expert Opin. Investig. Drugs 2011, 20, 859.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
Contents lists available at ScienceDirect
Bioorganic & Medicinal Chemistry journal homepage: www.elsevier.com/locate/bmc
Kinase hinge binding scaffolds and their hydrogen bond patterns Li Xing a,⇑, Jacquelyn Klug-Mcleod b, Brajesh Rai b, Elizabeth A. Lunney c a
Pfizer Worldwide Research and Development, 200 Cambridge Park Drive, Cambridge, MA 02140, United States Pfizer Worldwide Research and Development, 1 Eastern Point Road, Groton, CT 06340, United States c Pfizer Worldwide Research and Development, 10777 Science Center Drive, San Diego, CA 92121, United States b
a r t i c l e
i n f o
Article history: Received 21 May 2015 Revised 24 July 2015 Accepted 8 August 2015 Available online xxxx Keywords: Protein kinase Kinase hinge Hydrogen bonds Hinge scaffolds Potency and selectivity Crystal structure database
a b s t r a c t Protein kinases constitute a major class of intracellular signaling molecules, and describe some of the most prominent drug targets. Kinase inhibitors commonly employ small chemical scaffolds that form hydrogen bonds with the kinase hinge residues connecting the N- and C-terminal lobes of the catalytic domain. In general the satisfied hydrogen bonds are required for potent inhibition, therefore constituting a conserved feature in the majority of inhibitor-kinase interactions. From systematically analyzing the kinase scaffolds extracted from Pfizer crystal structure database (CSDb) we recognize that large number of kinase inhibitors of diverse chemical structures are derived from a relatively small number of common scaffolds. Depending on specific substitution patterns, scaffolds may demonstrate versatile binding capacities to interact with kinase hinge. Afforded by thousands of ligand–protein binary complexes, the hinge hydrogen bond patterns were analyzed with a focus on their three-dimensional configurations. Most of the compounds engage H6 NH for hinge recognition. Dual hydrogen bonds are commonly observed with additional recruitment of H4 CO upstream and/or H6 CO downstream. Triple hydrogen bonds accounts for small number of binary complexes. An unusual hydrogen bond with a non-canonical H5 conformation is observed, requiring a peptide bond flip by a glycine residue at the H6 position. Additional hydrogen bonds to kinase hinge do not necessarily correlate with an increase in potency; conversely they appear to compromise kinase selectivity. Such learnings could enhance the prospect of successful therapy design. Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction The protein kinases encoded in the human genome are involved in regulating essentially all aspects of cellular processes, from metabolism and cell cycle progression to differentiation and apoptosis. Small molecule kinase inhibitors play promising roles in treating a number of diseases including inflammation, neurodegenerative disorders, cardiovascular diseases and cancer.1 Hence they constitute a major component of the pipelines of drug discovery research and development in pharmaceutical industry. At the moment most of the approved kinase inhibitors for clinical use are in oncology, with Tofacitinib being the only marketed therapy for rheumatoid arthritis. Protein kinase enzymes phosphorylate the hydroxyl-containing amino acids (e.g. serine, threonine or tyrosine) under sequencespecific contexts. Despite the large variety of cellular functions Abbreviations: KD, kinase domain; CSDb, crystal structure database; HB, hydrogen bond; RMSD, root mean squared distance; P-loop, phosphate-binding loop; 3D, three-dimensional. ⇑ Corresponding author. Tel.: +1 (617) 665 5369; fax: +1 (617) 665 5575. E-mail address: [email protected] (L. Xing).
associated with protein kinases, their catalytic domains are structurally conserved. The approximately 300 amino acids of the kinase domain are structurally segregated into two distinct lobes. The smaller N-terminal lobe consists of mainly beta sheets and one conserved alpha helix (C-helix). The larger C-terminal lobe is mostly helical and contains the activation segment, which includes residues that in many kinases are phosphorylated upon kinase activation. The hinge region connects the two lobes. ATP binds in the folding cleft located between the two lobes forming conserved hydrogen bond interactions with the hinge. Protein kinases catalyze the transfer of the gamma phosphate group from ATP to the recipient substrate. Targeting the ATP binding site in drug design has led to discovery of numerous inhibitor templates that can bind to members of the kinase family. Protein kinases themselves can be activated upon phosphorylation, often on the activation loop. X-ray crystal structures have elucidated that catalytically active kinases share almost identical conformations, whereas unactivated kinases have diverse structures. Two frequently observed inactive conformations expose additional pockets that are accessible to kinase inhibitors, commonly termed DFG-out and C-helix-out.2,3 Partially occupying
http://dx.doi.org/10.1016/j.bmc.2015.08.006 0968-0896/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
2
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
the active site, the DFG-out conformation directly precludes binding of ATP. The C-helix-out state, on the other hand, can spatially accommodate ATP, but the catalytic machinery is disrupted. With the broken salt bridge between the conserved glutamate and lysine residues, the kinase is rendered inactive, incapable of turning over substrates. Based on surveys of published and Pfizer internal crystallographic data, the structures of over 200 distinct protein kinases have been resolved. Kinases with displaced C-helices or DFG-out conformation of the activation segment each account for approximately 30% of the total structures. Typically compounds containing certain chemical moieties help stabilize the inactive kinase conformation, for example, an urea and/or amide group extending toward the activation loop to facilitate DFG-out, or a hydrophobic aryl in the back pocket to secure the C-helix-out. Some kinases have been shown to naturally rest in the inactive states. For example, the C-helix rotates out in the crystal structure of the apo unphosphorylated BTK-KD.4 In complex with a number of compact ligands well confined in the adenine pocket, the DFG motif and activation loop of cFMS are in the canonical inactive conformation.5 Furthermore, a few kinases have shown unlimited versatility of adopting both DFG-out and C-helix-out capabilities, including SRC, ABL and CDK6.6–13 Kinase structures with simultaneous C-helix-out and DFG-out is possibly a rare event given the redundant mechanism of kinase inactivation, but evidently not mutually exclusive as demonstrated by X-ray structures including EGFR, FAK, BMX and PDK1 kinases.14–17 For ATP competitive inhibitors, hydrogen bond interaction with the kinase hinge is usually indispensable for potent inhibition, regardless of the kinase conformational state that is targeted. In order to make better decisions to select, design and prioritize multiple chemotypes of choice for lead optimization, understanding the interaction patterns of the primary scaffolds that anchor the compounds via hinge hydrogen bonds is of paramount importance. Recently putative hinge-binding fragments were compiled from decomposing known kinase inhibitors mapping to kinase pharmacophore models.18 We have previously reported the extraction of kinase hinge binding scaffolds from large body of X-ray structural data.19 Herein we focus on the statistical analysis of the signature hydrogen bonds in terms of their three-dimensional (3D) properties. With an attempt to understand the relationship between the propensity of the key molecular interactions and their implication for biological activity as well as kinome specificity, the learnings presented here could provide guidance for novel compound design. 2. Methods 2.1. Crystal structure database (CSDb) Kinase protein/ligand crystal complexes were retrieved from the in-house crystal structure database, a large collection of X-ray structure repository that contains both internally solved structures and those selected and imported from the Protein Data Bank.20 A total of 3980 crystal structures of kinases in complex with small, drug-like ligands exemplifying a variety of ligand binding modes were pulled from the database. All of them are better than 3.5 Å in resolution, of which 85% are higher resolution structures of 62.5 Å. These data were then utilized to mine and extract the hinge interacting fragments. An important aspect of this database was that all structures from a given family were aligned onto a common frame of reference based on superposition of preselected residues within the ATP site, making the subsequent analysis of the data derived from this database particularly efficient. The structural diversity of the compounds in the data set was analyzed. When subjected to a nearest-neighbor analysis, the compounds were found to cover different structural series of 670
clusters and 369 singletons, according to Ward’s similarities (distance cutoff of 0.1) using CDK-Daylight fingerprints. The diverse representation of the chemical classes attests the data set is not biased by any specific structural template, and the trends identified are generally applicable. 2.2. Extraction of hinge scaffolds The amino acids on kinase hinges are fully annotated from H1 to H13, with H3 corresponding to the gatekeeping residue.19 A set of criteria was defined to guide hinge extraction as described previously.19 Using the kinase annotation table, the ligand atoms that make hydrogen bonds with the hinge residues are identified. The rest of the molecules were then traversed along the connectivity map to complete the extraction of the core structures, including all ring fragments, the hetero-atoms and functional groups directly bonded to the rings containing cyano, nitro, nitroso, carbonyl, amide, urea, carboxylate, sulfone, sulfonamide, sulfoxide, sulfinic and sulfonic acids.19 The rest of the structural substituents comprising atoms and bonds that were farther away from the hinge were then removed. Hydrogen atoms were added to fill the open valences. The geometric criteria of a hydrogen bond interaction X–H Y–Z are: X Y distance <3.4 Å, and both XHY and HYZ angles >90°. In the end, a total of 595 unique hinge scaffolds were identified from the nearly 4000 kinase inhibitors in CSDb. 2.3. Binding modes of hinge scaffolds Each hinge binding scaffold, defined by a unique structural representation, could orient itself in many different ways in the ATP pocket. Frequently, each altered positioning presents distinct interactions with the kinase hinge. To account for such orientational and positional flexibility, each hinge scaffold is clustered into discrete representative conformations based on their in-place root-mean-squared-distance (RMSD) involving all heavy atoms. If the root-mean-squared-distance (RMSD) is less than or equal to 1 Å the positions are considered degenerative, otherwise a different binding conformation is recorded for the same hinge scaffold. 2.4. Kinase activity and selectivity data Kinase activity data against targets denoted by X-ray experiments were obtained by querying against Pfizer screening database for the corresponding screens, tagged with IC50 as the endpoint type. Multiple IC50 measurements for the same target were retained only if they fell within the same order of magnitude and the values were averaged. The final dataset consisted of 1502 ligands with ranging activity values from 0.01 nM to 100 lM. Kinase selectivity data were extracted from multiple screening panels measured both in-house as well as by commercial vendors, for compounds that appear in CSDb. For each selectivity panel, compounds were tested at either 1 lM and/or 10 lM concentrations at ATP Km for the corresponding kinase. The Gini coefficient at each screening concentration were computed using the percent inhibition data. With a range between zero and one, the Gini coefficient has been commonly used to describe selectivity of kinase inhibitors since its introduction by Graczyk.21 3. Results and discussion 3.1. Hinge binding scaffolds The protein kinases bind a common endogenous ligand. The ATP binding pocket is situated at the folding cleft of the N- and C-lobes, which are connected by the hinge region. Interactions are formed
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
3
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
H N
O
NH2
NH2 N
N
N
N
N
NH2
N N H
N O
N
N
N
H N
HO NH2
Core4 (42) Adenine
Core2 (24) Staurosporine
Core88 (22) 1H-pyrazole-3-amine
Core3 (20) 2-aminopyridine
NH2
H N
N
N
N
N
Core138(18) 1H-indazole
N H
Core64 (18) 7H-pyrrolo[2,3-d]pyrimidin-4-amine
H2N
Core124 (18) 1H-pyrrolo[2,3-b]pyridine
N
N
N
H N
N
NH2
Core13 (17) pyrimidine-4,6-diamine
Core0 (15) pyridine
Core285 (10) 7H-pyrrolo[2,3-d]pyrimidine
Figure 1. Ten most common hinge scaffolds. In parenthesis are the number of different kinases found in binary complexes of CSDb.
with the hinge backbone through two hydrogen bonds: the adenine N1 is an acceptor with the backbone NH (H6) and the 6-amino group is a donor to the backbone CO (H4). Furthermore, the carbonyl group of the same H6 residue is oriented toward the pocket for additional hydrogen bond formation with an exogenous ligand. Effective interaction in the ATP pocket that complements the topological and the electrostatic characteristics of the kinase hinge is essential for potent inhibition of protein kinases.22 The ten most common scaffolds found in the CSDb are illustrated in Figure 1, along with the number of kinases that they complex in. The adenine (Core4) is notably the most nondiscriminative core, crossing over to 42 different kinases. Most of the ensuing compounds are either ATP analogs or its derivatives, due to the fact that adenine has very few valences available for chemistry derivatization. The second most common scaffold is exemplified by staurosporine-type of compounds (Core2). Based on this potent yet non-selective template, many staurosporine analogs have been developed for kinase research. Most of the common scaffolds employ two HB features, specifically a donor– acceptor pair. Depending on the specific binding conformation, they either strictly replicate the hinge contacts of ATP, or alternatively form HBs with H6 NH and CO simultaneously. The amino pyrazole (Core88) scaffold has the potential of forming three hydrogen bonds. In the most productive binding mode, its protons and lone pair engages H4 CO, H6 NH and H6 CO in sequence as exemplified by multiple crystal structures including SLK (PDB code: 2JA0, 2UV2, CAMK1d (PDB code: 2JC6, 2VN9), Jak2 (PDB code: 2XA4, 3Q32), ITK (PDB code: 3MJ1), etc. On the other hand, the pyridine (Core0) can only form one HB by accepting a proton from H6 NH. More discussion on conformational variation in ATP pocket for a given scaffold follows in next section. It is noted from the analysis that the majority of the scaffolds (72%) are in unique complex with one specific kinase. They could be either under-explored by medicinal chemistry and/or lack of crystallographic execution. Additionally they may present valuable opportunities for improving kinome selectivity and/or generating chemical novelty in the exceedingly crowded kinase field.19 3.2. Varying binding modes of hinge scaffolds A single hinge scaffold could present itself in many different orientations in the ATP binding site. As depicted in Figure 2, in its canonical conformation of binding kinase, the adenine of ATP donates a proton to the backbone carbonyl of H4, and accepts a proton from the backbone NH of H6 via N1 of the purine. In
H4 N
H5
N
O
O N
H6 N
O
Figure 2. A single hinge scaffold could present multiple binding orientations as exemplified by adenine in CSDb. The canonical binding conformation corresponding to ATP binding is colored in green in ball-and-stick model.
addition, alternative orientations are identified from CSDb mining, of which two representations are illustrated in Figure 2. In the magenta orientation, the imidazole of the adenine donates its proton to H4, and the N3 of the purine accepts the NH proton from H6. Alternatively, the third mode colored in cyan is the result of a 120° counter-clock-wise rotation coupled with a translational move toward downstream residues of the hinge from the canonical state. In this mode, the adenine forms hydrogen bond with H6 but no longer with H4 residue. Specifically, the imidazole nitrogen accepts the proton from H6 backbone NH, and the amino moiety donates its proton to the backbone carbonyl of H6. In addition to these three representations, other variations are introduced by overall shifts along the hinge relative to the gatekeeper residue, movements toward or away from the hinge, as well as translation in the vertical direction relative to the P-loop. Out of the nearly 600 hinge scaffolds we identified in total 812 differential binding modes. For the top ten most common scaffolds, the number of binding modes is plotted in Figure 3, along with the percentage on the minor y-axis of the total number of binary complexes for the respective scaffolds. All of them exhibit some capacity of repositioning and/or re-orienting in the ATP site, albeit at different degrees of versatility. Adenine presents the most variations in the binding site, owning to its abundant composition
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
4
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
NH2 N
N
N
N
N
NH2 N
NH2
N
N N H
Core285
Core0
Core13
Core64
Core138
Core124
Core3
Core88
Core2
Core4
% Of Binding Modes
Number Of Binding Modes
N
Figure 3. Number of different binding modes observed for the top ten most common kinase scaffolds as plotted on the major y-axis. The minor y-axis depicts the different modes as a percentage of the total number of kinase complexes observed for each scaffold.
of heteroatoms affording multiple alternatives of hydrogen bond donating and accepting patterns. The large number of exemplary structures in CSDb also enriches its observed rotational as well as translational versatility. In resemblance, the aminopyrimidine and aminopyrrolopyrimidine are amongst the ones that afford varying modes of interactions with kinases. Different substitution patterns off the available valences would define different binding modes in order to present the overall molecule in its highest 3D complementarity in the protein pocket. In comparison, the binding mode of the staurosporine core is much more conserved. Except for the translational fluctuations occasionally observed in the binding site, no rotational motion was found. Similarly for the pyridine core various translational movements were observed given its small size, but no orientational changes due to its limited HB potential. 3.3. Hinge hydrogen bonds Hydrogen bond is among the most important specific interactions in biological recognition processes. In particular for kinases, hydrogen bonds with the hinge backbone residues are considered the anchor of ligand binding, a generally indispensable interaction for potent enzyme inhibition. We examine the intermolecular hydrogen bond patterns in detail in the inhibitor-kinase complexes. As illustrated in Figure 4a, the majority of the hydrogen bonds were made with the NH of H6, which accounts for twothirds of the crystal structures. In such context it is worth noting that the PIM kinases have a proline at position H6 that is conserved for the family (PIM1, 2, and 3), and thus are devoid of H6 NH donor. In CSDb a total of 12 PIM structures were available, of which the ligands make primary contacts with H4 CO for hydrogen bond interaction. In total, approximately one third of the ligands donate a proton to H4 CO, and slightly less than one-third donate to H6 CO. The predominant interactions with H6 NH is about twice the number of interactions with H4 CO. A former report based on PDB survey drew similar results with regard to hinge hydrogen bonds.23 It appears to be consistent with the theoretical projection
that unsatisfying a hydrogen bond donor in a buried protein environment is more energetically costly than burying an acceptor.24 The high propensity of the H6 NH pairing could also be facilitated by its spacial accessibility to the ligand molecule, situated between H4 CO and H6 CO. In contrast, the H4 is sterically more hindered given its close proximity to the gatekeeping residue H3, the side chain of which, depending on its character, typically forms important contacts with the kinase inhibitor. Large body of research has established that complementary interactions with the gatekeeper is critical to drive potency as well as selectivity, by targeting differences in amino acids at the gatekeeping position of related kinases.25 It is conceivable that intermolecular contact with the gatekeeper side chain would sterically block hydrogen bond engagement with H4 NH. As a matter of fact no HB is observed with H4 NH in our current analysis. Indicated by Figure 4a, some scaffolds extend hydrogen bond interactions toward downstream H7, by either accepting from H7 backbone amine, or less commonly donating to its backbone carbonyl. Hydrogen bond with H5 is extremely rare. The reason lies in the ensuing conformation of the peptide bonds along the kinase hinge. In the canonical kinase conformation denoted in Figure 2, both the H5 NH and CO point away from the binding pocket, geometrically precluding them from partnering with the ligand. However, there are exceptions. Through mining of thousands of crystal structures, we did observe the unusual hydrogen bond with H5 CO for a small number of structures, and it is made possible by a peptide bond flip connecting H5 and H6. Detailed discussion will be provided in the following section. The propensity of multiple hydrogen bond formation is of interest in comparison with the single point hinge contact. The pairwise hydrogen bond patterns with any two of the three hinge elements (H4 CO, H6 NH and H6 CO) are depicted in Figure 4b, along with the propensity of satisfying all three hydrogen bonds. Amongst them the dual hydrogen bonds with H4 CO and H6 NH is the most prominent. Almost a quarter of the CSDb structures present such hinge interactions, strictly replicating the hydrogen bond pattern
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
5
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
867 2623
577 413 1330 1042
(b)
H4_O & H6_N & H6_O
(a)
H6_N & H6_O
H7_N
H6_O
H6_N
H5_O
H5_N
H4_O
H4_N
H7_O
27
H4_O & H6_O
185 7
H4_O & H6_N
8
Figure 4. Hinge hydrogen bond distributions in CSDb. (a) Number of hydrogen bond occurrences for each bearing backbone of hinge residues. (b) Number of structures engaging multiple hydrogen bonds with kinase hinge.
of ATP. In comparison, the dual hydrogen bonds with H6 NH and H6 CO are less frequent, accounting for 15% of the total hinge interactions. Very rarely the dual hydrogen bond pattern with H4 CO and H6 CO is observed. In general it is challenging to install two donors in the conforming geometry to satisfy such 3D requirement. Furthermore, in the absence of water-mediated interaction, leaving H6 NH unsatisfied is thermodynamically disfavored, burying a hydrogen bond donor in an enclosed protein environment. Once the two donors are approximately placed in a scaffold, an accepting atom in-between could naturally add a third HB with H6 NH, depotting the scaffold into the class of three-point contacts. As depicted in Figure 4b, a number of structures utilize all three HB points from H4 CO, H6 NH and H6 CO. The frequency of three-point contacts is appreciably notable, accounting for more than 10% of the total kinase structures in CSDb. Hydrogen bond distances, a reflection of the strength of the interaction, are analyzed in Figure 5 by box plots. Both the longer distances and the broader distributions of the observed distances could be indicators of weaker HB interactions. The median
Hydrogen Bond Distance (Å)
3.4
3.2
3.0
2.8
2.6
2.4
H4_O
H6_N
distances for the backbone CO’s are almost identical: 2.88 Å for H4 CO and 2.85 Å for H6 CO. This suggests similar HB interactions with the two carbonyls of the kinase hinge, with H6 CO displaying a slightly broader distribution. On the other hand, the median distance with H6 NH (2.95 Å) is noticeably longer than the backbone carbonyls. In agreement, a previous survey of PDB reported a median distance of 2.9 Å for NH donors, approximately 0.15 Å larger than the OH donors.26 Possibly due to the larger van de Waals radius of nitrogen than oxygen atom, the hydrogen bonds of NH donors and N acceptors could exhibit slightly longer distance in distribution. We applied the general criteria of angular definition of hydrogen bonds in the detection process, but did not analyze the detailed angle distributions. The angular preferences of hydrogen bonds are important, but could be less profound in protein structures. Within boundary, the protein structure could relax to adapt to a specific ligand, in many more ways than the small molecule crystal structures. Additionally, when macromolecular X-ray structures are fitted to electron densities the small molecule conformation often play subordinate role. Given that the overall resolution of protein structure is much lower than that of small molecules, the angular terms could bear much imprecision and was used as a filtering criterion. Weak interactions involving CF H, C(Cl/Br) O@C, CH O and NH p have been observed to contribute to ligand–protein binding in a subtly directional manner. Sometimes they are termed pseudohydrogen bonds. Calculations suggested that CH O@C interactions are about one-half the strength of an NH O@C hydrogen bond.27 Similarly, halogen bonds are significantly weaker than hydrogen bonds. Ab initio calculations on formaldehydehalobenzene dimers estimated gas phase interaction energies below 2.5 kcal/mol, corresponding to the strength of CH O pseudohydrogen bonds.28 These categorically weaker interactions are second tier to hydrogen bonds. Intermolecular interactions involving pseudohydrogen bonds and/or halogen bonds await further analysis in a future communication.
H6_O 3.4. Unusual hydrogen bond with H5
Figure 5. Distribution of hydrogen bond distance between ligand and kinase hinge. Average of 2.9 Å is similar to survey from small molecule X-ray with broader distributions. Box regions correspond to 50% of the distribution, lines extend to max 1.5 times of this interval, and dots are outliers.
As evidenced by previous analysis, the most common HBs are formed with H4 CO, H6 NH and H6 CO, with varying numbers of
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
6
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
kinase inhibitors satisfying one, two, or three of the hydrogen bond donating and accepting partnerships. This pattern is largely attributed to the typical backbone orientations along the kinase hinge, which directs CO and NH groups in alternation between pointing toward and averting away from the ATP pocket in a sequential manner (as depicted in Fig. 2). As a result, whereas CO of H4, NH and CO of H6 are naturally oriented toward the ligand binding pocket, the NH and CO of H5 in their inherent orientations point away from the active site. Consequently, intermolecular hydrogen bonds are generally not observed with H5 residue. It is intriguing that an unusual hydrogen bond with the H5 CO emerged from our analysis (Fig. 6). This interaction is embodied by the PDB structure (code: 3HMP) for the dual-specificity protein kinase Mps1 in complex with Cpd 4, a low affinity compound (IC50: 38 lM),29 the diaminopyridine-based compounds,30,31 as well as several in-house crystal structures. Comparing the holo structure with the apo Mps1 protein we found an interesting conformational switch along the kinase hinge. In the apo Mps1 structure (PDB code: 3DBQ) the CO of H5 points in the opposite direction of ligand binding, as is generally observed with the canonical hinge conformations. Upon binding the ligand, an HB donor is presented by the 4-amino group attached to the quinazoline moiety of Cpd 4, which effectively shifts the thermodynamic equilibrium of protein conformations. As a result, the peptide bond linking H5 and H6 undergoes an almost 180° rotation to satisfy a non-conventional hinge hydrogen bond between the H5 CO and the amine of Cpd 4. Toward confirming this is not a crystallographic artifact, we identified five additional in-house X-ray structures of Mps1 with compounds of similar HB donating capacities. The different inhibitor-kinase complexes exhibit the consistent HB interaction with H5 CO, afforded by the backbone flip of the peptide bond. On the other hand, in the other Mps1 binary complexes (e.g. PDB code: 4JS8, 4JT3 etc.) where the ligands do not employ a proper donor in the vicinity of H5 CO, the observed backbone conformations are consistent with the apo structure. From mining the crystal structure database, the reversed conformation of H5–H6 peptide bond is only observed for Mps1 kinase. We think this is an uncommon event where certain sequences along the kinase hinge may be required. Specifically the glycine reside immediately downstream of H5, Gly605 (H6)
in Mps1, could be instrumental in providing the necessary flexibility to accommodate the backbone flip. Devoid of a side chain, glycine residue is unique in its superfluous flexibility that is otherwise restricted in other amino acids. In analogy, the H6–H7 peptide bond flip has been reported for p38 kinase, in response to binding of a pyridinone chemotype where the second hydrogen bond is engaged with the re-arranged hinge conformation.25 The H7 residue, Gly110 of p38, was found critical to allow the flipped conformation. We mined the human kinase genome for propensity of glycine at the H6 position, and found it is conserved in five additional kinases: SNRK and TSSK2 from CAMK family, COT from STE family, MOS and TOPK from others. Given the rare occurrence of a glycine at the H6 position, the probability of the H5–H6 amide flip is predicted to be extremely low. Crystal structures of the aforementioned kinases are in need to elucidate additional conformational flexibilities. 3.5. More hinge hydrogen bonds do not yield higher potency In a buried environment it is thermodynamically favorable for the NH and CO groups to form hydrogen bonds, and the number of observed hydrogen bonds increases with decreasing solvent accessibility.32,33 Since the hinge section of the ATP pocket is less solvent accessible, additional HB contacts could potentially contribute to binding affinity. In Figure 7 the inhibitory pIC50 values for the ligands bound to their respective kinases were presented in box plots. The x-axis denotes the hydrogen bond partners from the hinge. As discussed previously, very few examples present dual hydrogen bonds with H4 CO and H6 CO, and only two of them have activity data, thus they were taken out of the statistical analysis. The number of data points for H4 CO and H6 CO single HB patterns were also small, 22 and 20 respectively, thus they were excluded due to insufficient sample size. The focus was on the singular HB with H6 NH containing 432 data points, double HB patterns of H4_CO/H6_NH (491 data points) and H6_NH/H6_CO (265 data points), and triple hydrogen bonds with H4 CO, H6 NH and H6 CO (208 data points).
10 9
pIC 50
8 7 6 Cys604 (H5)
5
N N H
N
Cpd 4
Figure 6. Unusual hydrogen bond with H5 carbonyl of Mps1/TTK, afforded by protein backbone flip of H5–H6 upon ligand binding. Apo Mps1 kinase is in yellow (PDB code: 3DBQ), and the binary complex with Cpd 4 (in gold) is colored in cyan (PDB code: 3HMP).
Two hydrogen bonds
H4_O & H6_N & H6_O
Cl
H6_N & H6_O
Gly605 (H6)
H4_O & H6_N
H6_N
4
Figure 7. More hydrogen bonds doesn’t translate to higher potency. Box regions correspond to 50% of the distribution, lines extend to max 1.5 times of this interval, and dots are outliers.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
7
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
The distributions show no significant trend of activity values as a function of the number of HBs. The median IC50’s are within three fold for compounds making one, two or three hydrogen bonds with the kinase hinge, which is within experimental uncertainty. In agreement, the analysis on nine approved kinase drugs revealed no correlation between the number of hinge hydrogen bonds and ligand affinity.23 This suggests the contribution of hydrogen bond to binding affinity is highly context-dependent. Usually, leaving a buried donor or acceptor unsatisfied results in a free energy penalty. On the other hand, desolvation of the donor and acceptor must occur for the hydrogen bond to form, such that the effects of hydration and hydrogen bond formation nearly cancel out. Toward that end, it has been shown that removing a hydrogen bond by an NH through a methyl replacement hardly reduced the binding affinity of peptidomimetic thermolysin inhibitors.34 In the context of kinase hinge interaction, whereas one or more of the neighboring CO and/or NH may not be satisfied, they are in general shielded by a non-specific interaction with a heavy and/or hydrogen atom. Beyond hinge, complementarity of attractive and repulsive interactions in other parts of the ATP pocket is also essential in defining the ultimate potency of a kinase inhibitor. We also examine the relationship between potency and HB distance, which represents a geometric measure of the strength of the interaction. The hydrogen bonds with H6 NH were analyzed since it has the highest propensity, totaling 1396 data points. As shown in Figure S1, most of the hydrogen bond distances are within the optimal range of 2.8–3.2 Å (1093, 78%). The potency distributions are essentially comparable, except for the HB distances that fall remarkably out of the optimal range, for example, <2.6 Å and/or >3.4 Å, which suggests interactions in a forced environment could be detrimental to binding affinity. 3.6. More hydrogen bonds could compromise selectivity Since hinge hydrogen bonds are provided by backbone atoms in the segment of kinase structures that are largely conserved by conformation, they are generally regarded as potency anchors.
It is of interest to investigate the implication of the hydrogen bond patterns in defining kinome specificity. Using kinome screening panel results at 1 lM and 10 lM, the distributions of Gini coefficients as a selectivity measure were plotted against the number of hydrogen bonds (Fig. 8). The sample sizes are sufficient to render statistical significance: there are in total 811 Gini coefficients computed from 1 lM data and 782 from 10 lM, and each box distribution contains between 123 and 272 data points. Not surprisingly, the 1 lM percent inhibition data consistently suggests higher specificity than the 10 lM measurements. The median Gini coefficients range from 0.47 to 0.57 for 1 lM, and 0.32 to 0.53 for 10 lM respectively. The overall distribution profiles are similar for the two measured concentrations. Distribution of Gini coefficients is also examined with respect to hydrogen bond distance of H6 NH for 1 lM and 10 lM concentrations separately (Fig. S2). For the full range between 2.6 Å and 3.4 Å, the median Gini coefficients fluctuate between 0.3 and 0.6. It appears that compounds of one hydrogen bond have higher specificity, and those with three hydrogen bonds are among the ones of lowest Gini coefficients. This supports the notion that enhancing interactions with the conserved regions of protein structures could compromise specificity in general. At a minimum, kinome selectivity does not seem to benefit from additional hydrogen bonds with the kinase hinge. Not only the scaffolds alone, but also substitution patterns beyond the core region have large impact on the selectivity profiles. The practice of tailoring design to complement specific interactions employing other structural elements of the ATP site, for example, the P-loop, the activation loop, and/or the C-helix, has been successfully employed by medicinal chemists. Built upon even the most common hinge element, prominent selectivity can be achieved for the kinase target of interest by exploiting subtle differences in amino acids that compose the ATP binding site. Further investigation reveals that due to the stringent 3D configurations of the donor–acceptor-donor pattern, many three-point contacts display one or more poor hydrogen bonds, for example, heavy-to-heavy atom distance of less than 2.8 Å and/or greater
1µM
Two hydrogen bonds
Three hydrogen bonds
H6_N & H6_O
H4_O & H6_N
H6_N
Gini Coefficient
10µM
Figure 8. Gini coefficients computed from 1 lM (blue) and 10 lM (yellow) kinase panel screen. More hydrogen bonds could compromise selectivity. Box regions corresponds to 50% of the distribution, lines extend to max 1.5 times of this interval, and dots are outliers.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006
8
L. Xing et al. / Bioorg. Med. Chem. xxx (2015) xxx–xxx
than 3.2 Å (330 out of 410, 80%). Thus the additional hydrogen bonds to the kinase hinge may not add to binding free energy as favorably as expected due to their sub-optimal geometry. This could impact both potency toward the primary target as well as selectivity against the off-targets. In medicinal chemistry design, we are usually faced with choices of hinge templates that are of one-, two-, or three-point contacts. While it is tempting to regard the scaffolds of more hinge contacts as superior, our data suggest the decision should be made in a case-specific manner. There is no general advantage for either potency or selectivity to fully satisfy all three hinge hydrogen bonds. Toward that end, the simple pyridine has emerged from our analysis as one of the most common scaffolds. The popularity of this single-point scaffold speaks to its utility and value, from which a number of late stage kinase inhibitors have been developed for oncology indication.35–37 4. Conclusion Kinase inhibitors can exert their modulation of target function by elaborate mechanisms. Type I inhibitors form H-bonds with the kinase hinge and occupy the adenine binding pocket. Type II inhibitors occupy the adenosine pocket, but also capture conformational changes that is typically either DFG-out or C-helix out, or both. The scope of the current investigation covers both type I and type II kinase inhibitors. Nearly 600 different hinge scaffolds were extracted from CSDb composed of Pfizer internal and public crystal structures of kinase-inhibitor complexes. Coordinates of crystal structures made possible the analysis of 3D properties characteristic of protein–ligand binding, in particular the hinge hydrogen bonds that are generally critical for potent kinase inhibition. The majority of hydrogen bonds are formed with H6 NH, a spatially accessible feature located in the middle of the kinase hinge. Dual hydrogen bonds are commonly observed with additional recruitment of H4 CO upstream and/or H6 CO downstream. Triple hydrogen bonds accounts for 10% of the total binary complexes, engaging all three prevailing features via the donor–acceptor-donor pattern of the inhibitor. Combining with the biological assay data, we demonstrate that more hydrogen bonds in the hinge region do not necessarily afford higher potency. Conversely, they may have deleterious effects on kinome selectivity. A single scaffold can adopt different orientations to interact with kinase hinge, a versatility that would be constrained by explicit substitution patterns beyond the hinge binding scaffold. Furthermore, an unusual hydrogen bond with a canonically unavailable conformation is observed, requiring a peptide bond flip conceivably made possible by a glycine residue. The compiled collection of hinge scaffolds along with their binding coordinates could serve as basis set for hinge hopping, a practice frequently employed to generate novel invention as well as to optimize existing leads in medicinal chemistry. Acknowledgement This study was sponsored by Pfizer Inc. Supplementary data Supplementary data (plot of activity (pIC50) versus hydrogen bond distance, plot of Gini coefficient versus hydrogen bond distance computed using 1 lM and 10 lM percent inhibition data, kinase scaffolds along with their 3D coordinates and activity data) associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.bmc.2015.08.006.
References and notes 1. Cohen, P. Nat. Rev. Drug Disc. 2002, 1, 309. 2. Jacobs, M. D.; Caron, P. R.; Hare, B. J. Proteins 2008, 70, 1451. 3. Brooijmans, N.; Chang, Y. W.; Mobilio, D.; Denny, R. A.; Humblet, C. Protein Sci.: Publ. Protein Soc. 2010, 19, 763. 4. Mao, C.; Zhou, M.; Uckun, F. M. J. Biol. Chem. 2001, 276, 41435. 5. Schubert, C.; Schalk-Hihi, C.; Struble, G. T.; Ma, H. C.; Petrounia, I. P.; Brandt, B.; Deckman, I. C.; Patch, R. J.; Player, M. R.; Spurlino, J. C.; Springer, B. A. J. Biol. Chem. 2007, 282, 4094. 6. Dar, A. C.; Lopez, M. S.; Shokat, K. M. Chem. Biol. 2008, 15, 1015. 7. Georghiou, G.; Kleiner, R. E.; Pulkoski-Gross, M.; Liu, D. R.; Seeliger, M. A. Nat. Chem. Biol. 2012, 8, 366. 8. Seeliger, M. A.; Nagar, B.; Frank, F.; Cao, X.; Henderson, M. N.; Kuriyan, J. Structure 2007, 15, 299. 9. Hennequin, L. F.; Allen, J.; Breed, J.; Curwen, J.; Fennell, M.; Green, T. P.; Lambert-van der Brempt, C.; Morgentin, R.; Norman, R. A.; Olivier, A.; Otterbein, L.; Ple, P. A.; Warin, N.; Costello, G. J. Med. Chem. 2006, 49, 6465. 10. Nagar, B.; Hantschel, O.; Seeliger, M.; Davies, J. M.; Weis, W. I.; Superti-Furga, G.; Kuriyan, J. Mol. Cell 2006, 21, 787. 11. Levinson, N. M.; Kuchment, O.; Shen, K.; Young, M. A.; Koldobskiy, M.; Karplus, M.; Cole, P. A.; Kuriyan, J. PLoS Biol. 2006, 4, e144. 12. Russo, A. A.; Tong, L.; Lee, J. O.; Jeffrey, P. D.; Pavletich, N. P. Nature 1998, 395, 237. 13. Cho, Y. S.; Borland, M.; Brain, C.; Chen, C. H.; Cheng, H.; Chopra, R.; Chung, K.; Groarke, J.; He, G.; Hou, Y.; Kim, S.; Kovats, S.; Lu, Y.; O’Reilly, M.; Shen, J.; Smith, T.; Trakshel, G.; Vogtle, M.; Xu, M.; Xu, M.; Sung, M. J. J. Med. Chem. 2010, 53, 7938. 14. Muckelbauer, J.; Sack, J. S.; Ahmed, N.; Burke, J.; Chang, C. Y.; Gao, M.; Tino, J.; Xie, D.; Tebben, A. J. Chem. Biol. Drug Des. 2011, 78, 739. 15. Michalczyk, A.; Kluter, S.; Rode, H. B.; Simard, J. R.; Grutter, C.; Rabiller, M.; Rauh, D. Bioorg. Med. Chem. 2008, 16, 3482. 16. Iwatani, M.; Iwata, H.; Okabe, A.; Skene, R. J.; Tomita, N.; Hayashi, Y.; Aramaki, Y.; Hosfield, D. J.; Hori, A.; Baba, A.; Miki, H. Eur. J. Med. Chem. 2012. 17. Nagashima, K.; Shumway, S. D.; Sathyanarayanan, S.; Chen, A. H.; Dolinski, B.; Xu, Y.; Keilhack, H.; Nguyen, T.; Wiznerowicz, M.; Li, L.; Lutterbach, B. A.; Chi, A.; Paweletz, C.; Allison, T.; Yan, Y.; Munshi, S. K.; Klippel, A.; Kraus, M.; Bobkova, E. V.; Deshmukh, S.; Xu, Z.; Mueller, U.; Szewczak, A. A.; Pan, B. S.; Richon, V.; Pollock, R.; Blume-Jensen, P.; Northrup, A.; Andersen, J. N. J. Biol. Chem. 2011, 286, 6433. 18. Zhao, H.; Caflisch, A. Bioorg. Med. Chem. Lett. 2015, 25, 2372. 19. Xing, L.; Rai, B.; Lunney, E. A. J. Comput. Aided Mol. Des. 2014, 28, 13. 20. Berman, H. M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T. N.; Weissig, H.; Shindyalov, I. N.; Bourne, P. E. Nucleic Acids Res. 2000, 28, 235. 21. Graczyk, P. P. J. Med. Chem. 2007, 50, 5773. 22. Noble, M. E.; Endicott, J. A.; Johnson, L. N. Science 2004, 303, 1800. 23. Ghose, A. K.; Herbertz, T.; Pippin, D. A.; Salvino, J. M.; Mallamo, J. P. J. Med. Chem. 2008, 51, 5149. 24. Barratt, E.; Bronowska, A.; Vondrasek, J.; Cerny, J.; Bingham, R.; Phillips, S.; Homans, S. W. J. Mol. Biol. 2006, 362, 994. 25. Xing, L.; Shieh, H. S.; Selness, S. R.; Devraj, R. V.; Walker, J. K.; Devadas, B.; Hope, H. R.; Compton, R. P.; Schindler, J. F.; Hirsch, J. L.; Benson, A. G.; Kurumbail, R. G.; Stegeman, R. A.; Williams, J. M.; Broadus, R. M.; Walden, Z.; Monahan, J. B. Biochemistry 2009, 48, 6402. 26. Bissantz, C.; Kuhn, B.; Stahl, M. J. Med. Chem. 2010, 53, 5061. 27. Vargas, R.; Garza, J.; Dixon, D. A.; Hay, B. P. J. Am. Chem. Soc. 2000, 122, 4750. 28. Riley, K. E.; Merz, K. M., Jr. J. Phys. Chem. A 2007, 111, 1688. 29. Chu, M. L.; Lang, Z.; Chavas, L. M.; Neres, J.; Fedorova, O. S.; Tabernero, L.; Cherry, M.; Williams, D. H.; Douglas, K. T.; Eyers, P. A. Biochemistry 2010, 49, 1689. 30. Kusakabe, K.; Ide, N.; Daigo, Y.; Itoh, T.; Higashino, K.; Okano, Y.; Tadano, G.; Tachibana, Y.; Sato, Y.; Inoue, M.; Wada, T.; Iguchi, M.; Kanazawa, T.; Ishioka, Y.; Dohi, K.; Tagashira, S.; Kido, Y.; Sakamoto, S.; Yasuo, K.; Maeda, M.; Yamamoto, T.; Higaki, M.; Endoh, T.; Ueda, K.; Shiota, T.; Murai, H.; Nakamura, Y. ACS Med. Chem. Lett. 2012, 3, 560. 31. Kusakabe, K.; Ide, N.; Daigo, Y.; Itoh, T.; Yamamoto, T.; Kojima, E.; Mitsuoka, Y.; Tadano, G.; Tagashira, S.; Higashino, K.; Okano, Y.; Sato, Y.; Inoue, M.; Iguchi, M.; Kanazawa, T.; Ishioka, Y.; Dohi, K.; Kido, Y.; Sakamoto, S.; Ando, S.; Maeda, M.; Higaki, M.; Yoshizawa, H.; Murai, H.; Nakamura, Y. Bioorg. Med. Chem. 2015, 23, 2247. 32. McDonald, I. K.; Thornton, J. M. J. Mol. Biol. 1994, 238, 777. 33. Efimov, A. V.; Brazhnikov, E. V. FEBS Lett. 2003, 554, 389. 34. Morgan, B. P.; Scholtz, J. M.; Ballinger, M. D.; Zipkin, I. D.; Bartlett, P. A. J. Am. Chem. Soc. 1991, 113, 297. 35. Wood, J. M.; Bold, G.; Buchdunger, E.; Cozens, R.; Ferrari, S.; Frei, J.; Hofmann, F.; Mestan, J.; Mett, H.; O’Reilly, T.; Persohn, E.; Rosel, J.; Schnell, C.; Stover, D.; Theuer, A.; Towbin, H.; Wenger, F.; Woods-Cook, K.; Menrad, A.; Siemeister, G.; Schirner, M.; Thierauch, K. H.; Schneider, M. R.; Drevs, J.; Martiny-Baron, G.; Totzke, F. Cancer Res. 2000, 60, 2178. 36. Langenberg, M. H.; Witteveen, P. O.; Roodhart, J.; Lolkema, M. P.; Verheul, H. M.; Mergui-Roelvink, M.; Brendel, E.; Kratzschmar, J.; Loembe, B.; Nol-Boekel, A.; Christensen, O.; Schellens, J. H.; Voest, E. E. Ann. Oncol.: Off. J. Eur. Soc. Med. Oncol./ESMO 2011, 22, 2508. 37. Raghav, K. P.; Blumenschein, G. R. Expert Opin. Investig. Drugs 2011, 20, 859.
Please cite this article in press as: Xing, L.; et al. Bioorg. Med. Chem. (2015), http://dx.doi.org/10.1016/j.bmc.2015.08.006