Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2023 Feb 23;19(5):1615–1628. doi: 10.1021/acs.jctc.2c01171

Strategy toward Kinase-Selective Drug Discovery

Mingzhen Zhang †, Yonglan Liu ‡, Hyunbum Jang †, Ruth Nussinov †,§,*
PMCID: PMC10018734  PMID: 36815703

Abstract

graphic file with name ct2c01171_0011.jpg

Kinase drug selectivity is the ground challenge in cancer research. Due to the structurally similar kinase drug pockets, off-target inhibitor toxicity has been a major cause for clinical trial failures. The pockets are similar but not identical. Here, we describe a transformation invariant protocol to identify distinct geometric features in the drug pocket that can distinguish one kinase from all others. We integrate available experimental structures with the artificial intelligence-based structural kinome, performing a kinome-wide structural bioinformatic analysis to establish the structural principles of kinase drug selectivity. We generate the structural landscape from the experimental kinase–ligand complexes and propose a binary network that encapsulates the information. The results show that all kinases contain binary units that are shared by less than seven other kinases in the kinome. 331 kinases contain unique binary units that may distinguish them from all others. The structural features encoded by these binary units in the network represent the inhibitor-accessible geometric space that may capture the kinome-wide selectivity. Our proposed binary network with the unsupervised clustering can serve as a general structural bioinformatic protocol for extracting the distinguishing structural features for any protein from their families. We apply the binary network to epidermal growth factor receptor tyrosine kinase inhibitor selectivity by targeting the gate area and the AKT1 serine/threonine kinase selectivity by binding to the αC-helix region and the allosteric pocket. Finally, we develop the cross-platform software, KDS (Kinase Drug Selectivity), for customized visualization and analysis of the binary networks in the human kinome (https://github.com/CBIIT/KDS).

Introduction

Kinases, the enzymes that catalyze the transfer of a phosphate group from ATP to the substrate, play pivotal roles in critical cell functions including cell cycle progression, growth, apoptosis, and metabolism.1,2 Dysregulated kinases are chief actors in instigating and promoting cancer, making them vital and urgent pharmaceutical targets.2−6 It has been estimated that over ∼30% of the drug discovery efforts worldwide have been focusing on kinase families.7 ∼62 small molecule kinase drugs have been approved by the US Food and Drug Administration (FDA), and hundreds are in the clinical trials.7−10

The clinical successes of kinase inhibitors depend on a balanced profile between efficacy and toxicity.11,12 The human genome encodes over 500 kinases (referred to as the human kinome).13 They are highly similar in their catalytic domain structures, with the ATP pocket in the cleft between the N- and C-lobes.14 While allosteric kinase inhibitors, with mechanisms involving stabilizing the inactive state and/or destabilizing the inactive state15 are increasingly being designed, to date, most inhibitors act by competing with ATP for the pockets16 or by an allosteric/orthosteric combination.17 Due to the highly similar ATP pockets, a kinase inhibitor designed to target one kinase frequently also binds to another.18,19 Off-target toxicity has been common in kinase inhibitors failing clinical trials.20 Numerous structure-based strategies have been proposed to investigate the kinase selectivity,21 including those targeting the ATP pocket,22 the allosteric pocket,23 and the DFG conformations.24 Achieving kinome-wide drug selectivity has been a long-term pursuit and remains the ground challenge both in academia and industry.25−28

Drug pockets of human kinases are highly similar but not identical.29,30 Here, we ask whether there are inhibitor-accessible local features in the kinase drug pockets in one kinase but not in all others in the human kinome that have been overlooked despite the extensive efforts. If so, their identification could guide the design and optimization of drugs with superior kinome-wide selectivity.31 To date, such structural features of kinase inhibitor selectivity have not been reported. We surmise that this could be due to two reasons: (i) a lack of structures for many of the kinases in the human kinome, which prevents a comprehensive kinome-wide structural bioinformatic analysis. There are 287 typical protein kinases with crystal/cryo-EM kinase domain structures in the protein data bank (PDB). The structures of other 208 typical protein kinases are unavailable (Table S1). (ii) A lack of a structural bioinformatic protocol for extracting atomic-resolution structural features from the highly dynamic kinase drug pockets. The kinase structural landscape is broad, with multiple states, including with αC-helix “IN” and “OUT”, and DFG “IN” and “OUT” conformations, in the “ON” (active) or “OFF” (inactive) states.24,32 A protocol based on static, sequence-based comparisons can provide important insights but only limited three-dimensional (3D) structural information.22 It may fail to capture the detailed, dynamic local conformational features in the geometric space of the kinase drug pockets, which is essential in kinase-selective drug discovery.33

In this work, we integrate the experimental kinase structures available in the PDB database and the structural kinome from the artificial intelligence (AI)-driven AlphaFold234,35 and describe a transformation invariant binary network protocol to reveal the structural principles of kinase drug selectivity at atomic resolution. Computer vision-based transformation invariant description is powerful, having shown its merit in structural biology in diverse problems, including amino acid sequence order-independent protein structure comparisons36 and docking.37 Here, we establish the inhibitor-accessible geometric space in the kinase drug pockets, in which a total of 44 structurally conserved inhibitor-accessible residues are identified. We calculate the kinase structural landscape from the experimental kinase structures and design a binary network that encapsulates the dynamic structural information. The basic component, referred to as the “binary unit”, consists of residue pairs in the kinase drug pocket. The 151 binary units in the network represent the feasible structural features that can be targeted by inhibitors for kinome-wide selectivity. Iterative comparisons show that all kinases contain a binary unit that is shared by less than seven other kinases in human kinome. 331 kinases, ∼66.9% of human kinome, contain a unique binary unit that can distinguish them from all other kinases. We apply the binary network protocol to epidermal growth factor receptor (EGFR) tyrosine kinase and AKT serine/threonine kinase, illustrating how the structural principles of kinase inhibitor selectivity established in this work may benefit kinase drug discovery toward enhanced kinome-wide selectivity, including for AKT isoform selectivity. A link to the cross-platform software, KDS (Kinase Drug Selectivity), is provided (https://github.com/CBIIT/KDS).

Results

Inhibitor-Accessible Geometric Space in Kinases

Here, we focus on 10 typical eukaryotic kinase families, AGC (58 kinases), CAMK (75 kinases), CK1 (12 kinases), CMGC (62 kinases), NEK (11 kinases), RGC (5 kinases), STE (56 kinases), TKL (34 kinases), TYR (90 kinases), and OTHER (81 kinases). Thirteen kinases, including OBSCN, RPS6KA1, RPS6KA2, RPS6KA3, RPS6KA4, RPS6KA5, RPS6KA6, SPEG, JAK1, JAK2, JAK3, TYK2, and EIF2AK4, have two kinase domains. Two kinases, PLK5 in the OTHER family and kinase domain 1 of OBSCN in the CAMK family, are excluded since their kinase domains are incomplete/truncated. As a result, a total of 495 kinase domains were collected and analyzed (Table S1). The atypical kinase families were not included here since many of them show distinct pocket morphologies and may follow different structural activation mechanisms.38−40

We first collected and analyzed the 4913 crystal/cryo-EM kinase structures available in the PDB to explore the inhibitor-accessible geometric space in kinase drug pockets. Among these, 4554 (∼93%) contain ligands (Table S2). Since some kinase structures contain multiple chains of kinase domains and the kinase domains in the same structure are not always structurally identical, the 4554 kinase structures were further split into a total of 6709 kinase–ligand complexes. By aligning the kinase–ligand complexes, the interactions of the inhibitors with the kinase drug pockets, which represent the entire inhibitor-accessible geometric space in the human kinome, can be visualized. We observed that inhibitors with highly diverse structures can access most parts of the cleft between the N-lobe and the C-lobe in the kinase domain (Figure 1A). For clarity, the inhibitor-accessible geometric space is divided into six regions based on their locations and functions, including ATP, αCbot, αCtop, β1, αD, and allosteric regions (Figure 1B). The ATP region is responsible for ATP binding. Type I kinase inhibitors usually bind to this region (Figure 1C).16 The αCbot region is between the bottom region of the αC-helix and β-strands 4–5 (β4−β5) with hydrophobic residues from the R-spine.41 Type I1/2 kinase inhibitors can access this area (Figure 1D).16 The allosteric region is in between the αC-helix in the N-lobe and αE-helix in the C-lobe, which can be targeted by type II kinase inhibitors (Figure 1E).16 The allosteric region usually becomes available when the DFG motif in the A-loop flips out. The αCtop region is above the hydrophobic αCbot region. Type I inhibitors such as the ERK inhibitor (PDB ID: 6G9N), type II inhibitors such as the CDK16 inhibitor (PDB ID: 5G6V), and type III inhibitors such as the MEK inhibitor (PDB ID: 3PP1) can interact with the αCtop region (Figure 1F). The β1 region and αD regions are more solvent exposed. They can be visited by type I, type I1/2, and type II kinase inhibitors, as well as the linkers in the kinase PROTACs.42,43

Figure 1.

Figure 1

Open in a new tab

Inhibitor-accessible geometric space in the kinase drug pockets. (A) The kinase–drug crystal/cryo-EM structures indicate that kinase inhibitors may access most parts of the cleft between the N-lobe and the C-lobe. (B) The inhibitor-accessible geometric space in the drug pocket consists of six subregions, including β1, αD, ATP, allosteric, αCbot, and αCtop regions. The type I (C), type I1/2 (D), type II (E), and type III (F) kinase inhibitors fit to various subregions in the inhibitor-accessible geometric space.

Inhibitor-Accessible Residues in Kinases

The inhibitor-accessible geometric space is composed of a total of 14 sub-domains, including seven β-strands (β1−β5 in the N-lobe and β6−β7 in the C-lobe), three α-helices (αC in the N-lobe and αD−αE in the C-lobe), and four loops [the αC−β4 loop, the hinge region between β5 and αD, the catalytic loop between αE and β6, and the activation loop (A-loop) following β7] (Figure 2A). In these subdomains, the side chains of 44 residues point toward the pocket and exhibit a potential to form direct interactions with the inhibitors. These residues are structurally conserved across 495 kinases. The 44 inhibitor-accessible residues in the kinase pocket are numbered from the N-terminal to the C-terminal (referred to as X1 to X44) and colored based on their sub-domains (Figure 2B). Due to the lack of experimental structures of ∼42.0% of the human kinome, we established them from the high-quality predictions by AlphaFold2.34,35 We used AlphaFold2 to predict the structures of all 495 human kinase domains. For kinases with structures available in the PDB, we verified that the predicted residue locations in the kinase domain are highly similar to the experimental ones. This validated AlphaFold as a reliable predictor of the kinase domain. For kinases without experimental structures, we then considered AlphaFold structures as the best estimations. We next superimposed all AlphaFold structures and manually inspected and validated the corresponding 44 inhibitor-accessible residues for all individual 495 kinases (Table S3). Of the 44 inhibitor-accessible residues for the 495, 14 are conserved, shared by over ∼50% (Figure 2C). These residues usually play essential roles in kinase functions. X3, X5, and X39 are the hydrophobic residues in the C-spine,44 stabilizing ATP’s adenine group. X6 is the basic residue in β3, interacting with the αC-helix and ATP’s phosphate groups. X11 is the acid residue in the αC-helix, interacting with X6 in the ON-state kinase. X14, X19, X36, and X44 are the residues in the R-spine,41 where X36 is His in the HRD motif in the catalytic loop, and X44 is Phe in the DFG motif in the A-loop. X25 and X29 are the residues in the hinge loop. X38–X40 are in the ATP region, and X43 is Asp in the DFG motif in the A-loop. X24 is the gatekeeper residue. The regions with X2–X3, X12–X14, X16–X20, and X22–X24 evolved to be hydrophobic across the 495 kinases. The αD region with X30–X31 shows more acidic residues. The positions of X26, X34, X36, and X44 favor residues with aromatic rings.

Figure 2.

Figure 2

Open in a new tab

Inhibitor accessibility in kinase drug binding pockets. (A) A total of 14 subdomains in the kinase domain are involved in the drug binding pocket, in which (B) 44 structurally conserved inhibitor-accessible residues can be identified. (C) A summary of the 44 inhibitors across 495 kinases indicates that some functionally important residues are highly conserved.

Binary Networks

In this work, we introduce a binary network to explore atomic-level structural information encoded by 44 inhibitor-accessible residues in drug binding pockets that may distinguish one kinase from all other human kinome structures. In the inhibitor-accessible geometric space, atoms of one residue may determine the interactions with inhibitors by forming hydrogen bonds, salt bridges, or hydrophobic/hydrophilic interactions, while the space between two residues may determine the geometric shape matching the inhibitors. Both are essential for determining kinase selectivity (Figure 3A). Residue-based comparisons emphasize the contributions of the residues to specific interactions but cannot fully reflect their effects on the geometric shape matching. To extract the vital structural information, here, we propose a description of drug pockets through binary networks, consisting of binary units (Figure 3B). The “binary units” are defined as two residues (residue pair) that may accommodate the same motif/region of the inhibitors in the kinase pocket local environment (Figure 3B). In the analysis, they are considered as “unique” if they are not found in any of the other kinases in the human kinome or “non-unique” if they are shared by other kinases. It is expected that upon drug binding, they will undergo a minor induced fit optimization of the interactions. However, since we used a heterogeneous ensemble, the already built-in flexibility alleviates this concern. Such a protocol has three advantages. (i) The binary unit contains structural information of not only two residues but also the geometric space between them. This may ensure that the comparisons of the binary networks across the 495 kinases can be sufficiently sensitive to detect local geometric differences in the inhibitor-accessible geometric space. (ii) Any larger geometric space can be projected from the basic units in the network. Thus, the complete structural information in the kinase drug binding pocket can be included in the established binary networks. (iii) The binary units can indicate the residue contacts between the two subdomains in the kinase pocket, providing a vehicle for incorporating structural information of the kinome conformational landscape (Figure 3C).

Figure 3.

Figure 3

Open in a new tab

Binary network for a kinase drug pocket. (A) A residue pair in the kinase drug pocket contains the structural information for specific interactions of two residues and the geometric space between the two residues for shape matching. (B) The binary network uses the residue pair as the basic unit (referred as binary unit). (C) Any large geometric space in the drug pocket can be projected from the binary unit in the network. A total of (D) 67 in-domain and (E) 29 inter-domain static binary units can be identified. The inhibitor-accessible residues in (D,E) are colored based on the subdomains.

The binary network of individual kinases was established based on the 44 inhibitor-accessible residues. If the side chains of two residues in the pockets are visually observed to be adjacent and able to approach each other to form the local area in the pocket for accommodating the inhibitors, they are considered as the “binary unit”, with no distance cutoff. Some binary units can be static and independent of the kinase conformations, while others are highly dynamics and vary across them. Examples of the static binary units include those in the β-sheets of the N-lobe, such as X3–X4, X3–X5, and X5–X6. They are in the same subregions of the kinase domain and exhibit minor changes in the relative positions. We identified a total of 67 such in-domain static binary units (Figure 3D). Binary units formed by two residues from different subregions can also be static since these regions in the kinase domain are generally rigid. Examples include the binary units formed between β1−β5 and the hinge region, between the αD helix and the ATP region, between β1−β5 and the αC−β4 loop, and between the ATP region and the αC−β4 loop. Overall, we identified 29 inter-domain static binary units (Figure 3E). The dynamic binary units will be discussed below.

Binary Networks in the Structural Landscape

Like other proteins, kinases are dynamic and can adopt multiple conformations in the structural landscape.45 It has been well-established that typical protein kinases follow the OUT-to-IN structural mechanism of activation.45 The αC-helix in the N-lobe and the A-loop in the C-lobe are essential components of the OFF-to-ON transition. The αC-helix is OUT in the OFF state, forming hydrophobic interactions with the collapsed A-loop. It becomes IN in the ON state, with the extended A-loop. The DFG motif in the A-loop is important for loading ATP. The Phe residue of DFG is IN in the ON state, while it is OUT in the OFF state. These conformational changes may greatly alter the dynamic inter-domain binary units in the network.

If two residues in the pocket can get close to each other in certain kinase conformations in the landscape, they are defined as dynamic binary units. For example, X-18 in the αC−β4 loop and X44 in the A-loop are close in the inactive conformation (DFG-out). Although this may not be observed in other conformations, X18–X44 can be defined as a dynamic binary unit since they have a potential to form the local environment to accommodate the inhibitors in the inactive conformation. To generate the dynamic inter-domain binary units in the network, we established the structural landscape of kinases from the crystal/cryo-EM structures and used unsupervised machine learning algorithms46,47 to extract structural information. The salt bridge between X6 in β3 and X11 in αC was used as the residue-based feature to characterize the αC-IN or αC-OUT conformations, and the distance between X17 in β4 and X44 in the A-loop was calculated to quantify the DFG-IN and DFG-OUT conformations.48 The k-means algorithm46,47 was used to perform the clustering analysis. To avoid bias from the hyper parameters, the elbow method was first used to determine the optimal cluster number (K) in the k-means analysis (Figure 4A). The sum of the squared distances of sampled conformations to the closest cluster center (inertia) decreases linearly after the cluster number reaches 4, which indicates that the kinases have four major conformational states in the structural landscape (Figure 4B). Representative structures from each cluster were then selected as the structural references to generate the dynamic binary units in the network (Figure 4C).

Figure 4.

Figure 4

Open in a new tab

Kinase structural landscape from kinase-drug complexes. (A) The elbow algorithm is used to determine the optimal cluster number in the clustering analysis. (B) The k-means analysis indicates four major states in the kinase structural landscape, which correspond to the (C) kinase conformations with αC-in/DFG-in, αC-out/DFG-out, αC-in/DFG-out, and αC-out/DFG-in.

The dynamic binary units in the kinase conformations exhibit notable differences (Figure 5A). Inspection of the structural landscape indicated that the four states cover all the combinations of αC-helix IN or OUT and DFG IN or OUT conformations and they are not highly isolated (Figure 4). This implies that the kinase domain is structurally flexible and can accommodate the inhibitors and the transition from one state to another without huge energy barriers. Thus, the generated static and dynamic binary units from different states in the landscape can be combined (Figure 5B), and the combined units in the binary network may represent feasible structural features that can be targeted by inhibitors for kinome-wide selectivity. Finally, a total of 151 binary units are defined in the binary network for individual 495 kinases (Table S4).

Figure 5.

Figure 5

Open in a new tab

Dynamic binary units in the binary network. (A) The dynamic binary units for four states in the structural landscape indicate notable differences. (B) The combination of static and dynamic binary units results in a total of 151 binary units in the binary network.

Binary Units for Kinase Selectivity

Iterative comparisons of the binary networks across 495 kinases were performed to extract the structural features that can distinguish one kinase from all others. The comparisons are based on binary units in the network, aiming at identifying the less shared binary units for individual kinases in the kinome (Table S4). The results are shown in Figure 6A. The different colors indicate the number of the kinases that share the searched binary unit. All kinases contain the binary units that are shared by less than seven other kinases (Figure 6B). 490 (∼99.0% of 495) kinases contain the units that are shared by less than four other kinases. 474 (∼95.7% of 495) kinases contain the units that are shared by less than two other kinases. 420 (∼84.8% of 495) kinases contain the binary units that are only shared by one other kinase. 331 (∼66.9% of 495) kinases contain the unique binary unit that only exist in it but not in all others.

Figure 6.

Figure 6

Open in a new tab

Binary units for kinase selectivity. (A) All 495 kinases contain the binary units that are shared by less than seven other kinases in human kinome. (B) 331 kinases (∼67.2%) show unique binary units that can distinguish it from all others in the 495 kinases. (C) The unique binary units are more located at β1, αC, β4−β5, and the hinge region in the kinase drug pocket. (D) Kinase families, such as RGC, TKL, TYR, and OTHER exhibit more unique binary units than others including AGC, CAMK, and CMGC. The kinase PDB ID can be found in Table S1. In (C), the weights of the binary units are indicated by the width of the blue lines.

The unique binary units in the 331 kinases distinguish a targeted kinase from all others. The local structural features represented by these unique network units can be the hot spot for the ultimate kinase selectivity. They are at the β1, αCbot, and αCtop regions, involving β1, β4−β5, the αC-helix, and the hinge region. They may fit type II and type III kinase inhibitors (Figure 6C). Kinase families including AGC, CAMK, and CMGC, are more structurally conserved and appear to have fewer unique binary units, while those including RGC, TKL, TYR, and OTHER show more. From the structural standpoint, this indicates that inhibitors targeting the RGC, TYL, TYR, and OTHER families may more readily achieve kinome-wide selectivity than other kinase families. Kinases that do not contain single unique binary units may require a combination of two or more binary units.

EGFR Tyrosine Kinase

The EGFR, a tyrosine kinase in the TYR family, provides one example to our protocol. Its activation triggers the MAPK and PI3K/AKT pathways, triggering cell proliferation, division, and growth. Its dysfunction has been implicated in cancers.49 Four drugs targeting EGFR have been approved by the US FDA.7 Ten are in clinical trials.8,9 185 EGFR kinase domain structures are available in the PDB database.

Our network for EGFR identifies binary units in the αD, ATP, αCbot, and αCtop regions that can be exploited in kinase inhibitor selectivity (Figure 7A,B). FDA-approved EGFR drugs including erlotinib, gefitinib, lapatinib, and osimertinib that belong to the type I or type I1/2 kinase inhibitors bind to these regions (Figure 7C). The binary unit of X30–X31 in the αD region corresponds to the Cys797–Asp800 in EGFR (Figure 7B). It contains a cysteine residue, Cys797, which has been the primary site for EGFR covalent inhibitors (PDB IDs: 4G5J, 4I24, and 2JIV).50 Lapatinib (1XKK) and gefitinib (2ITO) show extensive atomic contacts with this region as well. In the kinome, 11 kinases including MAP2K7, BLK, BTK, EGFR, ERBB2, ERBB4, ITK, JAK3, TEC, TXK, and BMX have a cysteine residue at the same position in the αD region. The binary unit, Cys797–Asp800, is only found in four other kinases, BLK, ERBB2, ITK, and JAK3. This indicates that in addition to covalently bonding to Cys797, atomic contacts with Asp800 and geometric shape matching with the space between Cys797 and Asp800 may benefit EGFR inhibitor selectivity.

Figure 7.

Figure 7

Open in a new tab

Binary network for EGFR tyrosine kinase. (A) The less shared binary units in the network of EGFR are located at ATP, αD, αCbot, and αCtop regions. The blue lines indicate the unique binary units that only exist in EGFR. (B) The unique binary units, Met766–Cys775, Cys775–Thr790, Cys775–Met793, and Cys775–The854, are at the gate areas. αD region shows the binary unit of Cys797–Asp800, in which Cys797 is the primary site for EGFR covalent inhibitors. (C) The FDA-approved EGFR drugs that belong to type I and type I1/2 kinase inhibitors fit to the binary units in the network. (D) The cysteine residue, Cys775, and the binary units of Cys775–Met793, and Cys775–Thr854 are barely visited by the existing EGFR inhibitors. (E) The example of an inhibitor-specific binary unit for EGFR includes the binary unit of Cys775–Cys797 for dual covalent bond inhibitors. It only exists in EGFR and not in all other typical protein kinases.

The unique binary units that only exist in EGFR are X14–X18, X18–X24, X18–X27, and X18–X42, which are Met766–Cys775, Cys775–Thr790, Cys775–Met793, and Cys775–Thr854 at the gate area of EGFR’s kinase domain (Figure 7B). These unique binary units represent structural elements in EGFR kinome-wide selectivity. They are mostly in the ATP pocket and are static and independent of the kinase conformations in the landscape. Cys775 is involved in four unique binary units. Many of the EGFR inhibitors in the PDB database and the FDA-approved drugs do not show highly specific interactions with this cysteine residue in the αC−β4 loop. The geometric space formed by the unique binary units of Cys775–Met793 and Cys775–Thr854 are also not highly visited (Figure 7D). Notably, type III EGFR inhibitors, such as EAI001 and EAI045, show a potential to interact with the identified binary units in EGFR.51,52 They are located at the aC region and establish interactions in this region (Figure S1). Several similar inhibitors have been designed.53,54 It would be reasonable to expect that tested designs optimized to accommodate these structural features at the EGFR gate area may show improved kinase selectivity.

AKT Serine/Threonine Kinase

In a second example, we show the binary network of AKT (protein kinase B), a serine/threonine kinase in the ACG family.55 AKT kinases are downstream effectors of PI3K in the PI3K/AKT pathway for cell survival, metabolism, and proliferation. They are among the most frequently mutated oncogenes in cancer. Given their essential roles in the PI3K/AKT pathway for cell proliferation and their frequent oncogenic mutations, AKT kinases are promising therapeutic targets in cancers.56 However, compared to EGFR, the development of AKT inhibitors has been less successful. To date, no AKT drug has been approved by the US FDA. Four are in clinical trial.8

AKT has three isoforms, AKT1, AKT2, and AKT3 with isoform-specific functionalities in cells, among which AKT1 dysfunctions have been reported as occurring more frequently in human cancers.57 The isoform-specific roles indicate that gain of isoform selectivity is essential for improving the therapeutic window of ATK inhibitors.58 The binary network suggests that AKT1 contains 10 unique units that are not found in any other kinases and five binary units that are only shared by one or two other kinases (Figure 8A). These units in AKT1 are mostly located at the ATP region, the αC region, and the allosteric pocket, formed between the αC-helix and β3−β5 and between the αC-helix and the αE region (Figure 8B). They are dynamic and can be considerably affected by kinase domain conformations. Analysis of the structural landscape suggests that the availability of these binary units to inhibitors would be greater when the kinase adopts the cluster 2 conformations with αC-OUT and DFG-OUT, that is, the inactive state. This indicates that the type II drug scaffold, which is more accessible to the αCbot region and the allosteric pocket in the kinase domain, may be promising AKT1 inhibitors (Figure 8C). Clinically evaluated AKT inhibitors, including afuresertib, capivasertib, ipatasertib, and uprosertib, and AKT inhibitors in development (PDB IDs 3O96, 4EJN, and 5KCV) belong to either type I or type III (allosteric) kinase inhibitors (Figure 8D). Indeed, these inhibitors are mostly pan-inhibitors, lacking isoform specificity.58,59 In the three AKT isoforms, the binary units around X12 (Asn199 in AKT1, Ser201 in AKT2, and Ser197 in AKT3) and X16 (Ser205 in AKT1, Thr207 in AKT2, and Thr203 in AKT3) show differences that could be structural targets for amplifying the AKT1 inhibitor isoform selectivity versus ATK2 and AKT3 (Figure 8E–G).

Figure 8.

Figure 8

Open in a new tab

Binary network for AKT serine/threonine kinases. (A,B) The less shared binary units in the network of AKT1 are located at ATP, αC, and allosteric regions. The blue lines indicate the unique binary units that only exist in AKT1. The green lines indicate the binary units that are only shared by one or two other kinases. (C) The binary units are more accessible to the type II kinase inhibitors when the kinase domain adopts the conformation with αC-out and DFG-out. (D) The clinically evaluated AKT1 inhibitors that belong to type I kinase inhibitors and the allosteric AKT1 inhibitors that belong to type III kinase inhibitors have not formed the specific interactions with these binary units. (E–G) The residues in αC and allosteric regions may serve as the structural targets for improving the isoform specificity of AKT1 inhibitors.

KDS—A Cross-Platform Software

To facilitate a customized visualization and analysis of the binary networks, we developed a cross-platform software, KDS. KDS is based on an interactive algorithm and the binary networks proposed in this work. It uses a real-time 3D engine for interactively visualizing the binary networks in the kinase pockets and the user interface (UI) system to present the related information in the binary networks. The technical description of the KDS software can be found in the Methods and Materials section and the Supporting Information.

Figure 9 illustrates the main scenes in KDS, including a splash screen, a kinase scene, and a network scene. The splash scene provides the KDS’s basic information and initializes the network database at the backend (Figure S2). The kinase scene is designed for visualizing and analyzing the binary networks of the kinases in the human kinome. It has two panels, the selection panel on the left side and the visualization panel on the right side (Figure 9A). The kinase of interest can be selected by either clicking on the provided buttons or searching with the key words in the selection panel, after which the related information of the binary network for the selected kinase would be shown in the visualization panel (Figure S3). Kinases are frequently mutated in cancer. The network scene in KDS provides an interface for generating the customized binary networks in various biological contexts (Figures 9B and S4). The mutations can be introduced to the binary networks through the mutation panel and are summarized in the network panel. The KDS software for Windows and MacOS platforms is freely available at https://github.com/CBIIT/KDS.

Figure 9.

Figure 9

Open in a new tab

Snapshots of KDS software. KDS contains (A) kinase scene and (B) network scene. The kinase scene achieves the real-time visualization of the binary networks for 495 kinases in the kinome. The network scene provides the interface to generate the customized binary networks by introducing the mutations.

Discussion and Conclusions

In this work, we perform a structural bioinformatic analysis of 495 human kinases in 10 typical kinase families, identifying the inhibitor-accessible structural features that can distinguish one kinase from all others in the human kinome. These atomic-level features in the drug pockets may serve as the basis for structure-based kinase drug design/optimization. We integrate the existing crystal/cryo-EM kinase structures in the PDB database and the high-quality structural predictions by AlphaFold234,35 and propose a novel binary network protocol to extract atomic-resolution structural information from the dynamic kinase drug pockets. By exploring the structural landscape, we established and interactively compared the binary network with 151 binary units, achieving coverage of the complete kinome structural information with remarkable sensitivity. The results show that all 495 kinases contain binary units that are shared by less than seven other kinases. 331 kinases have at least a single unique binary unit that may distinguish it from all other kinases. We suggest that these can be potential targets in kinome-wide selective drug development. We expect that the proposed binary network in combination with the unsupervised clustering used in this work can serve as a structural protocol to extract the distinguishing structural features for any protein from their respective families.

Our analysis obtained 44 structurally conserved inhibitor-accessible residues in the accessible geometric space of the kinase pockets, which may contribute to kinase inhibitor selectivity. We also noticed that some kinases exhibit regional variations in the drug pockets, and these local variations can be unique and contribute to the selectivity (Figure S5). These include the hinge region (X29 to X30), the N-terminal loop of the αC-helix (X16 to X17), and the C-terminal loop of the αE helix (X35 to X36). The hinge region in most kinases has five residues. However, many kinases show a hinge region of three, four, or six residues. In STK31 in the CAMK family, the hinge region even consists of 19 residues. Similarly, the αC-helix at the N-terminal loop normally has five residues. In some kinases, including VRK1, VRK2, and VRK3 in the CK1 family and PINK1 in the OTHER family, the αC loop has more than 20 residues. The αE C-terminal loop from the HRD motif mostly has seven residues. However, in some kinases, including MKNK1 in the CAMK family and activin receptors in the TKL family, the αE loop is long, with more than 15 residues. These, shorter or longer, hinge regions and loops alter the local morphology in the drug pocket while preserving the potential for direct selective drug interactions. Residues in the A-loop after the DFG motif can also establish interactions with type II and III kinase inhibitors.

In the comparisons, pseudokinases are not separated from the 495 kinases. This is due to two reasons. First, despite the lack of catalytic activity, pseudokinases can be pharmaceutical targets.60 Pseudokinases, such as the kinase domains in JAK kinases and KSR,61−63 are frequently involved in kinase dimerization/oligomerization, allosterically promoting full activation of another kinase, thereby stimulating signaling. Numerous inhibitors have been designed to target pseudokinases.64 Second, the definition/identification of pseudokinases has not been conclusive. A pseudokinase is typically defined as a kinase with the kinase domain lacking evolutionally conserved functional residues. However, this is not always the case. For example, CASK lacking the Asp of the DFG motif in the A-loop has Mg2+-independent activity.65

The binary network contains a total of 151 binary units from the kinase structural landscape. These units represent conserved local features in the drug pockets that may confer inhibitor selectivity. The binary network can also be extended to include inhibitor-specific binary units with the two residues further apart. For example, in EGFR, two cysteine residues, Cys797 in the αD region and Cys775 in the αC−β4 loop, can be combined as an inhibitor-specific unit in the binary network for developing bivalent covalent EGFR inhibitors (Figure 7E). This Cys775–Cys797 binary unit only exists in EGFR, not in any other kinases in the entire human kinome, which implies that inhibitors that can form two covalent bonds with the cysteine residues at both ends may achieve EGFR kinome selectivity.

Collectively, our kinome-wide structural bioinformatic study constructed the inhibitor-accessible geometric space in the drug pockets from 6709 crystal/cryo-EM kinase structures, related to 287 kinases. The geometric space obtained from the experimental kinase–ligand complexes constitutes our structural kinase landscape. We identified 44 inhibitor-accessible residues in the drug pockets within these. We then used the high-quality predictions from AlphaFold234,35 to predict the structures of all (495) kinases, obtaining the structural kinome, and validated the 44 inhibitor-accessible residues on this inclusive kinase set. To extract the structural information encoded by the 44 inhibitor-accessible residues, we constructed a binary network that incorporates the structural features of the landscape. The results identify the unique binary units that may distinguish one kinase from all others. They do not provide a complete solution to kinase drug selectivity. The binary units do not account for the full energy landscape of individual kinases that cannot be obtained from crystal/cryo-EM structural snapshots, or from the rigid AI-based predictions, especially for the flexible pockets.66 The affinity and specificity for case-based designs are also missing. They can however be useful in in silico drug repurposing and altogether offer a step forward.

In this work, we focus on wild-type kinases, which may help in countering aberrant proliferation from overexpression in cancer and other diseases, including neurodevelopmental disorders.67,68 Kinases are however frequently mutated.69 Kinase domains can harbor single, or frequently double, or multiple mutations.70,71 Mutations occur in multiple kinases in cancer cells, in the same or different pathways.72,73 Since same-protein mutations perturb the conformations, they may considerably alter the binary networks and the corresponding comparison profiles. In our KDS software, we provided the interface for introducing mutations into the binary networks in the kinome.

To conclude, here, we described an innovative kinase selective protocol by identifying inhibitor-accessible local features in the drug pockets in one kinase but not in all others across the structural human kinome. We represent the structural features by binary units in binary networks and suggest that unique binary units can serve as the basis for structure-based design/optimization of highly selective inhibitors. As examples, we apply the protocol to a tyrosine and serine/threonine kinases. In the latter (AKT) case, the protocol successfully distinguishes among isoforms. Improved kinase selectivity is the holy grail of drug discovery to ease toxicity. Nonetheless, it will not overcome drug resistance, which will arise including from proliferating, rare cell populations with pre-existing insensitive mutations that were not captured in the networks.74 Single cell technologies and combinatorial drug regimens targeting the mutational variants could extend cancer-free time scales while avoiding toxicity. Distinct mutation profiles can make the generated binary network mutation specific. On the plus side, this can benefit precision medicine. Generating comprehensive binary networks, for diverse kinase mutations in clinical settings, is an important aim, albeit here hampered by the unavailability of the conformationally heterogeneous landscape.

Methods and Materials

495 Kinase Domains

The kinase families were obtained from the Uniprot database.75 We collected the gene name, uniport ID, kinase family, and structures of 495 kinase domains. A total of 4913 experimental crystal/cryo-EM structures were obtained from the PDB database. Based on the kinase domain sequences from Uniprot, we confirmed that 287 kinases with crystal/cryo-EM structures contained catalytic kinase domains. The 4913 structures contained 7439 chains of kinase domains. Within the 4913 kinase domain structures, the ligands were found in 4554. Among these 4554, structures with multiple chains of kinase domains were split into the 6907 kinase–ligand structures. The PDB structures and the ligand names are listed in Table S2.

∼42.0% (208) of the kinases in the human kinome do not have experimental domain structures. To perform the kinome-wide structural bioinformatic analysis, AlphaFold234,35 was used to construct the structural kinome. The structures were predicted for 483 kinases, 13 of which, including RPS6KA1, RPS6KA2, RPS6KA3, RPS6KA4, RPS6KA5, RPS6KA6, SPEG, JAK1, JAK2, JAK3, OBSCN, TYK2, and EIF2AK4, had two kinase domains. The kinase domain in PLK5 was truncated. OBSCN had two kinase domains, in which the kinase domain at the N-terminal (residues 6468–6721) cannot be obtained from AlphaFold2,34,35 which implied that it may not follow the typical kinase structure. These two kinase domains were not included in the analysis. Finally, a total of 495 kinase domain structures were generated and analyzed (Table S1).

Inhibitor-Accessible Residues

The inhibitor-accessible geometric space in the kinase drug binding pockets were generated from the kinase–ligand complexes. Based on these, 44 inhibitor-accessible residues were confirmed. These structurally conserved residues were validated from the structural kinome (Table S2). Notably, some kinases showed some structural variations in β1, β4, αC-helix, and hinge regions. These regions are more exposed to the solvent and can be impacted by crystal packing. Similar disruptions were also observed in the experimental structures in the PDB, such as the αC-helix in PDB IDs 1MRV, 1GZN, and 2JDO for AKT2. Thus, it has been unclear whether these variations were from the predictions or the inherent properties of these kinases. These features were preserved in the 44 inhibitor-accessible residues. The missing conserved residues in the β1 region are labeled as “O” and the potential affected residues are labeled with “*” in Table S3.

Binary Network

Our premise is that the binary network incorporates the structural information of the kinase drug binding pockets. A geometrically adjacent residue pair with side chains orienting toward interactions with inhibitors was used as the basic unit (referred as the binary unit) in the network. The structures from AlphaFold234,35 were aligned as the structural basis for constructing the static basic units. To define the dynamic binary units in the network, the kinase structural landscape was established from the clustered crystal/cryo-EM structures as described below. To eliminate the noise in the analysis, the crystal/cryo-EM structures were screened based on two criteria: (i) the complete densities for the 44 inhibitor-accessible residues. Structures with sequence labeling errors were ignored. (ii) The structures with the 44 inhibitors had no mutations. A total of 5793 crystal/cryo-EM structures were used in the analysis (Table S2). The k-means algorithm46,47 was used to conduct the clustering analysis for the structural landscape. Structures from each cluster of the landscape were selected to build the dynamic binary units. A total of 151 binary units were finally defined in the binary network (Table S3).

Based on the identified 44 inhibitor-accessible residues for individual kinases, the binary networks were established for 495 kinase domains. Interactive comparisons of the 495 binary networks were performed to search for the less shared binary units for individual kinases in the human kinome. Each binary unit in the network was compared with the corresponding ones in the other 494 networks. The searched binary units were considered as “different” if any of two residues in two compared binary units mismatched. This can ensure that the binary-unit-based comparisons reflect the geometric contributions from individual residues to drug binding. A summary of the binary units and the corresponding comparison profiles can be found in Table S4.

KDS Software

KDS software was developed using the Unity (Unity 2021.3.8f1). The 3D assets in the software were generated by Blender 3.1 (python 3.10.2) and PyMol. The binary networks for 495 typical kinases in 10 families were embedded. The post-process layer was used to produce the bloom volume effects. The asynchronous programming model was used to initialize and update the binary networks, while other functions were generally achieved in the main thread. The default username and password were “KDS” in the login panel of the splash screen. In the kinase scene, the representative kinase structure and the binary network in the visualization panel were generated based on the kinase ATK1. The 3D structure can be controlled through the mouse clicks. The buttons for individual binary units were colored based on the number of the kinases sharing the searched binary units. In the network scene, the amino acids for 44 inhibitor-accessible residues for the selected kinase were shown as a one-letter code. Additional details for the UI system and 3D visualization panels can be found in Figures S2–S4. The KDS software for Windows and MacOS platforms can be downloaded from https://github.com/CBIIT/KDS.

Acknowledgments

The calculations had been performed using the high-performance computational facilities of the Biowulf PC/Linux cluster at the National Institutes of Health, Bethesda, MD (https://hpc.nih.gov/).

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jctc.2c01171.

  • Analysis of the type III inhibitor in the EGFR pocket; snapshots of the KDS software; regional variations in 495 kinases (PDF)

  • Summary of 495 kinases and their experimental structures (XLSX)

  • Summary of 4913 experimental kinase domain structures (XLSX)

  • Summary of 44 inhibitor-accessible residues for 495 kinases (XLSX)

  • Summary of 74,745 binary units in the binary networks of 495 kinases (XLSX)

Author Contributions

M.Z.: conceptualization, investigation, formal analysis, software development, and writing—original draft. Y.L.: writing—review and editing. H.J.: conceptualization and writing—review and editing. R.N.: conceptualization, supervision, and writing—review and editing.

This project has been funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health, under contract HHSN261201500003I. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does the mention of trade names, commercial products, or organizations imply endorsement by the US Government. This research was supported [in part] by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research.

The authors declare no competing financial interest.

Supplementary Material

ct2c01171_si_001.pdf (643.1KB, pdf)
ct2c01171_si_002.xlsx (27.2KB, xlsx)
ct2c01171_si_003.xlsx (232.1KB, xlsx)
ct2c01171_si_004.xlsx (170.9KB, xlsx)
ct2c01171_si_005.xlsx (1.2MB, xlsx)

References

  1. Cohen P.; Alessi D. R. Kinase Drug Discovery - What’s Next in the Field?. ACS Chem. Biol. 2013, 8, 96–104. 10.1021/cb300610s. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Klaeger S.; Heinzlmeir S.; Wilhelm M.; Polzer H.; Vick B.; Koenig P. A.; Reinecke M.; Ruprecht B.; Petzoldt S.; Meng C.; Zecha J.; Reiter K.; Qiao H.; Helm D.; Koch H.; Schoof M.; Canevari G.; Casale E.; Depaolini S. R.; Feuchtinger A.; Wu Z.; Schmidt T.; Rueckert L.; Becker W.; Huenges J.; Garz A. K.; Gohlke B. O.; Zolg D. P.; Kayser G.; Vooder T.; Preissner R.; Hahne H.; Tõnisson N.; Kramer K.; Götze K.; Bassermann F.; Schlegl J.; Ehrlich H. C.; Aiche S.; Walch A.; Greif P. A.; Schneider S.; Felder E. R.; Ruland J.; Médard G.; Jeremias I.; Spiekermann K.; Kuster B. The target landscape of clinical kinase drugs. Science 2017, 358, eaan4368 10.1126/science.aan4368. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Cohen P.; Cross D.; Jänne P. A. Kinase drug discovery 20 years after imatinib: progress and future directions. Nat. Rev. Drug Discov. 2021, 20, 551–569. 10.1038/s41573-021-00195-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Attwood M. M.; Fabbro D.; Sokolov A. V.; Knapp S.; Schiöth H. B. Trends in kinase drug discovery: targets, indications and inhibitor design. Nat. Rev. Drug Discov. 2021, 20, 839–861. 10.1038/s41573-021-00252-y. [DOI] [PubMed] [Google Scholar]
  5. Modi V.; Dunbrack R. L. Jr. A Structurally-Validated Multiple Sequence Alignment of 497 Human Protein Kinase Domains. Sci. Rep. 2019, 9, 19790. 10.1038/s41598-019-56499-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Laufer S.; Bajorath J. New Horizons in Drug Discovery - Understanding and Advancing Different Types of Kinase Inhibitors: Seven Years in Kinase Inhibitor Research with Impressive Achievements and New Future Prospects. J. Med. Chem. 2022, 65, 891–892. 10.1021/acs.jmedchem.1c02126. [DOI] [PubMed] [Google Scholar]
  7. Roskoski R. Jr. Properties of FDA-approved small molecule protein kinase inhibitors: A 2020 update. Pharmacol. Res. 2020, 152, 104609. 10.1016/j.phrs.2019.104609. [DOI] [PubMed] [Google Scholar]
  8. Carles F.; Bourg S.; Meyer C.; Bonnet P. PKIDB: A Curated, Annotated and Updated Database of Protein Kinase Inhibitors in Clinical Trials. Molecules 2018, 23, 908. 10.3390/molecules23040908. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Bournez C.; Carles F.; Peyrat G.; Aci-Sèche S.; Bourg S.; Meyer C.; Bonnet P. Comparative Assessment of Protein Kinase Inhibitors in Public Databases and in PKIDB. Molecules 2020, 25, 3226. 10.3390/molecules25143226. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Roskoski R. Jr. Orally effective FDA-approved protein kinase targeted covalent inhibitors (TCIs). Pharmacol. Res. 2021, 165, 105422. 10.1016/j.phrs.2021.105422. [DOI] [PubMed] [Google Scholar]
  11. Fujita K. I.; Ishida H.; Kubota Y.; Sasaki Y. Toxicities of Receptor Tyrosine Kinase Inhibitors in Cancer Pharmacotherapy: Management with Clinical Pharmacology. Curr. Drug Metab. 2017, 18, 186–198. 10.2174/1389200218666170105165832. [DOI] [PubMed] [Google Scholar]
  12. Esposito A.; Viale G.; Curigliano G. Safety, Tolerability, and Management of Toxic Effects of Phosphatidylinositol 3-Kinase Inhibitor Treatment in Patients With Cancer. JAMA Oncol. 2019, 5, 1347–1354. 10.1001/jamaoncol.2019.0034. [DOI] [PubMed] [Google Scholar]
  13. Manning G.; Whyte D. B.; Martinez R.; Hunter T.; Sudarsanam S. The protein kinase complement of the human genome. Science 2002, 298, 1912–1934. 10.1126/science.1075762. [DOI] [PubMed] [Google Scholar]
  14. Kanev G. K.; de Graaf C.; de Esch I. J. P.; Leurs R.; Würdinger T.; Westerman B. A.; Kooistra A. J. The Landscape of Atypical and Eukaryotic Protein Kinases. Trends Pharmacol. Sci. 2019, 40, 818–832. 10.1016/j.tips.2019.09.002. [DOI] [PubMed] [Google Scholar]
  15. Nussinov R.; Jang H.; Gursoy A.; Keskin O.; Gaponenko V. Inhibition of Nonfunctional Ras. Cell Chem. Biol. 2021, 28, 121–133. 10.1016/j.chembiol.2020.12.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Roskoski R. Jr. Classification of small molecule protein kinase inhibitors based upon the structures of their drug-enzyme complexes. Pharmacol. Res. 2016, 103, 26–48. 10.1016/j.phrs.2015.10.021. [DOI] [PubMed] [Google Scholar]
  17. Teng M.; Luskin M. R.; Cowan-Jacob S. W.; Ding Q.; Fabbro D.; Gray N. S. The Dawn of Allosteric BCR-ABL1 Drugs: From a Phenotypic Screening Hit to an Approved Drug. J. Med. Chem. 2022, 65, 7581–7594. 10.1021/acs.jmedchem.2c00373. [DOI] [PubMed] [Google Scholar]
  18. Hantschel O. Unexpected off-targets and paradoxical pathway activation by kinase inhibitors. ACS Chem. Biol. 2015, 10, 234–245. 10.1021/cb500886n. [DOI] [PubMed] [Google Scholar]
  19. Bello T.; Gujral T. S. KInhibition: A Kinase Inhibitor Selection Portal. iScience 2018, 8, 49–53. 10.1016/j.isci.2018.09.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Lin A.; Giuliano C. J.; Palladino A.; John K. M.; Abramowicz C.; Yuan M. L.; Sausville E. L.; Lukow D. A.; Liu L.; Chait A. R.; Galluzzo Z. C.; Tucker C.; Sheltzer J. M. Off-target toxicity is a common mechanism of action of cancer drugs undergoing clinical trials. Sci. Transl. Med. 2019, 11, eaaw8412 10.1126/scitranslmed.aaw8412. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Bajusz D.; Ferenczy G. G.; Keseru G. M. Structure-based Virtual Screening Approaches in Kinase-directed Drug Discovery. Curr. Top. Med. Chem. 2017, 17, 2235–2259. 10.2174/1568026617666170224121313. [DOI] [PubMed] [Google Scholar]
  22. Vulpetti A.; Bosotti R. Sequence and structural analysis of kinase ATP pocket residues. Farmaco 2004, 59, 759–765. 10.1016/j.farmac.2004.05.010. [DOI] [PubMed] [Google Scholar]
  23. Sturm N.; Tinivella A.; Rastelli G. Exploration and Comparison of the Geometrical and Physicochemical Properties of an αC Allosteric Pocket in the Structural Kinome. J. Chem. Inf. Model. 2018, 58, 1094–1103. 10.1021/acs.jcim.7b00735. [DOI] [PubMed] [Google Scholar]
  24. Vijayan R. S.; He P.; Modi V.; Duong-Ly K. C.; Ma H.; Peterson J. R.; Dunbrack R. L. Jr.; Levy R. M. Conformational analysis of the DFG-out kinase motif and biochemical profiling of structurally validated type II inhibitors. J. Med. Chem. 2015, 58, 466–479. 10.1021/jm501603h. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Davis M. I.; Hunt J. P.; Herrgard S.; Ciceri P.; Wodicka L. M.; Pallares G.; Hocker M.; Treiber D. K.; Zarrinkar P. P. Comprehensive analysis of kinase inhibitor selectivity. Nat. Biotechnol. 2011, 29, 1046–1051. 10.1038/nbt.1990. [DOI] [PubMed] [Google Scholar]
  26. Karaman M. W.; Herrgard S.; Treiber D. K.; Gallant P.; Atteridge C. E.; Campbell B. T.; Chan K. W.; Ciceri P.; Davis M. I.; Edeen P. T.; Faraoni R.; Floyd M.; Hunt J. P.; Lockhart D. J.; Milanov Z. V.; Morrison M. J.; Pallares G.; Patel H. K.; Pritchard S.; Wodicka L. M.; Zarrinkar P. P. A quantitative analysis of kinase inhibitor selectivity. Nat. Biotechnol. 2008, 26, 127–132. 10.1038/nbt1358. [DOI] [PubMed] [Google Scholar]
  27. Smyth L. A.; Collins I. Measuring and interpreting the selectivity of protein kinase inhibitors. J. Biol. Chem. 2009, 2, 131–151. 10.1007/s12154-009-0023-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Knight Z. A.; Shokat K. M. Features of selective kinase inhibitors. Chem. Biol. 2005, 12, 621–637. 10.1016/j.chembiol.2005.04.011. [DOI] [PubMed] [Google Scholar]
  29. Wang H.; Qiu J.; Liu H.; Xu Y.; Jia Y.; Zhao Y. HKPocket: human kinase pocket database for drug design. BMC Bioinf. 2019, 20, 617. 10.1186/s12859-019-3254-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Kanev G. K.; de Graaf C.; Westerman B. A.; de Esch I. J. P.; Kooistra A. J. KLIFS: an overhaul after the first 5 years of supporting kinase research. Nucleic Acids Res. 2021, 49, D562–D569. 10.1093/nar/gkaa895. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Zhang C.; Lopez M. S.; Dar A. C.; LaDow E.; Finkbeiner S.; Yun C. H.; Eck M. J.; Shokat K. M. Structure-guided inhibitor design expands the scope of analog-sensitive kinase technology. ACS Chem. Biol. 2013, 8, 1931–1938. 10.1021/cb400376p. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Modi V.; Dunbrack R. L. Jr. Defining a new nomenclature for the structures of active and inactive kinases. Proc. Natl. Acad. Sci. U.S.A. 2019, 116, 6818–6827. 10.1073/pnas.1814279116. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Modi V.; Dunbrack R. L. Kincore: a web resource for structural classification of protein kinases and their inhibitors. Nucleic Acids Res. 2022, 50, D654–D664. 10.1093/nar/gkab920. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Jumper J.; Evans R.; Pritzel A.; Green T.; Figurnov M.; Ronneberger O.; Tunyasuvunakool K.; Bates R.; Žídek A.; Potapenko A.; Bridgland A.; Meyer C.; Kohl S. A. A.; Ballard A. J.; Cowie A.; Romera-Paredes B.; Nikolov S.; Jain R.; Adler J.; Back T.; Petersen S.; Reiman D.; Clancy E.; Zielinski M.; Steinegger M.; Pacholska M.; Berghammer T.; Bodenstein S.; Silver D.; Vinyals O.; Senior A. W.; Kavukcuoglu K.; Kohli P.; Hassabis D. Highly accurate protein structure prediction with AlphaFold. Nature 2021, 596, 583–589. 10.1038/s41586-021-03819-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Varadi M.; Anyango S.; Deshpande M.; Nair S.; Natassia C.; Yordanova G.; Yuan D.; Stroe O.; Wood G.; Laydon A.; Žídek A.; Green T.; Tunyasuvunakool K.; Petersen S.; Jumper J.; Clancy E.; Green R.; Vora A.; Lutfi M.; Figurnov M.; Cowie A.; Hobbs N.; Kohli P.; Kleywegt G.; Birney E.; Hassabis D.; Velankar S. AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Res. 2022, 50, D439–D444. 10.1093/nar/gkab1061. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Nussinov R.; Wolfson H. J. Efficient detection of three-dimensional structural motifs in biological macromolecules by computer vision techniques. Proc. Natl. Acad. Sci. U.S.A. 1991, 88, 10495–10499. 10.1073/pnas.88.23.10495. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Fischer D.; Norel R.; Wolfson H.; Nussinov R. Surface motifs by a computer vision technique: searches, detection, and implications for protein-ligand recognition. Proteins 1993, 16, 278–292. 10.1002/prot.340160306. [DOI] [PubMed] [Google Scholar]
  38. Zhang M.; Jang H.; Nussinov R. The mechanism of PI3Kα activation at the atomic level. Chem. Sci. 2019, 10, 3671–3680. 10.1039/c8sc04498h. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Zhang M.; Jang H.; Nussinov R. Structural Features that Distinguish Inactive and Active PI3K Lipid Kinases. J. Mol. Biol. 2020, 432, 5849–5859. 10.1016/j.jmb.2020.09.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Zhang M.; Jang H.; Nussinov R. PI3K inhibitors: review and new strategies. Chem. Sci. 2020, 11, 5855–5865. 10.1039/d0sc01676d. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Kornev A. P.; Haste N. M.; Taylor S. S.; Ten Eyck L. F. Surface comparison of active and inactive protein kinases identifies a conserved activation mechanism. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 17783–17788. 10.1073/pnas.0607656103. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Schiemer J.; Horst R.; Meng Y.; Montgomery J. I.; Xu Y.; Feng X.; Borzilleri K.; Uccello D. P.; Leverett C.; Brown S.; Che Y.; Brown M. F.; Hayward M. M.; Gilbert A. M.; Noe M. C.; Calabrese M. F. Snapshots and ensembles of BTK and cIAP1 protein degrader ternary complexes. Nat. Chem. Biol. 2021, 17, 152–160. 10.1038/s41589-020-00686-2. [DOI] [PubMed] [Google Scholar]
  43. Posternak G.; Tang X.; Maisonneuve P.; Jin T.; Lavoie H.; Daou S.; Orlicky S.; Goullet de Rugy T.; Caldwell L.; Chan K.; Aman A.; Prakesch M.; Poda G.; Mader P.; Wong C.; Maier S.; Kitaygorodsky J.; Larsen B.; Colwill K.; Yin Z.; Ceccarelli D. F.; Batey R. A.; Taipale M.; Kurinov I.; Uehling D.; Wrana J.; Durocher D.; Gingras A. C.; Al-Awar R.; Therrien M.; Sicheri F. Functional characterization of a PROTAC directed against BRAF mutant V600E. Nat. Chem. Biol. 2020, 16, 1170–1178. 10.1038/s41589-020-0609-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Kornev A. P.; Taylor S. S.; Ten Eyck L. F. A helix scaffold for the assembly of active protein kinases. Proc. Natl. Acad. Sci. U.S.A. 2008, 105, 14377–14382. 10.1073/pnas.0807988105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. McClendon C. L.; Kornev A. P.; Gilson M. K.; Taylor S. S. Dynamic architecture of a protein kinase. Proc. Natl. Acad. Sci. U.S.A. 2014, 111, E4623 10.1073/pnas.1418402111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Pedregosa F.; Varoquaux G.; Gramfort A.; Michel V.; Thirion B.; Grisel O.; Blondel M.; Prettenhofer P.; Weiss R.; Dubourg V.; Vanderplas J.; Passos A.; Cournapeau D.; Brucher M.; Perrot M.; Duchesnay E. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 2011, 12, 2825–2830. [Google Scholar]
  47. Lloyd S. P. Least squares quantization in PCM. IEEE Trans. Inf. Theor. 1982, 28, 129–137. 10.1109/tit.1982.1056489. [DOI] [Google Scholar]
  48. Zhang M.; Maloney R.; Jang H.; Nussinov R. The mechanism of Raf activation through dimerization. Chem. Sci. 2021, 12, 15609–15619. 10.1039/d1sc03444h. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Roskoski R. Jr. Small molecule inhibitors targeting the EGFR/ErbB family of protein-tyrosine kinases in human cancers. Pharmacol. Res. 2019, 139, 395–411. 10.1016/j.phrs.2018.11.014. [DOI] [PubMed] [Google Scholar]
  50. Schwartz P. A.; Kuzmic P.; Solowiej J.; Bergqvist S.; Bolanos B.; Almaden C.; Nagata A.; Ryan K.; Feng J.; Dalvie D.; Kath J. C.; Xu M.; Wani R.; Murray B. W. Covalent EGFR inhibitor analysis reveals importance of reversible interactions to potency and mechanisms of drug resistance. Proc. Natl. Acad. Sci. U.S.A. 2014, 111, 173–178. 10.1073/pnas.1313733111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Niggenaber J.; Heyden L.; Grabe T.; Müller M. P.; Lategahn J.; Rauh D. Complex Crystal Structures of EGFR with Third-Generation Kinase Inhibitors and Simultaneously Bound Allosteric Ligands. ACS Med. Chem. Lett. 2020, 11, 2484–2490. 10.1021/acsmedchemlett.0c00472. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Zhao P.; Yao M. Y.; Zhu S. J.; Chen J. Y.; Yun C. H. Crystal structure of EGFR T790M/C797S/V948R in complex with EAI045. Biochem. Biophys. Res. Commun. 2018, 502, 332–337. 10.1016/j.bbrc.2018.05.154. [DOI] [PubMed] [Google Scholar]
  53. Caporuscio F.; Tinivella A.; Restelli V.; Semrau M. S.; Pinzi L.; Storici P.; Broggini M.; Rastelli G. Identification of small-molecule EGFR allosteric inhibitors by high-throughput docking. Future Med. Chem. 2018, 10, 1545–1553. 10.4155/fmc-2018-0063. [DOI] [PubMed] [Google Scholar]
  54. Foschi F.; Tinivella A.; Crippa V.; Pinzi L.; Mologni L.; Passarella D.; Rastelli G. Structure-activity exploration of a small-molecule allosteric inhibitor of T790M/L858R double mutant EGFR. J. Enzyme Inhib. Med. Chem. 2023, 38, 239–245. 10.1080/14756366.2022.2145284. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Manning B. D.; Toker A. AKT/PKB Signaling: Navigating the Network. Cell 2017, 169, 381–405. 10.1016/j.cell.2017.04.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Wang Q.; Chen X.; Hay N. Akt as a target for cancer therapy: more is not always better (lessons from studies in mice). Br. J. Cancer 2017, 117, 159–163. 10.1038/bjc.2017.153. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Shariati M.; Meric-Bernstam F. Targeting AKT for cancer therapy. Expet Opin. Invest. Drugs 2019, 28, 977–988. 10.1080/13543784.2019.1676726. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Quambusch L.; Landel I.; Depta L.; Weisner J.; Uhlenbrock N.; Müller M. P.; Glanemann F.; Althoff K.; Siveke J. T.; Rauh D. Covalent-Allosteric Inhibitors to Achieve Akt Isoform-Selectivity. Angew. Chem., Int. Ed. Engl. 2019, 58, 18823–18829. 10.1002/anie.201909857. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Nitulescu G. M.; Margina D.; Juzenas P.; Peng Q.; Olaru O. T.; Saloustros E.; Fenga C.; Spandidos D.; Libra M.; Tsatsakis A. M. Akt inhibitors in cancer treatment: The long journey from drug discovery to clinical use (Review). Int. J. Oncol. 2016, 48, 869–885. 10.3892/ijo.2015.3306. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Kung J. E.; Jura N. Prospects for pharmacological targeting of pseudokinases. Nat. Rev. Drug Discov. 2019, 18, 501–526. 10.1038/s41573-019-0018-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Babon J. J.; Lucet I. S.; Murphy J. M.; Nicola N. A.; Varghese L. N. The molecular regulation of Janus kinase (JAK) activation. Biochem. J. 2014, 462, 1–13. 10.1042/bj20140712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Lavoie H.; Sahmi M.; Maisonneuve P.; Marullo S. A.; Thevakumaran N.; Jin T.; Kurinov I.; Sicheri F.; Therrien M. MEK drives BRAF activation through allosteric control of KSR proteins. Nature 2018, 554, 549–553. 10.1038/nature25478. [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Maloney R. C.; Zhang M.; Liu Y.; Jang H.; Nussinov R. The mechanism of activation of MEK1 by B-Raf and KSR1. Cell. Mol. Life Sci. 2022, 79, 281. 10.1007/s00018-022-04296-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Byrne D. P.; Foulkes D. M.; Eyers P. A. Pseudokinases: update on their functions and evaluation as new drug targets. Future Med. Chem. 2017, 9, 245–265. 10.4155/fmc-2016-0207. [DOI] [PubMed] [Google Scholar]
  65. Mukherjee K.; Sharma M.; Urlaub H.; Bourenkov G. P.; Jahn R.; Südhof T. C.; Wahl M. C. CASK Functions as a Mg2+-independent neurexin kinase. Cell 2008, 133, 328–339. 10.1016/j.cell.2008.02.036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Nussinov R.; Zhang M.; Liu Y.; Jang H. AlphaFold, Artificial Intelligence (AI), and Allostery. J. Phys. Chem. B 2022, 126, 6372–6383. 10.1021/acs.jpcb.2c04346. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Nussinov R.; Tsai C. J.; Jang H. Neurodevelopmental disorders, immunity, and cancer are connected. iScience 2022, 25, 104492. 10.1016/j.isci.2022.104492. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Nussinov R.; Tsai C. J.; Jang H. How can same-gene mutations promote both cancer and developmental disorders?. Sci. Adv. 2022, 8, eabm2059 10.1126/sciadv.abm2059. [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Hu R.; Xu H.; Jia P.; Zhao Z. KinaseMD: kinase mutations and drug response database. Nucleic Acids Res. 2021, 49, D552–D561. 10.1093/nar/gkaa945. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Yavuz B. R.; Tsai C.-J.; Nussinov R.; Tuncbag N.. Discovery of latent drivers from double mutations in pan-cancer data reveal their clinical impact. 2021, bioRxiv 438239 [Google Scholar]
  71. Nussinov R.; Zhang M.; Maloney R.; Jang H. Drugging multiple same-allele driver mutations in cancer. Expet Opin. Drug Discov. 2021, 16, 823–828. 10.1080/17460441.2021.1905628. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Lahiry P.; Torkamani A.; Schork N. J.; Hegele R. A. Kinase mutations in human disease: interpreting genotype-phenotype relationships. Nat. Rev. Genet. 2010, 11, 60–74. 10.1038/nrg2707. [DOI] [PubMed] [Google Scholar]
  73. Nussinov R.; Tsai C. J.; Jang H. Are Parallel Proliferation Pathways Redundant?. Trends Biochem. Sci. 2020, 45, 554–563. 10.1016/j.tibs.2020.03.013. [DOI] [PubMed] [Google Scholar]
  74. Nussinov R.; Tsai C. J.; Jang H. Anticancer drug resistance: An update and perspective. Drug Resist. Updates 2021, 59, 100796. 10.1016/j.drup.2021.100796. [DOI] [PMC free article] [PubMed] [Google Scholar]
  75. UniProt: the universal protein knowledgebase in 2021. Nucleic Acids Res. 2021, 49, D480–D489. 10.1093/nar/gkaa1100. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

ct2c01171_si_001.pdf (643.1KB, pdf)
ct2c01171_si_002.xlsx (27.2KB, xlsx)
ct2c01171_si_003.xlsx (232.1KB, xlsx)
ct2c01171_si_004.xlsx (170.9KB, xlsx)
ct2c01171_si_005.xlsx (1.2MB, xlsx)

Articles from Journal of Chemical Theory and Computation are provided here courtesy of American Chemical Society

RESOURCES