Using the structural kinome to systematize kinase drug discovery

Kinase-targeted drug design is challenging. It requires designing inhibitors that can bind to specific kinases when all kinase catalytic domains share a common folding scaffold that binds ATP. Thus, obtaining the desired selectivity, given the whole human kinome, is a fundamental task during early-stage drug discovery. This begins with deciphering the kinase-ligand characteristics, analyzing the structure-activity relationships, and prioritizing the desired drug molecules across the whole kinome. Currently, there are more than 300 kinases with released PDB structures, which provides a substantial structural basis to gain these necessary insights. Here, we review in silico structure-based methods - notably, a function-site interaction fingerprint approach used in exploring the complete human kinome. In silico methods can be explored synergistically with multiple cell-based or protein-based assay platforms such as KINOMEscan. We conclude with new drug discovery opportunities associated with kinase signaling networks and using machine/deep learning techniques broadly referred to as structural biomedical data science.

in 2001, a significant breakthrough in kinase drug design for cancer treatment 12 , 63 small molecule kinase inhibitors have been approved by the FDA 13-14 as of Feb. 12, 2021. These drugs provide a variety of disease treatments, such as for non-small cell lung cancer (NSCLC) 15 , chronic myelogenous leukemia (CML) 16 , rheumatoid arthritis 17 , breast cancer 18 , and acute lymphoblastic leukemia (ALL) 19 . However, in practice, the off-target toxicities and other adverse effects, such as congestive heart failure and cardiogenic shock in some CML patients 20 , require the further development of more effective, highly selective inhibitors 3 .
Attaining such high selectivity is a daunting task since the inhibitor should bind to a specific primary kinase or select kinases, yet all kinase catalytic domains share a common folding scaffold that binds ATP 21 . To validate selectivity, kinome-scale screening of lead compounds has been attracting more attention [22][23] . Indeed, there are a number of experimental kinome-scale screening methods 22,24 , such as KinaseProfiler 25 , KinomeScan 26 , and KiNativ 27 . Although kinase profiling technologies are gradually maturing, they are expensive, especially for screening a large compound library against the whole kinome, which remains impractical.
With the availability of an increasing number of kinase structures, virtual structure-based drug screening provides a low-cost and effective way to filter a large compound library and identify the most likely compounds at an early stage of drug screening. Used concurrently with experiential profiling platforms in silico methods provide early-stage kinome-scale drug screening. Based on structural insights, the atom-level binding characteristics of every compound can be revealed and can be used as a guideline for further compound identification and optimization. Given the more than 300 kinases with released PDB structures, subtle differences have been found in the vicinity of the binding site where the adenine base of ATP binds, as well as binding sites away from the ATP binding site, such as in the C lobe of the kinase domain [28][29] . This structural corpora provides insights towards achieving the desired selectivity.
In this chapter we describe the characterization of the whole structure kinome to facilitate drug development. Specifically, we use the function-site fingerprint method to analyze the structural kinome providing systematic insights into kinase drug discovery. With increased knowledge of kinase-driven signaling pathways new kinase targets are continuously being explored for related disease treatment. Looking ahead structural biomedical data science, combining structure-based polypharmacology with machine/deep learning new challenges and opportunities are discussed.

Kinome-level profiling
Due to the common ATP-binding pocket, the kinase domain was thought to be undruggable prior to the 1990s 30 . With advances in protein-and cell-level experimental techniques and an increase in structure-based knowledge of protein kinases, variation among different kinases became apparent 31 . However, possible specificity requires kinome-scale validation. Moreover, with the increased knowledge of kinase signal pathways, comparison of traditional "one-drug-onetarget" models, have been replaced by the acceptance of polypharmacology. Examples include the FDA-approved drug Crizotinib targeting ALK and Met for treating NSCLC, and Cabozantinib targeting VEGFR, MET, RET, FLT1/3/4, AXL, and TIE2 for treating thyroid cancer. Hence, kinome-level profiling is an important step in confirming the selectivity of multi-target drugs.
Multiple commercial platforms provide kinome profiling services with panels ranging from 30 to 715 kinases 1, 32-33 (Table 1). nucleophilic/electrophilic, which sub-pocket is hydrophobic, or which amino acids can provide covalent interactions. These atom-level interaction details provide the basic principles by which to modify the functional groups of the given compound. Iteratively combining compound optimization with kinome profiling establishes lead compounds for further testing.  [44][45] . The 3-D kinase structure corpus provides a basis for structural kinome-based drug discovery. Scientists can not only directly review the compound binding details against the specific target, but they can also compare nuances -similarities and differences -among different kinase targets. For example, in comparing the ATP binding mode, 63 FDA-approved small-molecule kinase drugs can be divided into Type-I, II, III, or IV inhibitors 29,46 . Similarly, based on the possible existence of a covalent interactions, these kinase drugs can be divided into covalent (irreversible) inhibitors and noncovalent (reversible) inhibitors [47][48] (Table 2). Overall, the desired selectivity is achieved by utilizing every nuance of the different binding sites and accommodating the different sub-pockets of the binding sites among the different kinases 13,46,49 . As such, deciphering the whole structural kinome will be very useful in enhancing kinase inhibitor screening, optimization, and prediction. To this end, we introduce the alignment of the binding sites across the structural kinome and describe the characteristics of the aligned binding sites for achieving the desired selectivity. The approach comprises three steps ( Figure 1). First, preparing the structural kinome database. All The FsIFP approach, which examines the specificity among binding sites has been explored to design high-selectivity kinase inhibitors. Beyond the ATP binding pocket, there are other binding sites to be validated 59 , such as hydrophobic segment, allosteric segment, DFP motif area, and G-rich-loop region (Figure 1a). Corresponding to these binding regions, inhibitors are classified as Type-I, Type-II, and Type-III.
Type-I kinase inhibitors mainly bind to the ATP-binding site in the "DFG-in" conformation.
To obtain stronger binding affinity and greater selectivity than ATP, besides occupying the ATPbinding space, Type-I inhibitors extend into different proximal regions, specifically referred to as the front pocket region, the hydrophobic pocket region, the DFG motif, or the G-rich-loop region 13,59 . For example, Gefitinib is one Type-I drug for the treatment of non-small cell lung cancer

Challenges and Opportunities
Since the launch of the first kinase drug, Imatinib in 2001, kinase targeted drug discovery has been on a fast track. In the last six years, an average of eight small molecule kinase drugs have been approved per year. This tremendous success benefits patients, but also highlights our ability to achieve drug discovery outcomes 9, 64 . However, challenges still remain in the development of efficient, non-toxic kinase-targeted drugs 3 .
Clinical adverse effects are one major challenge. For example, kinase drugs affect the digestive system and cause nausea, vomiting, and/or diarrhoea 65 . Further, most kinase inhibitors cause serious adverse effects, such as different degrees of cytopenia 66 . These side effects typically result from off-targets effects. To avoid such side effects, a highly selective drug is desired. Alternatively, adverse effect can be due to on-target toxicities involving the intrinsic mechanisms of the drugs 67 .
At this point in the evolution of small molecule kinase drugs novel compound scaffolds are needed.
to reduce adverse effect as much as possible. Scaffolds that are available for early-stage screening.
The increasing availability of panels of phenotypic assays may provide one strategy to profile selectivity by combining virtual structure-based kinome screening, which can filter a huge compound library into a highly focused kinase library 68 .
Another challenge is acquired drug resistance 58,69 . In clinical practice, kinase-targeted drugs are frequently subject to drug resistance, which has become a primary vulnerability in targeted cancer therapy. The first difficulty is exploring resistance mechanisms due to the diversity of specific drug-binding mechanisms. For example, drug resistance of Erlotinib, which is one FDA-approved kinase drug used to treat patients with EGFR-overexpression induced NSCLC, is caused by the gatekeeper T790M mutation, which increases the binding affinity of ATP to the EGFR kinase 70 .
In another example, Crizotinib was often found to be ineffective in the majority of patients after 1−2 years' treatment against ALK-positive NSCLC due to the acquired ALK L1196M mutation, which decreased the binding affinity of Crizotinib 71 .
Nevertheless, these challenges also provide unique opportunities to develop new approaches and applications. Currently, in vitro and/or in vivo kinome-scale and proteome-scale profiling methodologies have been merged into the drug design pipeline, which potentially provides a thorough understanding of targets and selectivity of kinase inhibitors. Combined with the diseases' signal pathway, the target spectrum can be further applied to "one-drug-multiple-target" drug design. For example, Midostaurin is a multi-target kinase drug 72 used to treat adult patients with newly diagnosed FLT3-mutated acute myeloid leukemia (AML). For a "multiple-drug-multiple-target" combination therapy strategy, Capmatinib (a MET inhibitor) and Gefitinib (an EGFR inhibitor) had been approved to treat patients with EGFR-mutated-MET-dysregulated -in particular, MET-amplified -NSCLC 73 . We can expect profiling methodologies will be further developed to cover the whole kinome and even the proteome. one virtual profiling assay model to support virtual screening, compound repurposing, and the detection of potential off-targets 37 . Here, we collate the free databases of available kinase-inhibitor activity ( Table 3). It is worth noting that the ChEMBL Kinase SARfari database, which contains ~54,000 compounds, ~980 kinases targets, and the corresponding approximately 530K structureactivity data points 78 , has been used to predict kinome-wide profiling of small molecules [79][80][81] .
Taken together, data-driven methods and applications will further experimental protocols and facilitate the drug discovery processes.