Understanding selectivity of CLK kinase inhibitors: structural insights with 3decision®
The development of kinase-targeting compounds is a very active field in drug discovery. Protein kinases regulate many cellular processes, and so their inhibition is an attractive therapeutic strategy. However, kinase inhibitors are known to be associated with selectivity issues. Existing drugs often target the same, highly conserved ATP-binding site, mimicking the binding mode of ATP. This leads to side effects in pharmacological therapy and ultimately to the failure of drug discovery efforts.
For this reason, there is a high interest in the development of novel, non-ATP-mimetic kinase inhibitors that will overcome selectivity issues, which translates into increased success rates of drug discovery programs in this research area.
In this blog post, we are going to present the structural basis of the CLK kinase inhibitors selectivity using the core features of 3decision®– Discngine’s 3D protein structure data management and analytics software. We will show you how to efficiently navigate through CLK protein structural data and quickly get relevant structural insights to promote ideation and drive drug design.
Used to watching videos? We have a complete visual representation of the same use case in the video tutorial. Access it below!
Introduction: CLK Kinase family selective inhibitors
O. Fedorov et al. reported that derivatives of the natural bioactive compound Bauerine C are highly specific inhibitors for some members of the CDC2-like kinase (CLK) protein family: CLK1 and 4 are inhibited, while CLK3 is not.
What are the structural bases for the observed selectivity profile?
In a recent study, M. Schröder et al. elucidated this behavior through a mutagenesis study. Following the publication, we reproduced this use case using 3decision:
First, we created a relevant data set and then, we examined and compared the differences in the binding mode, all in the same 3decision platform and in just a few clicks.
The illustrated workflow can be easily applied to a structure-based drug discovery campaign.
There is also a bonus at the end of the article, so be sure to follow it to the very end.
Creation of a data set of CLK structures
To start investigating the binding mode of the Bauerine C derivatives with the CLK family, we first need to create a data set of relevant protein structures. To do this, we can search for all the available structures of the CLK protein family bound to the dichloro-methyl-indole chemical moiety, which is contained in the Bauerine C structure.
We can easily collect them in 3decision using Advanced Search feature, that permits us to combine multiple queries, both by general properties and chemical structure. In our case, we will combine the searches (Figure 1):
By Protein Family: CMGC protein kinase CLK
By the dichloro-methyl-indole chemical substructure
Each search is run separately through all the structures deposited in the database (proprietary and public, experimental and models) and then the hit results are combined.
Figure 1: The query based on the combination of protein family name “CLK” and chemical substructure “dichloro-methyl-indole” in the 3decision Advanced search
In this way, we retrieved 8 CLK structures co-crystallized with Bauerine C derivatives, available in the 3decision database.
The search results can be evaluated and further refined on the results navigation page. By default, the results page is set on the “Structure” view mode, which gives a general overview of the structures found by the search (Figure 2). We can see from this page that we only have CLK1 and CLK3 3D structures in complex with the chemical substructure of interest, and none for the other members of the family CLK2 and CLK4.
Figure 2: The 8 CLK hit structures from the query in the “Structure” view mode of the 3decision results navigation page. Click to Zoom In
To better refine and navigate through the results, we can focus on the chemical matters bound to the proteins by switching to the “Ligand” view mode (Figure 3).
Figure 3: The search results in the “Ligand” view mode of the 3decision results navigation page showing the chemical matter found in the hit structures.
This view helps us to quickly see that we have three chemical scaffolds in complex with the proteins. We will focus our analysis on the three PDB structures that contain the whole Bauerine C scaffold (7,8-dichloro-1-oxo-β-carboline), which are the result of a mutagenesis study (M. Schröder et al.).
We can save this collection of structures by creating a Project, that also allows us to continue the work later and share this set of data with colleagues involved in a project. To visualize the 3D structures, we need to set a reference structure for the project. We will use the structure in complex with CLK1 - the member of the CLK family that we know is effectively inhibited by the ligand (PDB: 6YTG). In the setting of the project reference, we need to select a pocket in the structure, which will be the reference pocket to which all other structures in the project will be aligned, and finally, we can set the viewpoint (Figure 4).
Figure 4: Set up of a reference structure for the collection of structures saved in the 3decision Project.
Assessment of 3D structural differences
Now that we have a desired data set, we are ready to analyse it, to understand the binding mode of the ligands and the selectivity profile of the CLK family.
We open our saved project in the Workspace, which is the core of 3decision. Here we can visualize the 3D structures of proteins, retrieve all associated structural information, and perform several analyses. (To understand which information is associated with the 3D structures registered in 3decision, have a look at our video).
When the reference structure (CLK1, PDB: 6YTG) is opened in the 3D viewer of the workspace, protein-ligand contacts are automatically displayed. Protein-ligand interactions are calculated by 3decision upon registration of the structures in the database. In this way, we can easily inspect the ligand binding, without the need to export the structure in a different program to calculate the protein-ligand contacts.
If we focus on the ATP binding site, we can see that the ligand does not establish any canonical hydrogen bond interactions with the hinge region (Figure 5A). This is different from the binding mode of the ATP and ATP-mimicking inhibitors where the purine ring is forming hydrogen bonds (H-bonds) with the carbonyl and amine groups on the backbone (Figure 5B).
Interestingly, the Bauerine C scaffold (7,8-dichloro-1-oxo-β-carboline) in the reference structure interacts with the hinge through the chlorine atoms on the indole portion, instead of interacting canonically through the lactam portion of the molecule, which would be capable of forming H-bonds with the hinge. Surprisingly, the pyridone and spiropiperidine rings are pointing toward the back of the binding pocket.
This structural consideration already explains the selectivity of these kinase inhibitors over other chemical scaffolds.
Figure 5: ATP binding site for kinases: A) CLK1:bauerine C derivative complex (blue, PDB: 6YTG) non-canonical binding mode, halogen bonds between chlorine atoms (in green) and the hinge region; B) CDK7:ATP complex (orange, PDB: 1UA2) showing the canonical ATP binding mode.
To understand the selectivity profile observed within the CLK family, we need to superpose the reference with the other two available structures in our data set, which are two structures of the CLK3 protein in complex with the same ligand. In the 3decision workspace, you can superpose structures on-the-fly, in one click. The superposition is pocket-based: the structures are aligned on the given reference pocket.
We can immediately see that there is a difference in the binding mode for one of the structures (PDB: 6YU1): the orange ligand is flipped compared to the other two structures. It interacts with the hinge region through the lactam moiety, forming H-bonds in a canonical ATP-like modality (Figure 6), while the hydrophobic dichlorobenzene points into the back pocket.
Figure 6: Superposition of all the project structures in the 3decision 3D Viewer. The blue (CLK1, PDB: 6ytg) and white (CLK3 A319V, PDB: 6Z2V) have the same non-canonical binding pose (halogen bonds in purple), while the orange structure (CLK3, PDB: 6YU1) shows and ATP-like interaction pattern with the hinge region (H-bonds in yellow, indicated by the orange arrows).
To check the difference between the two CLK3 structures, we can look at the information panel (Figure 7): the structure that is showing the “flipped” binding pose is the wild type CLK3, while the other one carries a mutation at the DFG-1 residue of the ATP binding site (A319V mutant).
Figure 7: General information on the CLK3 A319V structure summarized in the 3decision Information panel on the left.
To investigate the structural basis for the different poses among the CLK family, we can activate the highlight mode in the 3D Viewer, which facilitates the comparison of multiple structures. With this modality, only the differences between the residues in the reference pocket and the others loaded in the workspace are shown. The residues that are detected as different (which are above the RMSD threshold set), are represented as sticks. We set the RMSD to 1.3.
If we focus on the mutation at the DFG-1 position, the CLK1 and the A319V CLK3 mutant structures show a Valine, while the wild-type CLK3 has an Alanine (Figure 8). The ligand binds both CLK1 and A319V CLK3 with the same, non-canonical pose, while in the wild type CLK3 it assumes a canonical, ATP-like pose. This indicates that the presence of the bulky Valine residue at the DFG-1 position induces the non-canonical pose of the ligand, with the dichloroindole oriented towards the hinge, stabilizing the carboline portion in the back pocket. On the contrary, the smaller Alanine at position 319 in the CLK3 structure opens a larger pocket that allows it to accommodate the ligand with the canonical ATP-mimicking orientation. Therefore, this residue difference in the binding site between CLK1 and CLK3, is responsible for the different binding modes, and consequently, for the selectivity, as was also confirmed by the comparison with the mutant A319V.
With the highlight mode, we could rapidly rationalize the observed differences in the mutagenesis campaign on the CLKs.
Figure 8: Differences between the structures highlighted with the 3decision highlight mode. The residues at the DFG-1 position are indicated in the green circle.
Conclusion
This mutagenesis study proves that the DFG-1 site can be of great interest in overcoming the selectivity issues in kinases. In fact, most kinases harbour a small residue, and so the few kinases that present bulky residues (such as valine, isoleucine and leucine) could be selectively targeted by exploiting this difference.
In this blog post, we showed how 3decision can help you to:
efficiently navigate through structural
data quickly retrieve valuable structural knowledge
In the described CLK use case, we were able to collect the data of interest for investigating the structural basis of the selectivity observed in the protein kinase CLK family, using a Bauerine C derivative.
Bonus: explore 3D structure through annotations
The residue difference highlighted in the use case is so important since it is in a crucial position of the ATP-binding site: DFG-1. To facilitate the detection of important regions on the protein, 3decision provides a tool called the Annotation Browser. Here, sequence annotations of the protein are listed and synched with the 3D Viewer, to easily spot their location in the 3D structure. The sources of these annotations are the Uniprot and Pfam, and we also have a “User annotation” section that is customized by 3decision team, depending on the protein family.
In this case, we are looking at kinase-specific annotations. If we look at the V324 in CLK1, we can see that it corresponds to the “x motive preceding DFG” annotation, hence the DFG-1 residue (Figure 9). Using this tool, the navigation through different domains and residues of the protein becomes very easy and intuitive.
Figure 9: The DFG-1 Valine residue of CLK1 is highlighted in red in the 3D structure, using the Annotation Browser.
References
Fedorov, O.; Huber, K.; Eisenreich, A.; Filippakopoulos, P.; King, O.; Bullock, A. N.; Szklarczyk, D.; Jensen, L. J.; Fabbro, D.; Trappe, J.; Rauch, U.; Bracher, F.; Knapp, S. Specific CLK Inhibitors from a Novel Chemotype for Regulation of Alternative Splicing. Chem. Biol. 2011, 18 (1), 67–76. https://doi.org/10.1016/j.chembiol.2010.11.009.
Schröder, M.; Bullock, A. N.; Fedorov, O.; Bracher, F.; Chaikuad, A.; Knapp, S. DFG-1 Residue Controls Inhibitor Binding Mode and Affinity, Providing a Basis for Rational Design of Kinase Inhibitor Selectivity. J. Med. Chem. 2020, 63 (18), 10224–10234. https://doi.org/10.1021/acs.jmedchem.0c00898.
If you wish to learn more about how to uncover the full potential of your 3D protein structures to drive drug discovery, get the free copy of our whitepaper.