(338c) Discovery of CDK8 Inhibitors By Novel Docking Algorithm with CDK8-Targeted Scoring Function

Introduction

Cycline C dependent kinase 8 (CDK8) is an interesting oncogenic kinase target to treat cancers like acute myeloid leukemia, breast and colorectal cancer [1]. CDK8 regulates β-catenin-dependent transcription following the activation of WNT signaling [2]. CDK8 is regarded as an early clinical stage drug target to treat cancer [3]. IC50 is a basic parameter to evaluate the binding property of the drug-receptor [5]. Experimental methods (e.g., radioligand binding assay [6], surface plasmon resonance (SPR) [7], fluorescence energy resonance transfer method (FRET) [8], affinity chromatography [9], and isothermal titration calorimetry [10]) are typically used to screen ligands (also named drugs or compounds) to inhibit CDKs. However, it is costly and time-consuming to determine IC50 for thousands of ligands using these experimental approaches. This explains why only a few CDK inhibitors like seliciclib [11] and flavopiridol [12] were tested in clinical trials. It is thus in an urgent demand to develop affordable and efficient approaches to infer IC50 to accelerate the discovery of CDK inhibitors for clinical trials.

Docking-based drug design is widely used in lead optimization and contributes to many approved drugs. Although docking programs and severs are available, most of docking functions that predict binding affinity are performed for all-purpose use. The prediction may be not accurate for one particular target [14]. A comparation study suggested that docking functions are either linear or nonlinear regression models for quantifying the correlation between predicted binding energies of large-scale crystal structure data and experimental binding affinity (e.g., IC50 or Ki). However, the correlation from these approaches was generally low (i.e., R² less than 0.3) [15]. In this study, we aim to develop a novel docking algorithm for CDK8-targeted scoring function that specifically evaluates the affinity of known CDK8 inhibitors and screen new compounds to inhibit CDK8.

Method

Chemical structure and in vitro inhibition (IC50) of CDK8 inhibitors was obtained from the literature [13, 16, 17]. Structures of CDK8 in complex with selected inhibitors were downloaded from Protein Data Bank, including 5FGK, 5HBE, 5HBH, 5HBJ, 5I5Z, 5ICP, 5IDP, and 5IDN [18]. As Molsoft ICM-pro was able to generate a 4D flexible docking receptor from multiple crystal structures and return 93% accuracy in flexible docking, it was used to calculate binding energies [19, 20]. The structures of CDK8 protein were modified and merged into a 4D flexible receptor in Molsoft ICM, with the deletion of ligand and water, addition of hydrogens, and generation of stack from an ensemble of the crystal structures. The compounds were docked into pocket of ATP binding site of CDK8 with 10 effectiveness. Docking poses were identical with ligand conformations in crystal structures. Therefore, data of free energies/entropy calculated by ICM was collected, including Natom, Hbond, Hphob, VwInt, Eintl, Dsolv, SolEl, dTSsc.

On the basis of docking data returned from ICM and inhibition IC50 data from the literature, various models have been developed in this study to infer IC50 value from the compound’s binding data (i.e., Natom, Hbond, Hphob, VwInt, Eintl, Dsolv, SolEl, dTSsc). In particular, The following models were developed and implemented in the R programming platform: principle component regression (PCR), partial least squares regression (PLSR), logistic regressions, support vector machine, and fuzzy regression. Data of 103 compounds was randomly half-half spliced into the training and testing sets for 1000 simulations. For each simulation, the training data set was used to estimate parameters in the aforementioned models, while the testing data set was implemented to validate prediction capability of those models using criteria like Coefficient of determination (R²) and Root-mean-square deviation (RMSE).

Major findings

It turns out that the PCR and PLSR models return the best predicted IC50 from binding affinity, with R² as 0.5 and RMSE of pIC50 less than 0.5. The models were then applied to evaluate hundreds of derivatives with our novel scoring function. Those compounds with the best inhibition potential provide good candidates for future in vitro experiment validation. Therefore, the models developed in this work for predicting compounds’ inhibitory effect would reduce time and cost for high-throughput screening and accelerate the development of novel CDK8 inhibitors.

Reference

1. Xi, M., et al., CDK8 as a therapeutic target for cancers and recent developments in discovery of CDK8 inhibitors. European Journal of Medicinal Chemistry, 2019. 164: p. 77-91.
2. Ferguson, F.M. and N.S. Gray, Kinase inhibitors: the road ahead. Nature Reviews Drug Discovery, 2018. 17: p. 353.
3. Liang, J.X., et al., CDK8 Selectively Promotes the Growth of Colon Cancer Metastases in the Liver by Regulating Gene Expression of TIMP3 and Matrix Metalloproteinases. Cancer Research, 2018. 78(23): p. 6594-6606.
4. Rzymski, T., et al., CDK8 kinase—An emerging target in targeted cancer therapy. Biochimica et Biophysica Acta (BBA) - Proteins and Proteomics, 2015. 1854(10, Part B): p. 1617-1629.
5. Weiland, G.A. and P.B. Molinoff, Quantitative-Analysis of Drug-Receptor Interactions .1. Determination of Kinetic and Equilibrium Properties. Life Sciences, 1981. 29(4): p. 313-330.
6. Frey, K.A. and R.L. Albin, Receptor binding techniques. Curr. Protoc. Neurosci. Chapt. 1., 2001. Unit 1.4.
7. Brannstrom, K., et al., The N-terminal Region of Amyloid beta Controls the Aggregation Rate and Fibril Stability at Low pH Through a Gain of Function Mechanism. Journal of the American Chemical Society, 2014. 136(31): p. 10956-10964.
8. Capraro, B.R., et al., Kinetics of Endophilin N-BAR Domain Dimerization and Membrane Interactions. Journal of Biological Chemistry, 2013. 288(18): p. 12533-12543.
9. Kim, H.S. and D.S. Hage, Chromatographic analysis of carbamazepine binding to human serum albumin. Journal of Chromatography B-Analytical Technologies in the Biomedical and Life Sciences, 2005. 816(1-2): p. 57-66.
10. Li, L.C., et al., Discovery of two aminoglycoside antibiotics as inhibitors targeting the menin-mixed lineage leukaemia interface. Bioorganic & Medicinal Chemistry Letters, 2014. 24(9): p. 2090-2093.
11. Raje, N., et al., Seliciclib (CYC202 or R-roscovitine), a small-molecule cyclin-dependent kinase inhibitor, mediates activity via down-regulation of Mcl-1 in multiple myeloma. Blood, 2005. 106(3): p. 1042-1047.
12. Senderowicz, A.M., Flavopiridol: the first cyclin-dependent kinase inhibitor in human clinical trials. Investigational New Drugs, 1999. 17(3): p. 313-320.
13. Mallinger, A., et al., Discovery of Potent, Selective, and Orally Bioavailable Small-Molecule Modulators of the Mediator Complex-Associated Kinases CDK8 and CDK19. Journal of medicinal chemistry, 2016. 59(3): p. 1078-1101.
14. Guedes, I.A., F.S.S. Pereira, and L.E. Dardenne, Empirical Scoring Functions for Structure-Based Virtual Screening: Applications, Critical Aspects, and Challenges. Frontiers in pharmacology, 2018. 9: p. 1089-1089.
15. Warren, G.L., et al., A critical assessment of docking programs and scoring functions. J Med Chem, 2006. 49(20): p. 5912-31.
16. Mallinger, A., et al., 2,8-Disubstituted-1,6-Naphthyridines and 4,6-Disubstituted-Isoquinolines with Potent, Selective Affinity for CDK8/19. ACS medicinal chemistry letters, 2016. 7(6): p. 573-578.
17. Czodrowski, P., et al., Structure-Based Optimization of Potent, Selective, and Orally Bioavailable CDK8 Inhibitors Discovered by High-Throughput Screening. Journal of Medicinal Chemistry, 2016. 59(20): p. 9337-9349.
18. Goodsell, D.S., et al., RCSB Protein Data Bank: Enabling biomedical research and drug discovery. Protein Sci, 2020. 29(1): p. 52-65.
19. Bursulaya, B.D., et al., Comparative study of several algorithms for flexible ligand docking. J Comput Aided Mol Des, 2003. 17(11): p. 755-63.
20. Neves, M.A., M. Totrov, and R. Abagyan, Docking and scoring with ICM: the benchmarking results and strategies for improvement. J Comput Aided Mol Des, 2012. 26(6): p. 675-86.