Introduction

Casein kinase 2 (CK2) is a highly pleiotropic serine-threonine kinase1 that is naturally endowed with constitutive activity and lacks gain-of-function mutants. It is often present as a heterotetramer composed of two catalytic α subunits (α or α′) and two regulatory β subunits in various combinations2,3. CK2 is a key regulator of various cellular events, and its phosphorylation targets include more than 300 proteins4 that are implicated in the regulation of many cellular functions, including signal transduction, transcriptional control, apoptosis, and the cell cycle5,6,7. On the other hand, CK2 is abnormally elevated in a wide variety of tumors6. Overexpression of CK2 was documented in kidney8, mammary gland9, lung10, head and neck11 and prostate cancers8. Thus, CK2 is currently considered to be a promising therapeutic target, and many CK2 inhibitors targeting the classical ATP substrate site have been developed12, such as emodin13, TBB14, (5-oxo-5,6-dihydroindolo[1,2-a]quinazolin-7-yl)aceticacid15, and CX-494516,17. Some CK2 inhibitors, such as 5,6-dichloro-1-b-D-ribofuranosylbenzimidazole (DRB)18, also block the interface between the CK2α and CK2β subunits. Nevertheless, the lack of selectivity of these known CK2 inhibitors has impeded their development in cancer therapy.

Allosteric inhibitors can avoid the shortcomings of most conventional ATP-competitive kinase inhibitors, such as poor selectivity, in vivo competition with a high ATP concentration, and reliance on highly conserved residues in the ATP binding site susceptible to mutation19. One way to discover new allosteric molecules is to identify a new allosteric pocket19,20. Inhibitors that bind these allosteric pockets exert control over the protein's function by perturbing the ensemble of structures that the protein adopts21,22,23. Such sites could have biological functions and serve as valuable targets for drug design. Unfortunately, the discovery of new allosteric pockets is challenging. Most known allosteric sites were serendipitously discovered from the results of protein-ligand structural complexes24,25. To date, only Jennifer Raaf has found an allosteric site in CK2α that can be alternatively filled by (DRB) or glycerol18 in the complex structure of CK2α and DRB. However, the specificity of the cavity is obviously low compared with that of other kinases, which is not good for developing selective inhibitors.

It is urgent to identify more allosteric pockets in CK2α for further inhibitor discovery. In the present study, computational and experimental experiments were performed to explore such allosteric sites for drug design in CK2α. Using Allosite prediction and point mutation analysis, we identified a new allosteric site in CK2α.

Materials and methods

Plasmid constructs and site-directed mutagenesis

The His-tagged CK2α construct was created by the direct insertion of the CK2α gene into the pET28a expression vector between the BamHI and HindIII restriction sites. The template of CK2α was amplified from pDB1-CK2α (Addgene, Cambridge, MA, USA). All DNA oligonucleotides were produced by Sangon Biotech Co, Ltd (Shanghai, China). Point mutations were created using the Site-Directed Mutagenesis Kit (Stratagene, La Jolla, California, USA) according to standard protocols. The sequences of the individual clones were verified.

Protein expression and purification

His-fusion vectors of CK2α and the indicated mutant vectors were transformed into BL21 Escherichia coli. All wild-type and mutated vectors were expressed and purified identically. Expression was induced with 0.5 mmol/L IPTG at 0.6–0.8 OD and proceeded for 12–16 h at 16 °C. Bacterial cells were harvested by centrifugation at 4000×g for 15 min, and the pellets were resuspended in lysis buffer containing 20 mmol/L TRIS-HCl at pH 7.5, 50 mmol/L NaCl, 1 mmol/L phenylmethylsulfonyl fluoride, and 20 mmol/L imidazole. The cells were lysed by sonication, and the lysates were cleared by centrifugation at 20 000×g for 40 min. The supernatants were incubated with 1 mL of previously washed Ni Sepharose Fast Flow (GE Healthcare, Chicago, Illinois, USA) per liter of bacterial culture and equilibrated with the lysis buffer. After 1 h at 4 °C, the beads were washed with lysis buffer and eluted with lysis buffer containing different concentrations of imidazole. After the Ni2+ affinity chromatography, CK2α and its mutants were purified by Superdex 200 gel filtration chromatography (GE Healthcare, Chicago, Illinois, USA) and equilibrated with a buffer containing 20 mmol/L TRIS-HCl at pH 7.5, 50 mmol/L NaCl, 1 mmol/L EDTA, and 2 mmol/L DTT. The peak fractions were combined and concentrated. Then, the purified proteins were aliquoted, snap-frozen in liquid nitrogen, and stored at −80 °C for subsequent use.

ADP-Glo™ CK2α kinase assay

Kinetic analyses of CK2α and the indicated mutants were performed using a luminescent ADP detection assay (Promega, Madison, WI, USA) by varying the concentration of the peptide substrate (RRRADDSDDDDD). The peptide substrate (>95% purity) was synthetized by Wuhan Holder Co, Ltd (Wuhan, China). White low-volume 384-well polystyrene plates (ProxiPlate-384 Plus, PerkinElmer, Waltham, MA, USA) were used for the kinase assay. In total, 2.5 μL of kinase were mixed with 2.5 μL of the kinase buffer (50 mmol/L TRIS at pH 7.6, 10 mmol/L MgCl2, 2 mmol/L DTT, varied concentrations of substrate, and 200 μmol/L ATP). The enzyme concentrations were chosen so that the reactions were linear during the 220-s incubation period, and all reactions were carried out in triplicate. Blank wells lacked the enzyme but included the kinase buffer, the substrate, and ATP. The plates were covered, and the reactions were carried out at 25 °C for up to 220 s. The reactions were stopped with the addition of 5 μL of ADP-GloTMReagent. After 40 min of incubation at room temperature, 10 μL of Kinase Detection Reagent were added, and the plates were incubated for another 60 min at room temperature. The plates were read on a Synergy H4 microplate reader (BioTek Instruments, Winooski, USA) with an integration time of 1 s per well. The apparent KM and Vmax values were obtained from a non-linear curve fitted to the Michaelis-Menten equation and analyzed using Prism 5.0 software (GraphPad, La Jolla, CA, USA). The kcat was calculated as Vmax divided by the valid enzyme concentration.

Prediction of the allosteric sites in CK2α

All 127 crystal structures of CK2α were downloaded from Protein Data Bank26. To cover potential sites as much as possible, all 127 structures with diverse conformations were used to identify allosteric sites in CK2α. These query structures were submitted into Allosite, and 1510 sites were found and merged into 31 primary pockets based on the overlay of the structures. First, 25 weighted descriptors (eg, Normalized maximum distance between two alpha spheres, normalized hydrophobicity density, flexibility score) in the Allosite model were chosen to characterize all primary pockets, and the feature score for each pocket was calculated. Second, normal mode analysis (NMA) was also performed to perturb each primary pocket with a dummy ligand. The extent of flexibility at each pocket was evaluated based on the difference between the B-factors before and after binding the ligand, and the dynamic score of each pocket was calculated using the flexibility. Then, the allosite score for each primary pocket was obtained according to the normalization of feature score and dynamic score. Finally, allosite scores larger than 0.5 were used for further validation.

Identification of the allosteric pathway between the allosteric site and the substrate site

The MCPath server (http://safir.prc.boun.edu.tr/clbet_server) is based on MC path generation runs with a PDB ID or an uploaded structure. Using the query structure from the server, the MCPath generation method was used to calculate all of the likely allosteric communication pathways by generating an ensemble of maximum probability paths, as well as an infinitely long path for plausible functional residues based on the graph centrality measures. Finally, the output of results included the top three populated pathways shown on the structure's ribbon diagram27.

Molecular graphics

All of the protein structure figures were generated with PyMOL Molecular Graphics System (Version 1.8; Schrödinger, LLC). The amino acid sequences of human CK2α (P68400), CK2α'(P19784), CDKL2(Q92772), CDKL3(Q82VW4), CDKL4(Q5MAI5), CDK1(P06493), CDK2(P24941), CDK5(Q00535), CDK7(P50613), CDK8(P49336), CDK9(P50750), CDK10(Q15131), CDK11B(P21127), CDK12(Q9NYV4), CDK14(O94921), CDK16(Q00536), CDK20(Q8IZL9), MK01(P28482), MK03(P27361), MK04(P31152), MK06(Q16659), MK07(Q13164), MK08(P45983), MK09(P45984), MK10(P53779), MK11(Q15759), MK12(P53778), MK13(O15264), MK14(Q16539), CLK1(P49759), CLK3(P49761), CLK4(Q9HAZ1), GSK3A(P49840) and GSK3B(P49841) were aligned using Clustal X28, and the figures were generated with ESPript (http://espript.ibcp.fr/ESPript/ESPript/)

Results

Computational detection of the known hidden allosteric site in CK2α

Using our algorithm Allosite24,25, we computationally identified the potential allosteric sites in CK2α. The Allosite results showed that there are 6 predicted allosteric sites distributed around the surface of CK2α (Figure 1 and Table 1). Interestingly, site 1 corresponded to the ATP substrate site18, whereas the DRB2-binding allosteric site was identified as site 6. These results could be considered an internal validation of the Allosite methodology.

Figure 1
figure 1

Predicted allosteric sites using the Allosite method. The structure of CK2α is represented by a cartoon, and the residues at the predicted sites are shown in sticks. Residues at site 1 (ATP substrate site) are colored in purple, residues at site 2 are colored in red, residues at site 3 are colored in gray, residues at site 4 are colored in cyan, residues at site 5 are colored in green, and residues at site 6 (DRB2-binding allosteric site) are colored in blue.

PowerPoint slide

Full size image
Table 1 The predicted allosteric sites in the CK2α proteins.
Full size table

In addition to the known sites, a new potential allosteric site at rank 5 (referred to as site 5 in Table 1) was predicted around the loop linking the β3 strand and the αC helix (Figure 1). The conformation of the αC helix has been reported to correlate with kinase activity in another kinase, PDK129, leading to the identification of an allosteric site in that kinase30. Then, further analysis of site 5 showed that the structural conservation of the pocket located on the surface of the αC helix was low, 27.5%, compared with other kinases. In addition, the pocket is present even in the absence of CK2α substrate. This type of pocket may be more suitable for an allosteric pocket20. Therefore, we expected that perturbations at this site could modulate the ensemble of CK2α conformations and regulate the protein's activity.

Effects of distal mutations in the predicted site on CK2α catalysis activity

The predicted site 5 was tested experimentally using site-directed mutagenesis, a method that has been useful for the determination of allosteric sites and key residues that play a role in signaling31,32. We selected residues exposed on the surface of site 5 for experimental validation after inspecting the CK2α structure using PyMOL. These mutations include I78, I82 and P109, which form the bottom of the pocket, and V31 and K75, which are on the flank of the pocket (Figure 2H). Among these residues, K75 and I78 are part of the αC helix, V31 is located in the N-terminal segment loop, and P109 is in the β5 strand. First, we chose residue I78, which was close to a basic center related to substrate recognition33. Our computational model also predicted that it could form a bump when the residue was mutated to tryptophan, and we hypothesized that this mutation minimized potential perturbations to the protein but changed the conformation of the pocket performing the reaction. Thus, we introduced a mutation into this region by replacing isoleucine with tryptophan or cysteine and performed an enzyme activity assay to determine the effect on CK2α activity. The kinase activities and catalytic rate of CK2α were measured with the ADP-Glo™ kinase assay34 using the peptide substrate RRRDDDSDDD35 and ATP. This mutation resulted in a decrease in the catalytic rate kcat from 290.5 min−1 for wild-type CK2α to 150.2 min−1 for the I78W mutant and 193.9 min−1 for the I78C mutant. The I78W and I78C mutants catalyzed the reaction at 52% and 66.7% of the rate of wild-type CK2α, respectively. The KM of I78C was not changed relative to wild-type CK2α, but the KM of I78W was reduced from 285.2 μmol/L to 179.9 μmol/L (Figure 2A, 2D, 2E and Table 2). The mutation of I78 to hydrophilic cysteine or hydrophobic tryptophan had a different influence on the activity of CK2α, which indicated that I78 is an important residue in maintaining the activity of CK2α.

Figure 2
figure 2

Michaelis–Menten plot of CK2α and its mutants with a peptide substrate. (A) Wild-type CK2α, (B) CK2α-V31R, (C) CK2α-K75E, (D) CK2α-I78C, (E) CK2α-I78W, (F) CK2α-I82C, and (G) CK2α-P109C. The reaction velocities for CK2α and its mutants were assayed in the presence of 3.9–4000 μmol/L of RRRDDDSDDD, the peptide substrate, and the results were fitted to the Michaelis–Menten equation. (H) Residues selected for mutation in the predicted pocket. Ribbon diagram of CK2α highlighting the residues on the bottom of the predicted allosteric pocket (I78 and I82 colored in magenta) and the flank of the predicted allosteric pocket (V31, K75 and P109 colored in yellow). The amino acids are shown in stick diagram. The αC helix and the activation loop are colored in light pink.

PowerPoint slide

Full size image
Table 2 The effect of mutations in the CK2α proteins on the kinetic constants for CK2-catalyzed substrate phosphorylationa.
Full size table

We also mutated four other important residues in the pocket. The kinetic constants of all of the CK2α mutants with the peptide substrate were determined and compared with wild-type CK2α. The kcat and KM values and the overall catalytic efficiencies expressed by the kcat/KM are summarized in Figure 2 and Table 2. Surprisingly, the mutants V31R, K75E, I82C and P109C displayed significantly increased catalytic efficiencies relative to that of wild-type CK2α, with each showing a 1.6-fold, 4.6-fold, 3.5-fold, and 2-fold increase in the value of kcat/KM, respectively. A detailed examination of the data in Table 2 allowed a rough subdivision of the mutants into the following two categories: (i) mutants with a KM for the substrate peptide that was only slightly decreased, whereas the kcat was actually increased or decreased relative to that of wild-type CK2α including V31R, I82C, P109C, and I78C; (ii) mutants with a KM and kcat for the peptide that were altered in a sharply different manner compared with wild-type CK2, including K75E and I78W. K75 and I78 are located in the αC helix and may have a stronger influence than the other four residues on CK2α activity. In fact, lysine mutated to glutamic acid has also been implicated in activating mutations in encephalocraniocutaneous lipomatosis36. At the same time, the large increase in the catalytic efficiencies of K75E, I82C and P109C resulted from significantly and reproducibly higher kcat values and lower KM values. A possible interpretation was that these mutants might also stimulate CK2α activity by interacting with V116, an important residue in the binding of ATP to CK217. These results showed that mutations in this site could both activate and inhibit the activity of CK2α.

The predicted allosteric pathway between the allosteric site and the active site

The MCPath generation approach is an effective tool for the prediction of key residues that mediate allosteric communication in an ensemble of functionally pathways27. Therefore, we employed the MCPath approach to investigate the potential communication between the predicted pockets and the active site, as indicated by correlations in the ensemble of structures that CK2α adopts. Residue H160 is homologous to PKA E170, which is responsible for the binding of the basic residue in the peptide substrate of CK2α33,37,38. H160 has been recently shown to contribute to the recognition of an acidic residue whenever it is present in the peptide substrates of CK239. The mutation of H160 caused the enzyme to be nearly inactive, hampering the calculation of kinetic parameters, which indicates the importance of H160 in the recognition of substrate39. Therefore, the intra-molecular communication pathways and functional sites between H160 and mutant residues were analyzed by the MCPath server. The apo-CK2α protein (PDB code 3AT2) was selected for analysis. The most popular pathways between the active site H160 and allosteric site using the MCPath method were mapped on the structure, as shown in Figure 3. Perturbation of the pathways could propagate, as visualized with spheres. The allosteric pathways of K75 and I78 to H160 are relatively short pathways, both passing through R80, D156, V157, and P159 (Figure 3B, 3C and 3F). The differences between the two pathways are the two residues with opposite charges, E180 in the K75-H160 allosteric pathway and K77 in the I82-H160 allosteric pathway. These results may correspond to the change in both the KM and kcat and may have the opposite influence on catalytic efficiencies. The allosteric pathways of I82 and P109 both generated 6 steps to H160 with two of the same residues, V162 and G177 (Figure 3D, 3E and 3F). The I82 and P109 mutants have nearly the same KM but different kcat values, which may be caused by the other four residues in the allosteric pathways. V31 generated an ensemble of pathways (10-steps-long) to H160 (Figure 3A and 3F), which the interaction transmits through the activation loop and then into the core of the domain. The whole pathway belongs to a loop that has good flexibility, which may exert the least influence after mutation. The results of these predicted allosteric pathways were consistent with the mutation analysis.

Figure 3
figure 3

The possible communication pathway between the active site His160 (colored in yellow) and the allosteric sites (colored in magenta) from CK2α (PDB code: 3AT2). The most populated pathways are shown in the sphere. The residues in the allosteric pathway are colored in orange. (A) The pathway from an ensemble of 12-step paths generated starting from V31. (B) The pathway from an ensemble of 7-step paths generated starting from K75. (C) The pathway from an ensemble of 7-step paths generated starting from I78. (D) The pathway from an ensemble of 8-step paths generated starting from I82. (E) The pathway from an ensemble of 8-step paths generated starting from P109. (F) The tables list the most populated pathway.

PowerPoint slide

Full size image

The conservation of the allosteric pocket on the CK2α surface in the CMGC kinase family

CK2α is a member of the CMGC family of eukaryotic protein kinases40. To determine the conservation of the residues in the newly discovered allosteric pocket of the whole CMGC kinase family, we aligned the protein sequences of 9 members of the CMGC kinase family, including cyclin dependent kinases (CDK), CDK-like kinases (CDKL), casein kinase 2 (CK2), glycogen synthase kinase(GSK), mitogen-activated protein kinase(MAPK), SR-rich protein kinase (SRPK), CDK-like kinase(CLK), dual specificity tyrosine regulated kinase (DYRK) and RCK41. A global score was calculated for all sequences. The similarity global score was defined as 0.7. The sequence alignment results of the ClustalX analysis of the CMGC family are shown in Figure 4. The predicted pocket consisted of 12 residues, with 3 residues (I78, I82 and P109) located on the bottom of the pockets and 9 residues (V31, W33, P72, V73, K74, K75, K102, R107 and T108) located on the side of the pockets. Thus, we assessed the conservation level of the amino acid residues in the predicted pocket of CK2α. The inspection of the amino acid sequences within this site showed that almost all residues were diverse, except for W33 and I82. V31, P72, V73, K74, K75, I78, K102, R107, T108 and P109 were not conserved throughout the whole CMGC family. Residue I82 showed considerable conservation in the CDK family, CDKL family and CLK family, and residue W33 showed considerable conservation in the MAPK family. In addition, none of these 12 residues showed conservation between CK2α and SRPK, DYRK and RCK kinases (data not shown). These unique residues could be good determinants for the design of specific inhibitors.

Figure 4
figure 4

Multiple sequence alignment of CK2α with the other CMGC kinases, CDK, CDKL, GSK, CLK and MAPK. The sequences were aligned using the ClustalX method. The figure was created with ESpript. The sequence of 30-110 of CK2α is shown. Residues with a conservation of greater than 70% are color-coded. Mutations are indicated at the corresponding positions. Residues that were not conserved (V31, W33, P72, V73, K74, K75, I78, K102, R107, T108 and P109) are shown in gray. A red character with a blue color scheme indicates similarity in a group (I82), and a red box with a white character indicates strict identity.

PowerPoint slide

Full size image

Discussion

There is currently increasing interest in identifying more specific next-generation CK2 inhibitors that do not directly compete with ATP and that exhibit different mechanisms of action. However, there are few known allosteric inhibitors of CK2. Allosteric pockets are being increasingly exploited for finding therapeutic molecules19. One way to find allosteric inhibitors is to locate a new allosteric pocket for which structure-based drug design can be used. We found a new allosteric pocket in CK2α by combining bioinformatics and biochemistry methods. Mutation of this pocket both activated and inhibited the activity of CK2α. Both the binding and the extent of activation or inhibition were highly dependent on the residues chosen.

It is intriguing that the mutation of a single allosteric site on a kinase, such as in pocket 2 on CK2α, can generate a continuum of agonistic and antagonistic effects. Previous studies have shown that even changing a single atom42 or a small group43 may result in opposite effects or large differences in efficacy. Furthermore, even the same allosteric ligand bound at the same site can lead to opposing effects in different environments44. As shown using MCPath, generating the predicted allosteric pathways may interrupt the opposing effects. Further experiments should be conducted to explain the results.

The pocket is not conserved among the whole CMGC kinase family, which suggests the potential for designing specific inhibitors. We should also avoid activating residues in this pocket, such as V31, K75, I82 and P109, when designing allosteric inhibitors based on this pocket.

Although we have not discovered new molecules that bind to this allosteric site, we note that the allosteric inhibitor discovered by Horn et al45. demonstrates that it is possible for small molecules to bind such hidden allosteric sites strongly enough to stabilize the open form of a pocket and to alter an enzyme's activity. In fact, Bowman et al found that almost any binding pocket could also serve as an allosteric site because of the coupling of many residues to different portions of the active site46. However, the drugability of different pockets is different. Given the central location of these residues within the allosteric pocket, new allosteric sites can be targeted with rational drug design or further studied to determine their biological relevance. In summary, this study yielded a new allosteric site in CK2α, which we believe will foster efforts to investigate its potential as a candidate pocket for subsequent drug design.