Docking and Structure Activity Relationship Studies of Potent and Selective Thiazolidinethione GSK-3 Inhibitors

Abstract

Glycogen synthase kinase-3 (GSK-3) plays a key role in several biochemical pathways and is an attractive target for pharmacological intervention. We prepared a series of analogs of a highly selective thiazolidinethione inhibitor of GSK-3. The structure-activity relationship indicated a precise structural requirement for potent inhibition. We used docking and bioinformatic analysis to explore the rationale for the potency and selectivity of this class of GSK-3 inhibitors. These computational studies identified residues unique to GSK-3 likely to play a role in ligand-specific induced fit interactions. Together, these studies highlight the structural stringency of specific kinase inhibition that can be achieved for GSK-3.

Graphical Abstract

Glycogen synthase kinase-3 (GSK-3) is a protein-serine kinase initially identified for its role in glucose metabolism and later found to play a role in a variety of cellular processes.^{1, 2} GSK-3 has α- and β-isoforms, with the β-isoform largely implicated in most pathological conditions. GSK-3 has been identified as a target for the treatment of Alzheimer’s disease,³ cancer,⁴ and diabetes.⁵ One of the first reported inhibitors of GSK was lithium, which directly inhibits GSK-3 by competing with magnesium ions.⁶ Indeed other metals such as zinc, copper and mercury have also been shown to be potent inhibitors of GSK-3.⁷ A variety of natural products and analogs have been reported as inhibitors of GSK-3.⁸ These include schisandrin B,⁹ the indirubin class of compounds,¹⁰ the iridoid glycosides (e.g. loganin),^{11 12} and hymenialdisine.¹³

By far the largest class of GSK-3 inhibitors are synthetic compounds.^{14, 15} The thiadiazolidinedione, tideglusib is a member of a well studied class of synthetic compounds.^16–19 Other select classes of synthetic GSK-3 inhibitors include, pyrazolopyrimidine,²⁰ maleimides,²¹ pyridines,²² and aminofurazans.²³ A few of these compounds including 9-ING-41²⁴, LY-2090314,²⁵ and tideglusib²⁶ have undergone or are currently in clinical trials.

We previously reported the thiazolidinethione compounds COB-187 and COB-152 as inhibitors of GSK-3²⁷ and reported on the pharmacological activity of the structurally related phenylmethimazole derivative COB-141 as an inhibitor of IL-6 expression in triple negative breast cancer cells.²⁸ Here we report the synthesis of analogues of COB-187 and COB-152 and their structure activity relationship for GSK-3 inhibition as well as docking to GSK-3.

Compounds COB-187 (3c) and COB-152 (3b) and analogues were prepared as shown in Scheme 1. The reaction of an amine (1) with a phenacyl bromide (2) provides hydroxylthiazolidinone 3 in generally good yield.²⁹ This compound was then treated with acid to provide the elimination product 4. The analogues of 3 and 4 were evaluated in a series of pharmacological assays for GSK-3 inhibition.

Scheme 1.

A series of analogues of 3 were prepared where R² remained constant as a simple unsubstituted phenyl ring and R¹ was varied. When R¹ was a phenyl (3a), or substituted phenyl (4-Cl, 4-NO₂, 4-OCH₃) no activity was observed (defined as an IC₅₀ >10μM). Similarly aliphatic chains PhCH₂, Ph(CH₂)₂, nC₅H₁₁) showed no activity (See Supplementary Material for a list of inactive compounds). A series of heterocyclic derivatives were also prepared and assayed. Of these the 3-pyridyl (3b) and 4-pyridyl (3c) showed excellent activity (Table 1). Compound 3b showed an IC₅₀ of 106 nM for GSK-3α and an IC₅₀ value of 253 nM for GSK-3β. Compound 3c showed an IC₅₀ of 18 nM for GSK-3α and an IC₅₀ value of 12 nM for GSK-3β. Interestingly, the 2-pyridyl derivative (3d) showed no activity. Other heterocyclic derivatives such as 5-indole or 2-benzimidazle again showed no activity. Given the excellent activity of 3b and 3c, a pyrimidine analog was prepared in order to mimic the 3-pyridyl, however only limited activity was observed. A 4-N-methylpiperidine analog was also prepared in order to generate a more basic compound than the active 4-pyridine. Again no activity was observed (Supplementary Material). In addition to the heterocyclic derivatives at R¹, a small group of substituted phenyl derivatives were prepared at R² when R¹ was either 3-pyridyl or 4-pyridyl. These compounds showed very similar activity to the parent compounds, 3b and 3c, with no significant improvements in activity and only minor decreases in activity (Supplementary Material). Overall, the structure activity relationship of this series of compounds implicates a stringent, precise structural requirement in binding GSK-3.

Table 1.

GSK Inhibition of Compounds 3a - 3d

Compound #	Tested Kinase	% Inhibitiona	Errorb	IC50 (nM)
3a	GSK-3α	16.4	2.8	>10,000
	GSK-3β.	24.1	0.9	>10,000
3b	GSK-3α	94.3	0.9	106
	GSK-3β.	93.9	0.6	253
3c	GSK-3α	94.9	0.3	18.2
	GSK-3β.	99.0	1.0	12.2
3d	GSK-3α	27.4	7.1	>10,000
	GSK-3β.	38.3	3.5	>10,000

^aTested at 10μM, average of 2 data points;

^bReported as standard deviation

To explore the importance of the 4-OH group, two additional compounds 4a and 4b were prepared that were similar to the previously reported COB-141.²⁸ Both compounds were prepared by acid-catalyzed elimination of 3b and 3c respectively. Even at 10 μM, neither of these compounds showed greater than 40% inhibition of either GSK-3α or GSK-3β. Clearly unsaturation of the thiazolidine ring provides no improvement in activity. This significant reduction in activity in the dehydrated analogues of 3b and 3c likely indicates that this 4-OH is directly involved in binding GSK-3 or the stereo-electronic features of that tertiary alcohol are important for binding.

Given the structure activity relationship results discussed above, we sought to use computational docking to identify an explanation for the kinase specificity and potency of 3b-3d. While there are many crystal structures of GSK-3β and intense drug discovery interest, there are no reports of potent, highly selective inhibitors derived from de novo docking studies. One report comes close, but the successful docking prediction was derived from a crystal structure of an initial lead compound rather than a structurally unrelated compound.³⁰ Recently, a large compound screening, coupled with pharmacophore modeling and molecular dynamics identified several compounds with low micromolar IC₅₀ values, but no assessment of specificity for GSK-3β compared to other kinases was reported.³¹ Consequently, we first examined the existing GSK-3β crystal structures to determine which would be best to use. We focused our docking on the ATP binding region since our previous studies indicated that one of the residues in this region, Cys199, was involved in the mechanism of inhibition.³²

To determine the best PDB structure of GSK-3β to use for the docking, we analyzed the extent of residue position differences in 35 crystal structures compared to a reference PDB structure. Initially, we superimposed the backbone of non-phosphorylated GSK-3β PDB structures using Maestro (Schrödinger) (Supplementary Material,Table S4) to that of PDB ID 1PYX.³³ Residue position differences were then evaluated by calculating only the amino acid RMSD differences (AA RMSD In Place) for select amino acids surrounding the ATP binding region. The ATP binding region consists of residues in the glycine rich loop; the hinge region; the catalytic site and the interface between the ATP binding site, and; the substrate binding site (sites 1 and 2 respectively).³⁴ We chose 1PYX as the reference PDB since the ligand in this PDB (adenyl imidodiphosphate (ANP)) is a close analog of the natural ATP substrate As shown in Figure 1, some GSK-3 residues had a wide range of amino acid structure RMSD variability compared to that observed for other residues. Residues with a greater range of variability compared to 1PYX may have a higher propensity for ligand-specific induced-fit interactions.

Figure 1.

Comparison of GSK-3β crystal structures reveals variable structure positions and unique outliers a) Active site region of GSK-3β 1PYX with co-crystalized ligand ANP in thick grey tube and residues analyzed in structural comparison studies colored by hydrophobic (green), acidic (red), basic (blue), Gly (pink), Cys (yellow), pro (orange), Asn/Gln (pale blue) and Thr (pale red). b) Violin plot of per-residue RMS differences for amino acid residues in GSK-3β PDBs compared to1PYX.

Given the extent of variability amongst the different GSK-3β crystal structures, we chose to dock our ligands to PDB ID 1PYX in a manner that explored possible induced-fit. In addition, since there are water-mediated ligand-receptor interactions in 1PYX, we also explored the effect of explicit water molecules. We docked the 4-, 3-, 2-pyridyl compounds 3c, 3b, 3d with Glide³⁵ and the OPLS2005 forcefield, using three different docking protocols (Supplementary Material), including a docking workflow to explore induced-fit involving water interactions. Recent computational studies of a GSK3-inhibitor complex have highlighted the importance of and sensitivity to slight adjustments to the location of a water residue.³⁶

The only docking where 3c was the best binder (i.e., correlated with GSK-3β inhibition data) was with water 616 and R-3c (Figure 2) using the docking workflow that avoided premature exclusion of poses prior to induced-fit MMGBSA calculations (Strategy 3, Supplementary Material). Ligand-GSK-3β contacts in this induced-fit structure include hydrogen bonds (thione to Cys199 thiol and 4-pyridyl N to Val135 NH), stacking contacts (phenyl to Arg 141 guanidinium group) and hydrophobic contacts (phenyl and the hydrophobic region at the base of the ATP binding region). In addition, the hydroxyl group of R-3c is positioned to form a hydrogen bond with the reoriented (induced-fit) 616 water bridge to ILE62. This water staples together an extended hydrogen-bonding network between Arg141, 3c and Ile62. The contacts between GSK-3 residues and the hydroxyl and pyridyl group contacts in the docked structure are consistent with the abolished activity in analogs lacking these groups (4a/4b and 3a).

Figure 2.

3c docked to GSK-3β a) Induced-fit binding mode to GSK-3β, b) Cys199 comparison to PDB ID 1H8F and c) Arg141 comparison to PDB ID 4ACG.

There were several notable conformational changes in this induced-fit docking compared to 1PYX. While the Cys199 conformation differs significantly from that in 1PYX, it is remarkably similar to that found in GSK-3 crystal structure 1H8F³⁷ (χ₁ of 62.5° vs. 61.2° in 1H8F vs. −41.7 in 1PYX, Figure 2b). In the amino acid conformer comparison across GSK3β PDBs, there was little variability for Cys199 except for 1H8F³⁷ which is the one outlier observed in the violin plot of Cys199 (Figure 1b). Having a direct Cys199-ligand binding interaction that requires a conformational change of Cys199 is consistent with the previous studies³² indicating a Cys199-dependent, slow, induced-fit binding mode of inhibition.

Another significant conformational change compared to 1PYX is with Arg141, particularly the χ₄ angle (113.5° vs. −156.3° in 1PYX, Supplementary Material, Figure S1a). This conformation still retains the salt bridge with Glu137 while also forming favorable cation-pi stacking contacts with the phenyl ring of 3c (Figure 2a). The amino acid conformer comparison (1.78 Å RMSD) is comparable to the mid-quartiles for Arg141 of all GSK3β PDBs (Figure 1b). Interestingly, the orientation of Arg141 in the docked structure is very similar to that in 4ACG³⁸ (RMSD 1.95 Å,Figure 2c ), where 4ACG is a representative example of other GSK3β PDBs with comparable conformations of Arg141 (Supplementary Material, Figure S2). Finally, the docked conformation of 3c is very similar to the starting low-energy conformer (0.5563 Å RMSD, Supplementary Material, Figure S1b). Binding of the lowest energy conformer would provide an entropic advantage for binding that could contribute to the low nanomolar binding affinity of 3c. Overall, the observed similarities of the induced-fit conformation of Cys199 and Arg141 to existing GSK3β crystal structures supports the structural relevance of this docking pose. Given the relatively small size of 3c, it is notable that the docking contacts with Val135, Cys199 and Arg141 span the entire ATP binding plus hinge region.

The 3c docked binding interactions, especially with Arg141 (Figures 2 and 3) are also consistent with previous kinase specificity data. In a broad kinase screen, the only kinases significantly inhibited by 3c were GSK-3β and GSK-3α.²⁷ Additional studies indicated that 3c was highly specific for GSK-3β and GSK-3α (>90% inhibition at 2 μM) with the inhibition dependent on Cys199.³² GSK-3β belongs to the Group 4 kinases containing a conserved CDFG sequence near the active site.³⁹ Of all these Group 4 kinases and the other cysteine-containing kinases tested previously,³² only GSK-3 contains both Cys199 and Arg141 (Figure 4), residues involved in direct contacts with 3c in the docked structure. In addition, Pro136, located in the hinge region, forms the floor of the side-tunnel (Figure 3) and is unique to only GSK-3. The only other kinase moderately inhibited by 3c was MAPK5/PRAK (60% inhibition at 2μM).³² MAPK5/PRAK has a Phe at the position corresponding to Arg141 that could potentially also form stacking interactions with the phenyl of 3c. While several of the other kinases tested have a lysine at a position comparable to Arg141 in GSK-3β (Figure 4), none were inhibited significantly by 3c (<30% inhibition at 2 μM).³² This is consistent with the GSK-3β – 3c docked structure (Figure 2). Previous detailed computational studies have shown that arginine forms stronger stacking interactions with pi aromatics than lysine does, especially in a hydrophilic environment.⁴⁰ Furthermore, the side-tunnel formed by Arg141 and Pro136 (Figure 3) might contribute to the observed limited R² ring substituent variability accommodated in the structure activity relationship of GSK-3β inhibition. As a result of 1) the unique correlation of Arg141 to kinase specific inhibition, 2) the stacking contacts with 3c, and 3) the role it plays in forming the side tunnel in the docked structure, we are referring to Arg141 as a “Sentry” residue to highlight its possible significance in conferring GSK-3 kinase specificity.

Figure 3.

Side tunnel formed by Arg141 in GSK-3/3c docked structure.

Figure 4.

Clustal Alignment of CDFGkinases previously assayed³² for inhibitory activity by 3c. a) conserved CDFG region b) region that includes kinase gatekeeper and hinge residues and residue 141 (ARG “sentry” residue in GSK3).

Overall, the unique sequence determinants of GSK-3 and the docking results are consistent with 3c kinase specificity, isoform selectivity and potency. In addition, we also identified GSK-3 residues likely to play a role in ligand-specific induced fit interactions and kinase specificity. The correlation between the observed stringent structure-activity relationship profile with unique sequence determinants of GSK-3 highlight future avenues drug development.

Highlight.

1. A series of GSK-3 inhibitors were prepared and screened for their GSK-3 inhibitory activity. A preliminary structure activity relationship was observed

2. Docking studies were carried out to identify key components of the highly specific interactions of the inhibitors with GSK-3