Inhibitors of Glycogen Synthase Kinase 3 with Exquisite Kinome-Wide

Jo , J., Whitcomb , D. J., Olsen , K. M., Kerrigan , T. L., Lo , S. C., Bru-Mercier , G., ..... Pan , J. Q., Lewis , M. C., Ketterman , J. K., Clore ,...
0 downloads 0 Views 7MB Size
Articles pubs.acs.org/acschemicalbiology

Inhibitors of Glycogen Synthase Kinase 3 with Exquisite KinomeWide Selectivity and Their Functional Effects Florence F. Wagner,*,† Joshua A. Bishop,‡ Jennifer P. Gale,† Xi Shi,† Michelle Walk,† Joshua Ketterman,† Debasis Patnaik,‡ Doug Barker,† Deepika Walpita,§ Arthur J. Campbell,† Shannon Nguyen,† Michael Lewis,† Linda Ross,⊥ Michel Weïwer,† W. Frank An,§ Andrew R. Germain,§ Partha P. Nag,§ Shailesh Metkar,§ Taner Kaya,† Sivaraman Dandapani,§ David E. Olson,† Anne-Laure Barbe,† Fanny Lazzaro,† Joshua R. Sacher,† Jaime H. Cheah,§ David Fei,∥ Jose Perez,∥ Benito Munoz,§ Michelle Palmer,∥ Kimberly Stegmaier,⊥ Stuart L. Schreiber,§ Edward Scolnick,† Yan-Ling Zhang,† Stephen J. Haggarty,†,‡ Edward B. Holson,† and Jen Q. Pan*,† †

Stanley Center for Psychiatric Research, §Center for Science of Therapeutics, ∥Therapeutics Platform, Broad Institute of MIT/Harvard, Cambridge, Massachusetts 02142, United States ‡ Chemical Neurobiology Laboratory, Departments of Neurology & Psychiatry, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts 02215, United States ⊥ Department of Pediatric Oncology, Dana-Farber Cancer Institute and Boston Children’s Hospital, Boston, Massachusetts 02215, United States S Supporting Information *

ABSTRACT: The mood stabilizer lithium, the first-line treatment for bipolar disorder, is hypothesized to exert its effects through direct inhibition of glycogen synthase kinase 3 (GSK3) and indirectly by increasing GSK3’s inhibitory serine phosphorylation. GSK3 comprises two highly similar paralogs, GSK3α and GSK3β, which are key regulatory kinases in the canonical Wnt pathway. GSK3 stands as a nodal target within this pathway and is an attractive therapeutic target for multiple indications. Despite being an active field of research for the past 20 years, many GSK3 inhibitors demonstrate either poor to moderate selectivity versus the broader human kinome or physicochemical properties unsuitable for use in in vitro systems or in vivo models. A nonconventional analysis of data from a GSK3β inhibitor high-throughput screening campaign, which excluded known GSK3 inhibitor chemotypes, led to the discovery of a novel pyrazolo-tetrahydroquinolinone scaffold with unparalleled kinome-wide selectivity for the GSK3 kinases. Taking advantage of an uncommon tridentate interaction with the hinge region of GSK3, we developed highly selective and potent GSK3 inhibitors, BRD1652 and BRD0209, which demonstrated in vivo efficacy in a dopaminergic signaling paradigm modeling mood-related disorders. These new chemical probes open the way for exclusive analyses of the function of GSK3 kinases in multiple signaling pathways involved in many prevalent disorders.

I

synthase).5 However, the specific proximal substrates for each paralog remain largely unknown and in some cases are misassigned to one or the other paralog through the use of nonselective pharmacological tools. Precisely how GSK3 maintains its pathway specificity at the crossroads of so many cellular processes is still unclear. GSK3 activity is modulated by phosphorylation at two distinct sites. Inhibitory phosphorylation at Ser21/9 of GSK3α/ β, by upstream kinases such as AKT, inactivates the kinase as the phosphorylated N-terminal tail serves as a pseudosubstrate, blocking incoming substrates and their access to the catalytic center.6 Conversely, autophosphorylation at Tyr279/216 is correlated with increased GSK3 activity.7,8 Constitutively active

nitially identified as a key regulator of insulin-dependent glycogen synthesis, glycogen synthase kinase 3 (GSK3) was subsequently shown to function as a master regulator of multiple signaling pathways including insulin signaling, neurotrophic factor signaling, Wnt signaling, neurotransmitter signaling, and microtubule dynamics.1 In mammals, GSK3 is encoded by two genes, GSK3A and GSK3B, which produce two highly homologous protein-serine/threonine kinases ubiquitously expressed in all tissues. Despite their high degree of structural similarity (67% overall amino acid identity and 98% identity in the ATP-binding site2) and functional overlap, the GSK3α and GSK3β paralogs appear to also have distinct functions.3,4 Known phosphorylation substrates for GSK3 include an array of proteins involved in transcription (c-Jun, NFAT, CREB), translation initiation (eIF2B), cell cycle (cyclin D1), Wnt pathway (Axin, APC, β-catenin), microtubule dynamics (CRMP, Tau), and glucose metabolism (glycogen © 2016 American Chemical Society

Received: April 4, 2016 Accepted: April 29, 2016 Published: April 29, 2016 1952

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 1. Existing GSK3 inhibitors and their kinome selectivities. (a) Representative GSK3 inhibitor chemotypes. A mobility shift microfluidic assay (Caliper, MA) was used to determine IC50 values for GSK3α and GSK3β. (b) Each inhibitor was screened against 402 kinases at 10 μM concentration (KINOMEscan); kinases with >50% inhibition are depicted (percentage inhibition is depicted by the size of the red dot). (c) X-ray crystal structure of AMP-PNP bound to hGSK3β (PDB ID: 1PYX). The different regions of the ATP-binding site are highlighted. (d) X-ray crystal structure of CHIR99021 bound to hGSK3β (3.10 Å resolution; PDB ID: 5HLN; Xtal BioStructures, Inc., MA). Dashed pink lines indicate hydrogen bonds between the adenine ring or CHIR99021 and the hinge region.

GSK3α or GSK3β may normalize the GSK3 signaling dysfunction in mood disorders.

GSK3 often serves as a negative regulator of downstream substrates, and an increase in GSK3 kinase activity often results in the attenuation of cellular signaling.9 Hyperactivity of both GSK3 paralogs has been implicated in the pathophysiology of a number of human disorders,10,11 including non-insulin-dependent diabetes mellitus,12 cardiac hypertrophy,13,14 cancer, and neurological or neurodevelopmental disorders such as Alzheimer’s disease,15 bipolar disorder,16 depression,17,18 and Fragile X syndrome.19−22 Additional human clinical and mouse genetic evidence as well as pharmacological studies support the notion that GSK3 inhibition may attenuate certain symptoms of psychiatric disorders. Patients with bipolar disorder show significant decrease in Ser9 GSK3β phosphorylation (enhanced activity) in peripheral blood mononuclear cells.16 Lithium, the first-line treatment for bipolar disorder, inhibits GSK3 kinase activity both directly (Ki = 2.0−3.5 mM),23 via competition with magnesium,24,25 and indirectly by increasing inhibitory phosphorylation of GSK3 at Ser21 of GSK3α and Ser9 of GSK3β.16,26−28 Furthermore, GSK3α null or GSK3β haploinsufficient mice phenocopy lithium’s effect in modulating behaviors modeling psychiatric symptoms.29,30 Conversely, mice overexpressing GSK3β or carrying mutations preventing inhibitory phosphorylation of GSK3α (Ser21) or GSK3β (Ser9) exhibit psychiatric-like behavior.16 Taken together, these data suggest that selective and potent inhibition of either



RESULTS AND DISCUSSION During the mid-1990s, as the role of the GSK3 kinases in insulin signal transduction and regulation of Tau protein phosphorylation emerged, the pharmaceutical industry and academia developed a plethora of highly potent GSK3 inhibitors, across a number of different chemotypes. As reviewed,9,31 small molecule GSK3 inhibitors include the natural product-based indirubins (e.g., 6-bromoindirubin-3′oxime, BIO), paullones (e.g., alsterpaullone), bisindolylmaleimides (e.g., SB-216763), and indolocarbazoles (e.g., staurosporine) and the synthetically derived aminopyrimidines (e.g., CHIR99021), ruthenium complexes (e.g., complex 532), thiadiazolidinones (e.g., TDZD-8), and imino-thiadiazoles (e.g., VP1.1433). Except for allosteric inhibitor TDZD-8, most of these ATP-competitive binders of GSK3 display poor to moderate selectivity when tested against the larger human kinome at a concentration comparable to cellular exposure (nonbiased conditions34). The biochemical selectivity of a representative set of GSK3 inhibitors (Figure 1a) was assessed against a panel of 402 kinases (Figure 1b; KINOMEscan, DiscoveRx at 10 μM concentration, Table S3). We found that, in addition to their GSK3 inhibitory activity, SB-216763, AR-A014418, and CHIR99021 inhibit 37, 9, and 18 kinases, respectively (number of non-mutant kinase with % 1953

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 2. HTS hit and its kinome selectivity. (a) BRD4003 kinome selectivity against 311 kinases at 10 μM. (Carna Biosciences; top 10 kinases shown). (b) IC50 values of individual enantiomers for GSK3 and CK1 kinases (Carna Biosciences).

ctrl ≤ 50). Moreover, none of these inhibitors demonstrate paralog selectivity between GSK3α and GSK3β (Figure 1a). These GSK3 inhibitors, exemplified by CHIR99021 (Figure 1a), adopt an “ATP-like” binding mode (Figure 1c,d) through the formation of hydrogen bonds to key “hinge” residues (backbone Val135-NH and Asp133-CO for AMP-PNP, Figure 1c; backbone Val135-NH and Val135-CO for CHIR99021, Figure 1d). CHIR99021, as well as most other GSK3 inhibitors, contains hydrophobic aromatic rings (high sp2 content), which form favorable interactions with hydrophobic regions I and II and the ribose region or DFG motif (Figure 1c,d). Consequently, the vast majority of these inhibitors tend to be relatively large (molecular weight > 500 g·mol−1) hydrophobic molecules with a high number of hydrogen-bond donors (≥2) and therefore possess poor blood−brain barrier properties (e.g., active P-glycoprotein efflux), limiting their translational ability to in vivo models of CNS diseases.35,36 Recently, two novel series of GSK3 inhibitors with good brain penetration and a suitable pharmacokinetic profile for use in vivo were reported. The carboxamide oxazole core37 was characterized by remarkable brain penetration (B:P ∼ 3); however, its selectivity against the broader kinome was not characterized (only 17 kinases tested). The nicotinamide scaffold38 reported by Luo et al. demonstrated excellent kinome selectivity, good in vitro permeability, and low efflux ratio (ER < 1), but it suffered from rapid clearance in rodents. Notable exceptions to ATP-competitive GSK3 inhibitors are thiadiazolidinones, which have been extensively used as pharmacological tools to study the roles of GSK3, including TDZD-8 and NP-12, two allosteric inhibitors of GSK3. However, NP-12 was reported to bind GSK3 irreversibly, suggesting a potential covalent interaction.39 To this day, the development of selective GSK3 inhibitors remains a dynamic research field. Existing pharmacological probes of GSK3, such as CHIR99021, have been used extensively to probe the biological function(s) of the GSK3 kinases both in vitro and in vivo.40−42 However, the largely uncharacterized and/or poor specificity of these inhibitors limit the ability to conclude whether the observed effects are due to the inhibition of GSK3 or other unintended targets. As the role

of GSK3 is being further investigated in multiple disorders, there is a clear need for specific GSK3 inhibitors to facilitate a less ambiguous probing of these enzymes’ multiple and critical functions. Herein, we report the discovery of a novel pyrazolotetrahydroquinolinone scaffold with high kinome-wide selectivity, a structural rationale and conceptual framework for the observed selectivity, and, lastly, the biochemical, cellular, and in vivo evaluations of this unique class of GSK3 inhibitors. Identification of a Highly Selective GSK3 Inhibitor via a Nonconventional High-Throughput Screen Analysis. In order to identify novel and selective inhibitors of GSK3, we developed a homogeneous biochemical assay using affinitypurified recombinant human GSK3β (GST tagged for purification) incubated with a peptide substrate and ATP (see Methods, ADP-Glo assay). In this assay, a luminescent signal, produced by the luciferase reaction, is proportional to the ADP produced during the kinase reaction with and without compound treatment and is correlated with GSK3 kinase activity. By detecting ADP production, this assay measures GSK3β activity at varied ATP concentrations up to 1 mM, allowing for the identification of ortho- and allosteric inhibitors. The Km values for ATP (13 μM) and the peptide substrate (1.33 μM) for the GSK3β fusion protein were determined using this assay and were consistent with previous reports.43,44 The assay parameters, including the concentrations of the enzyme, substrate peptide, and DTT as well as the incubation time, were optimized for signal linearity in 1536-well plates. A high-throughput screen of approximately 320 000 compounds (NIH, MLPCN library; http://mli.nih.gov/mli/ mlpcn/) was performed using the ADP-Glo kinase assay. Compounds were tested at 10 μM in a 1536-well plate format at 25 μM ATP concentration, slightly higher than the Km of ATP (13 μM), in order to capture both ATP-competitive and -noncompetitive inhibitors. The screen featured a Z′ > 0.8 at 10 μM GW8510, a nonselective kinase inhibitor. 45 Upon completion of the GSK3 inhibitor screen, we performed a search against known luciferase inhibitors in the same library by a similar assay (ATP-Glo assay) to rule out false positives,46−48 resulting in approximately 960 compounds with >25% inhibition of GSK3β at the test concentration. These actives, 1954

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 3. Structure−activity relationship. (a, b) Representative synthetic schemes of pyrazolo-dihydropyridinones IVa and IVb. (c) IC50 values for the inhibition of GSK3α and GSK3β in a mobility shift microfluidic assay (Caliper, MA) measuring the phosphorylation of a synthetic substrate. Values are the average of at least three experiments. Data are shown as IC50 values in μM ± standard deviation. Compounds were tested using a 12point dose curve with 3-fold serial dilution starting from 33 μM.

along with approximately 1200 closely related analogues in the library, were selected and tested using three different ATP concentrations (1, 25, and 250 μM) in a secondary screen to determine their half-maximal inhibitory concentration (IC50) and evaluate their ATP-dependent inhibitory properties to identify allosteric inhibitors. As a surrogate for broader kinase selectivity, a counter screen against cyclin D kinase 5 (CDK5), a highly homologous neuronal kinase, was implemented as part of our screening cascade. By prioritizing compounds based on potency, ATP-independent binding activity, and selectivity (against CDK5), compounds were selected and evaluated in multiple biochemical and cellular assays. All compounds selected were found to be ATP-competitive. Furthermore, such a prioritization scheme identified only compounds with poor kinome selectivity (data not shown), usually from already known GSK3 inhibitor scaffolds.

In order to discover compounds with kinome-wide selectivity, we then chose to reanalyze the full set of ∼2160 compounds in the secondary screen and prioritize based on selectivity against CDK5 (more than 5-fold) rather than emphasizing potency toward GSK3β. To capture more hits, we include compounds with relatively weak inhibitory activity toward GSK3β (IC50 < 10 μM). Finally, to further eliminate promiscuous kinase binders, we computationally filtered any structures resembling (75% similarity) known GSK3 kinase inhibitors based on a Bemis−Murcko substructure analysis49 of 81 known GSK3 inhibitors (Table S4). This computational filtering selected 13 compounds (Table S5) with biochemical GSK3β inhibitory activities ranging from low nanomolar to micromolar potencies with approximately equal inhibition of GSK3α. To evaluate their GSK3-dependent cellular activity, the compounds were screened in a TCF/LEF reporter assay using neuroblastoma SH-SY5Y cells. These cells stably express a 1955

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 4. Ligand-bound crystal structure. (a) Co-crystal structure of BRD3937 bound to hGSK3β (2.45 Å resolution; PDB ID: 5HLP; Xtal BioStructures Inc., MA). Hydrogen bonds are shown as pink dashed lines. (b) Molecular surface view of hGSK3β bound to BRD3937.

heated at 150 °C in a microwave in the presence of triethylamine for 15 min. Desired compounds IVa were obtained in moderate to high yields. When benzaldehyde IIa was replaced by acetophenone IIb, the yield suffered, and a two-step reaction had to be devised (Figures 3b and 5a). An electrophilic aromatic substitution under acidic conditions between aminopyrazole I and acetophenone IIb generated intermediate V, which was then reacted with dimedone III to obtain the desired quaternary tetrahydroquinolinone ring in IVb upon vinylogous amide formation and cyclization. Because all ATP-competitive inhibitors of GSK3 establish 1− 3 hydrogen bonds with the backbone atoms of Asp133 and/or Val135 in the hinge region,15 with classical donor−acceptor− donor motifs, we hypothesized that our scaffold most likely utilized one or more of the pyrazole (ring A) nitrogen(s) and tetrahydroquinolinone ring (ring B) nitrogen for the affinity to GSK3. We therefore decided not to alter the electronics of the pyrazole ring via substitution at R1 and kept this moiety constant throughout our initial SAR exploration (Figure 3c). IC50 values for GSK3α and GSK3β inhibition were measured in a mobility shift microfluidic assay (Caliper, MA) measuring the phosphorylation of a synthetic substrate, GSp-1 (see Methods), in order to capture potential bias between the two paralogs. We first probed whether the phenyl ring was necessary by either eliminating it entirely, compound 3, or replacing it with a cyclohexyl ring, compound 4. While the methyl in compound 3 resulted in only a 2-fold decrease in potency, the sp3-rich cyclohexyl ring in 4 was not tolerated in the ATP-binding pocket of either GSK3 kinase. Therefore, different substitutions of the phenyl ring were investigated next in order to probe the optimal electronic and steric requirements. Electron-deficient 2-pyridyl (5) and 3-pyridyl (6) were welltolerated but did not lead to increase activity. The 4-pyridyl (7), however, led to a greater than 10-fold loss of potency at both GSK3α and GSK3β, presumably due to disfavored hydrophobic interactions in the ATP-binding site of the GSK3 kinase. Next, the electron-withdrawing o-fluoro substitution in compound 8 increased the affinity for GSK3α to 18 nM (10-fold) and to a lesser degree to GSK3β (87 nM, 2.5fold). Consequently, compound 8 displayed moderate selectivity for GSK3α (4-fold). This slight paralog bias was unique to the ortho-fluorinated compound and did not translate to other ortho substitutions (Me in 11, OMe in 12−13, and CF3 in 14). The electron-withdrawing m-fluoro substitution in compound 9 led to a moderate increase in affinity to both paralogs (62 and 156 nM, respectively), whereas the p-fluoro substitution in compound 10 was less tolerated. While the electronics of the phenyl ring does not appear to impact potency (cf. compound 8 vs 12), o-substitution led to

firefly luciferase reporter gene that is induced, upon GSK3 inhibition, by accumulated nuclear β-catenin binding to a promoter containing 12 copies of a TCF/LEF binding sequence.50,51 Three of the 13 hits were active in the TCF/LEF reporter assay. One of these, pyrazolo-tetrahydroquinolinone compound BRD4003, a racemic mixture (Figure 2), emerged as a moderate inhibitor of GSK3β kinase activity (IC50 = 294 nM) with relatively weak cellular activity (10.4% TCF/LEF activation at 25 μM). Finally, to test whether our computational exclusion of known kinase motifs improved broader kinome selectivity, we profiled BRD4003 against a panel of 311 kinases at a concentration of 10 μM (Carna Biosciences; Table S6). BRD4003 demonstrated impressive kinome selectivity, inhibiting only three kinases other than GSK3α and GSK3β at greater than 50%: CK1α, CK1β, and CK1δ (Figure 2). By comparison, one of the most selective GSK3 inhibitors known, CHIR99021, inhibited 18 kinases by >50% at 10 μM in the same profiling assay (Carna Biosciences; data not shown). The enantiomers of BRD4003 were separated, and the dose−response inhibition of these five kinases was determined. The absolute stereochemistry of each enantiomer was determined indirectly via a high-resolution hGSK3β co-crystal structure of the closely related analogue. The (S) enantiomer, BRD4963, inhibits GSK3α and GSK3β with exquisite selectivity versus the CK1 kinases (>90-fold selectivity against the CK1’s). In contrast, the (R) enantiomer (ent-1) weakly inhibits GSK3α and GSK3β (IC50 = 4.77 and 10.18 μM, respectively) with 10-fold greater potency toward CK1δ (IC50 = 0.426 μM). Next, we evaluated whether the inhibition in an in vitro biochemical assay using recombinant enzyme translated into cellular activity by examining phosphorylation changes in direct substrates of GSK3. Upon treatment of neuroblastoma SHSY5Y cells with BRD4963, we observed a dose-dependent decrease in Tau phosphorylation with an IC50 of 11.5 μM. On the basis of this cellular activity, we chose to exploit this novel GSK3 inhibitor chemotype with unprecedented kinome selectivity. A series of analogues was designed to establish the structure−activity relationship (SAR) and increase potency to facilitate an X-ray structure in order to elucidate and exploit the binding of the compound to maintain the kinome selectivity and improve potency. Optimization Using Structure-Based Drug Design. The synthetically tractable scaffold of pyrazolo-tetrahydroquinolinone BRD4963 lends itself to rapid introduction of chemical diversity on the phenyl ring via a one-step multicomponent reaction (Figure 3a). An ethanolic solution of aminopyrazole I, benzaldehyde IIa, and dimedone III was 1956

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 5. Characterization of exquisitely selective GSK3 inhibitors. (a) IC50 values for inhibition of GSK3α and GSK3β in a microfluidic assay measuring the phosphorylation of a synthetic peptide substrate. Values are the average of at least three experiments. Data are shown as IC50 values in μM ± standard deviation. Compounds were tested using a 12-point dose curve with 3-fold serial dilution starting from 33 μM; EC50 values for the inhibition of Tau phosphorylation were obtained using an ELISA assay. (b) Kinome profile of BRD1652 (311 kinases at a 10 μM concentration, Carna Biosciences) and IC50 against all kinases inhibited at greater than 50% in panel. (c) Percent substrate conversion over time in a reversibility assay for BRD0209 (100-fold dilution; 2 nM final concentration) at 1 nM GSK3α and 0.1 nM GSK3β. The red circles represent compound dilution, and the blue circles represent DMSO dilution. Right panel: Inhibition curve of BRD0209 at GSK3α and GSK3β.

significant increases in potency (BRD3937 and compound 14), whereas p-substitutions were unproductive. With improved potency and sufficient solubility in BRD3937, we were poised for an X-ray study to understand the binding mode of this novel scaffold, rationalize its exceptional kinome selectivity, and inform analogue design. A high- resolution (2.45 Å) co-crystal structure of BRD3937 with human GSK3β (hGSK3β; Figure 4) was obtained. Consistent with previous X-ray crystal structures of hGSK3β (41 unique structures, Protein Data Bank), the GSK3 kinase domain contains a small N-lobe consisting primarily of β-sheets connected by a short hinge domain (in yellow) to a larger Clobe consisting mainly of α-helices (Figure 4). Confirming our biochemical data, BRD3937, a type-I inhibitor, binds in the ATP-binding domain between the N- and C-terminal lobes. BRD3937 is nestled between the hydrophobic residues (Figure 4b) that define the ATP-binding domain and contacts the hinge

residues via hydrogen-bond interactions. Unlike other small molecule GSK3 inhibitors, BRD3937 is anchored to the hinge domain via a unique tridentate hydrogen-bond interaction (pink dotted lines, Figure 4) through the hydrogen-bond donor−acceptor−donor motif embedded within the constrained tricyclic pyrazolo-tetrahydroquinolinone structure. The first H-bond donor interaction, through the pyrazoloNH-, engages the backbone carbonyl of Asp133, whereas the pyrazolo-N- forms an H-bond with the backbone N−H of Val135. A second H-bond donor interaction through the dihydropyridyl-NH- donor engages the backbone carbonyl of Val135. Additionally, the carbonyl oxygen of BRD3937 forms a hydrogen bond with a conserved water bridging the small molecule to the protein at Ile62. The tridentate interaction to the hinge is found in only two other crystal structures of GSK3β with nonselective indolinones (PDB IDs: 1Q41 and 3SAY).52 While these H-bond interactions are important in 1957

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 6. Characterization of cellular activities. (a) Western blot analysis of GSK3 and its substrates in SH-SY5Y cells treated with BRD0209 and BRD1652 (10 μM) for 30 min and 24 h. p-GSK3 indicates phosphorylated GSK3α-Tyr279 together with GSKβ-Tyr216. t-GSK3 indicates total GSK3 proteins (both α and β). (b) Quantification of fold changes in phosphorylation compared to vehicle (DMSO)-treated controls in each signal detected from western blots. Phospho and total protein levels were normalized to the GAPDH signal on the same blot. (c) DU-145 cells immunostained for β-catenin (green), Ki67 (red), and DAPI (blue) following 24 h treatment with vehicle (DMSO) and BRD0209. BRD0209 induced β-catenin stabilization (in green) and nuclear translocation compared to DMSO-treated controls.

enhancing affinity toward GSK3, we believe that they are necessary but not sufficient to provide broader kinome selectivity.15 We hypothesize that the exquisite kinome selectivity observed for this scaffold arises from the unique tridentate H-bond interactions anchoring the highly rigid (low entropy) pyrazolo-tetrahydroquinolinone system within the ATP-binding domain coupled with the orthogonal projection of the phenyl ring (81° planar angle, Schrödinger) relative to the tetrahydroquinolinone ring B system (Figure 3). The highly conformationally constrained nature of the pyrazolotetrahydroquinolinone tricyclic core, the distinct tridentate hinge binding mode, and the unique 3D steric requirements preclude productive engagement across the broader kinome within this highly conserved region. Informed by this crystal structure, further structure-based design (Figure 5a) led us to replace the methyl pyrazole with electron-withdrawing CF3 in order to strengthen the hydrogen bond between the pyrazole-NH and the Val135 backbone carbonyl in BRD1172. The potency for either paralog increased 5−10-fold, to 3 and 10 nM, respectively. Despite its excellent potency, BRD1172 was not a suitable chemical probe for cellular or in vivo assays due to its poor

chemical, plasma, and microsomal stabilities (Table S7). We therefore focused our effort to improve the metabolic stability of this class of compounds. The benzylic center of the dihydropyridine ring is prone to rapid oxidative aromatization, leading to an inactive pyrazolopyridine scaffold, exemplified by compound 15 (Figure 5a, GSK3β IC50 > 33.3 μM). In order to prevent the oxidation of the central dihydropyridine ring B, a quaternary methyl was installed by replacing benzaldehyde with acetophenone in our synthetic scheme (Figure 3b). Remarkably, the affinity of compound 16 for both GSK3α and GSK3β increased >40-fold to 8 and 26 nM, respectively, compared to that of BRD4963. The CF3 pyrazole in BRD1652 introduced another 10-fold increase in potency to single-digit nanomolar activity on both paralogs. Due to the electron-withdrawing effect of CF3, the nucleophilicity of the starting aminopyrazole was decreased and the overall chemical yield suffered (0.5% yield over two steps; Figure 3b). The weakly electron-donating cyclopropyl group in BRD0209 (GSK3α IC50 = 19 nM; GSK3β IC50 = 5 nM) retains the desired potency while dramatically improving reaction efficiency (65% yield; see the Supporting Information). 1958

DOI: 10.1021/acschembio.6b00306 ACS Chem. Biol. 2016, 11, 1952−1963

Articles

ACS Chemical Biology

Figure 7. In vivo characterization. (a) Pharmacokinetics and brain distribution of BRD1652 (30 mg/kg) and CHIR99021 (12.5 mg/kg) following a single intraperitoneal (i.p.) dose in male C57BL/6 mice. (b) Corresponding pharmacokinetic parameters for BRD1652 and CHIR99021. (c) Amphetamine-induced hyperactivity behavioral assay: distance traveled across 5 min bins was collapsed into (d) a total activity measure and analyzed via one-way analysis of variance (ANOVA) with Tukey’s LSD post hoc analysis (p1mg/kg,Veh = 0.28; p3mg/kg,Veh =