Discovery and Characterization of GSK2801, a Selective Chemical Probe for the Bromodomains BAZ2A and BAZ2B

Bromodomains are acetyl-lysine specific protein interaction domains that have recently emerged as a new target class for the development of inhibitors that modulate gene transcription. The two closely related bromodomain containing proteins BAZ2A and BAZ2B constitute the central scaffolding protein of the nucleolar remodeling complex (NoRC) that regulates the expression of noncoding RNAs. However, BAZ2 bromodomains have low predicted druggability and so far no selective inhibitors have been published. Here we report the development of GSK2801, a potent, selective and cell active acetyl-lysine competitive inhibitor of BAZ2A and BAZ2B bromodomains as well as the inactive control compound GSK8573. GSK2801 binds to BAZ2 bromodomains with dissociation constants (KD) of 136 and 257 nM for BAZ2B and BAZ2A, respectively. Crystal structures demonstrated a canonical acetyl-lysine competitive binding mode. Cellular activity was demonstrated using fluorescent recovery after photobleaching (FRAP) monitoring displacement of GFP-BAZ2A from acetylated chromatin. A pharmacokinetic study in mice showed that GSK2801 had reasonable in vivo exposure after oral dosing, with modest clearance and reasonable plasma stability. Thus, GSK2801 represents a versatile tool compound for cellular and in vivo studies to understand the role of BAZ2 bromodomains in chromatin biology.


■ INTRODUCTION
Bromodomain modules selectively recognize acetylated lysine (KAc) containing sequences, a post-translational modification that is highly abundant in histones and other nuclear proteins. 1−3 Interest in these so-called epigenetic reader proteins has risen as it has become clear that the interaction between the proteins and the acetylated histone tails can be inhibited with small molecules. A number of reported selective bromodomain inhibitors are drug-like molecules and have potential as therapeutics in a range of diseases, including cancer and inflammation. 4−7 Much of the work to date has focused on the BET family for which inhibitors have now entered clinical testing. 4,8,9 The BET subfamily of bromodomains (BRDT and BRD2−4) has proven to be remarkably druggable. Interestingly, some molecules in clinical development 10−12 have been reported to have BET bromodomain binding activity in addition to their previously known activity. The rapid exploration of the biology of BET bromodomains has been facilitated by the availability of potent and selective small molecule chemical tools that have been widely used to study chromatin biology. 13−20 With that in mind, we embarked on efforts to identify probes for bromodomains outside of the BET subfamily through a public private partnership coordinated by the Structural Genomics Consortium. Here we outline the collaborative discovery of GSK2801, a chemical probe for the bromodomain adjacent to zinc finger domain protein (BAZ) 2A/B. An alternative BAZ2A/B chemical probe that can be used as an additional independent validation tool has recently been published by our group in collaboration with an academic collaborator. 21 BAZ1A, BAZ1B, BAZ2A, and BAZ2B constitute a family of evolutionarily conserved and ubiquitously expressed proteins with conserved domain structure including PHD and bromodomain histone tail reader motifs. 22 BAZ2A is a component of the nucleolar remodeling complex (NoRC), a member of the "imitation switch chromatin remodeling complexes" (ISWI), 23 which play a role in regulating the expression of noncoding RNAs and in the formation of repressive heterochromatin in particular at centromeres and telomeres. 24,25 Association of BAZ2A with acetylated histone tails is mediated by its bromodomain. 26 Recently, high expression levels of BAZA have been reported in prostate cancer, and the protein may serve as a marker for metastatic potential. 27 Little is known about the BAZ2B subtype. However, single nucleotide polymorphisms (SNPs) in the BAZ2B gene locus have been associated with sudden cardiac death. High expression levels of BAZ2B have been associated with poor outcome of pediatric B cell acute lymphoblastic leukemia (B-ALL). 28 X-ray crystal structures of the BAZ2A/B bromodomains showed a shallow acetyl-lysine binding pocket, and they were predicted to be among the least druggable members of the protein family. 29 We therefore felt that identification of inhibitors for the BAZ2A/B bromodomains would be an interesting indicator of the chemical tractability of the broader protein family.  These illustrate that BRD2 appears to have a narrower channel that accommodates the pendant phenyl ring, suggesting this as an area to explore in order to enhance selectivity. (b) Chemical structure of compound 2 (left) and cocrystal structure of compound 2 with BAZ2B (right). The main interacting side chains are shown in ball and stick representation. Hydrogen bonds are shown as dotted lines. Conserved water molecules are depicted as spheres.

■ RESULTS AND DISCUSSION
A few weak binding acetyl-lysine mimetic fragments have been recently reported 30 in addition to a chemical tool compound, which showed an unusual aromatic π-stacking interaction of two aromatic moieties present in the inhibitor that efficiently filled out the binding pocket. 21 In our hands, screening of the BAZ2A bromodomain against a set of molecules that contained acetyllysine mimetics led to identification of a low molecular weight, chemically tractable indolizine, compound 1, which had been previously reported as an inhibitor of BET bromodomain subfamily proteins. Indolizine 1 had a potency of 1.5 μM in a BAZ2A AlphaScreen assay, which was comparable to its reported activity against the BET bromodomains ( Figure 1). 31,32 A crystal structure of compound 1 bound to BRD2 (PDB ID: 4A9I) revealed that the methyl ketone is serving as an acetyl lysine mimetic (Figure 2A). Compound 1 was modeled into the acetyl lysine binding site of BAZ2B ( Figure 2A). The channel that accommodates the indolizine R1 substituent appears wider in BAZ2B, suggesting that modification in this region will modulate selectivity. To facilitate the optimization of the indolizine hit, we cocrystallized an early representative compound (2) with BAZ2B ( Figure 2B). Compound 2 has a 2-pyridyl group at R1 and a morpholine substituent in the R2 position. As is the case for compound 1 bound to BRD2, 2 the methyl ketone of compound 2 functions as an acetyl-lysine mimetic. The carbonyl forms a hydrogen bond with the conserved asparagine N1944 and with a network of water molecules that are typically found at the bottom of the acetyl-lysine binding site. The indolizine ring forms a number of hydrophobic interactions, most notably to I1950 and V1893. The aromatic ring in position R1 forms an aromatic stacking interaction with W1887. The R2 morpholino group was oriented toward the solvent and formed no significant interactions with BAZ2B. Details describing protein purification, crystallization, and crystallography as well as in vitro screening methods are available in the Supporting Information.
To optimize the indolizine hit for activity on BAZ2A/BAZ2B and selectivity over the BET bromodomains, we explored substituents at the 7-position of the indolizine and modified the aryl group at the 1-position. Compounds were synthesized via two general routes. In route 1 (Scheme 1a), an appropriately parasubstituted pyridine A was N-alkylated with 1-chloropropan-2one. The resulting pyridinium salt B was then reacted in a dipolar cycloaddition reaction with an aryl alkyne and potassium carbonate in DMF at elevated temperatures in a microwave to give the desired ring system with the acetyl lysine mimetic methyl ketone present in the 3-position of the indolizine. In route 2 (Scheme 1b), the appropriate pyridinium salt B was reacted with methyl acrylate in the presence of triethylamine and manganese dioxide in toluene at 90°C for 1 h to give an indolizine C with a methyl ester in the 1-position, a methyl ketone in the 3-position, and the substituent that came from the starting pyridine in the 7-position. Saponification of the methyl ester with aqueous base, followed by bromination gave bromosubstituted indolizines of general structure D. In the final step, a Suzuki reaction was utilized to install the desired aryl group, R1, in the 1-position.
We first explored the structure−activity relationships (SARs) on the pendant phenyl group in the R1-position of the indolizine. In an AlphaScreen assay (Table 1), compounds 3−6 showed weaker activity at both BAZ2A and BAZ2B compared to 1. Replacing the pendant 2-pyridyl group with a meta-methoxy phenyl (compound 3) obliterated activity at both BAZ2A and BAZ2B. Compound 7, harboring an ortho-CH 2 OH, maintained activity at BAZ2A and gave a good increase in potency at BAZ2B, moving from 7 to 1 μM. Compound 12, which incorporates an ortho-methyl sulfone, improved potency on BAZ2A to 720 nM, and BAZ2B stayed in the low micromolar IC 50 range.
Based on the improved potency of the compounds on both BAZ2A and BAZ2B with the ortho-CH 2 OH compound 7 and the ortho-methyl sulfone 12, we explored substitution on the indolizine ring with these groups in place. In the context of the ortho-CH 2 OH substituted pendant phenyl (Table 2), a methyl

Journal of Medicinal Chemistry
Article group 13 and a methoxy group 14 improved BAZ2A activity relative to the unsubstituted parent, compound 7. Both the primary carboxamide substituted indolizine 16 and the phenoxy substituted compound 17 were weaker against BAZ2A. BAZ2B activity was improved in methyl substituted indolizine 13, methyl ether substituted indolizine 14, morpholine substituted indolizine 15, and phenyl ether substituted indolizine 17. With amide substituted compound 16, however, we saw a potency decrease on BAZ2A relative to unsubstituted compound 7.
Within a series of ortho-methyl sulfone phenyl analogues (Table 3), several indolizine substituents improved activity at both BAZ2A and BAZ2B relative to unsubstituted compound 12. The Boc-protected aminomethyl substituent 18 gave good potency increases. In the ortho-hydroxymethyl series described in Table 2, ethers off of the indolizine helped BAZ2B activity but in general were less helpful for BAZ2A activity. In the orthomethylsulfone context, however, the methyl ether 21, propyl ether 19, and phenyl ethers 20 enhanced both BAZ2A and BAZ2B activity.
Following identification of a handful of compounds with improved potencies, we sought to characterize their activity at other bromodomains. Table 4 shows that the two unsubstituted indolizines 7 and 12 were quite weak on BRD4 and BRD9. The Boc-protected aminomethyl compound 18, however, has low micromolar BRD9 activity and thus is not suitable as a BAZ2B/A probe. The ethers (19−21) were inactive or very weakly active against BRD4(1) and BRD9.
Modeling suggested that the improved selectivity of the bulkier ortho-substituted phenyls for BAZ2B over BET resulted from reduced planarity of these compounds with respect to ortho-H and 2-pyridyl analogues. The binding site of BRD2 is narrower than BAZ2B and binds compound 1 in essentially a planar conformation (Figure 2A). Docking of compounds 17 and 20 into the BRD2 site gave strained, high-energy poses consistent with their lower in vitro potency. In contrast, the more open site of BAZ2B was able to accommodate docked o-substituted phenyl indolizines with ease.
We cocrystallized compounds 17 and 20 with BAZ2B ( Figure 3). The structure confirmed that the introduction of the 2-hydroxyl-methyl group at the phenyl ring produced a 90°r otation of the phenyl ring incompatible with potent BET binding. In BAZ2B, the rotated phenyl ring makes an end-on π-stacking interaction with W1887 and orients the 2-hydoxylmethyl group to produce a hydrogen-bond with the ZA-loop backbone (N1894 amine). The methyl-sulfonyl substitution at position-2 of the phenyl ring produces an even more dramatic rotation to accommodate the more bulky sulfonyl group. In addition, the sulfonyl oxygen groups also make a pair of hydrogen bonds to the backbone NH of N1894.
Based on the potency and excellent selectivity over the BET bromodomains, we chose to look at compound 19, GSK2801, in more detail. We wanted to determine the potency of the compound in orthogonal assay systems. The interaction between the small molecule GSK2801 and the bromodomain was examined using biolayer interferometry (Octet-RED BLI) and isothermal titration calorimetry (ITC) (Figure 4a,b). Steady state fits of the measured dose response BLI data resulted in a dissociation constant of 60 nM (K D ). The inhibitor showed fast on-and off-rate binding (K on , 1.57 ± 0.02 × 10 5 1/(M s); K off , 6.95 ± 0.058 × 10 −3 1/s). ITC confirmed this value determining a K D of 136 and 257 nM for BAZ2B and BAZ2A and a stoichiometry of 1.0 ± 0.1, respectively. The cocrystal structure with GSK2801 showed that the observed binding mode of the probe is conserved when compared with the related compound 20 (Figure 4c).

Article
Next we assessed the wider selectivity of GSK2801 against other bromodomains in thermal stability assays. BAZ2A/B bromodomains are exceptionally stable interaction domains, a property that resulted in small temperature shifts in thermal stability assays even in the presence of tightly binding ligands. At an inhibitor concentration of 10 μM, GSK2801 resulted in a temperature shift of 4.1 and 2.7°C for BAZ2A and BAZ2B, respectively. Significant T m shifts were also observed for the second bromodomain in TAF1L(2) (3.4°C) and BRD9 (2.3°C) ( Figure 5a).
Because of the limited sensitivity of BAZ2A/B in the thermal stability assay we evaluated the selectivity of GSK2801 using an alternative assay format. Forty-two bromodomains modified with a Bir A biotin ligase targeting sequence were coexpressed in bacteria with the enzyme. This coexpression strategy resulted in homogeneously biotin-labeled bromodomains in vivo. 33 The successful labeling with biotin was verified using mass spectrometry (data not shown). We used the biotinylated proteins in BLI experiments probing the interaction of GSK2801 at two concentrations (0.2 and 1.0 μM) (Figure 5b). In agreement

Journal of Medicinal Chemistry
Article with the ΔT m data, BRD9 and TAF1(L) were detected as the major off-targets, while no other significant interactions were detected within the bromodomain family. We quantified the off-target interactions using ITC. These experiments showed that GSK2801 bound TAF1L(2) with an affinity K B of 0.31 (± 0.02) × 10 6 M −1 (K D : 3.2 μM) and a binding enthalpy change ΔH of −8.6 (±0.02) kcal/mol. ITC experiments using the bromodomain of BRD9 resulted in the determination of an affinity K B of 0.826 (±0.02) × 10 6 M −1 (K D : 1.1 μM) and ΔH of −9.8 (±0.01) kcal/mol. TAF1L is a retrotransposed gene only present in the primate lineage that has a very restricted expression to the testis compartment. 34 However, TAF1L has been reported to be mutated and expressed in a subset of melanoma cases. 35 BRD9 is expressed in most tissues and is an off-target of GSK2801.
Because of the off-target activity with BRD9, we developed a structurally highly related control compound that lacked activity on BAZ2A/B. Based on the low activity on BAZ2A and BAZ2B with a meta-methoxy substituted pendant phenyl (compound 3), we designed a compound that utilized this moiety in conjunction with the propyl ether in the indolizine 7-position. Using the established BLI cross-screening assay format, this compound (compound 23, GSK8573) proved to be inactive against BAZ2A/B and all other bromodomains except BRD9 ( Figure 6).

Journal of Medicinal Chemistry
Article Using ITC we determined an affinity for interaction of GSK8573 with BRD9 of 1.04 μM (K D ) associated with a favorable binding enthalpy change ΔH of −8.98 (± 0.152) kcal/mol. Thus, GSK8573 shared similar affinity for the off-target BRD9 while losing all activity against the rest of the bromodomain family. These results suggest that GSK8573 can be used as a structurally related negative control compound in biological experiments.
To test whether GSK2801 would also bind to endogenous BAZ2 proteins, we devised a chemoproteomic competition binding assay. The Boc-protecting group of compound 18 was removed to provide an amine functionalized analogue of GSK2801 that was immobilized on sepharose beads. The resulting affinity matrix was incubated with nuclear and chromatin enriched HuT78 extracts containing increasing concentrations of GSK2801 or the structurally related GSK8573. Bound proteins were eluted and quantified using an isobaric mass tagging strategy and targeted mass spectrometry. 36,37 Among the 18 endogenous full-length bromodomain proteins that bound to

Journal of Medicinal Chemistry
Article the matrix, only BAZ2A and BAZ2B displayed a dose-dependent reduction binding by GSK2801 suggesting a low micromolar potency (pK d = 5.5) (Figure 7a,b). Conversely, GSK8573 did not show any binding to BAZ2 proteins up to a concentration of 50 μM, but limited affinity to BRD9 (pK d = 4.5) and to SMARCA2 and 4 (pK d = 4.8), which is likely to be indirect due to the association of SMARCA2 and 4 to BRD9 in the BAF complex 38 (Figure7a).
In order to determine if GSK2801 could displace the BAZ2A/B from chromatin in cells we utilized FRAP experiments. Treatment of U2OS cells with the HDAC inhibitor SAHA induced hyperacetylated chromatin and a better activity window in the FRAP assay. SAHA-treated U2OS cells were transfected with a GFP-BAZ2A fusion construct. In parallel, U2OS cells were transfected with a GFP fusion with a BAZ2A mutant construct (N1873F) in which the conserved asparagine essential for recognizing the acetylated lysine had been mutated (Figure 8). GSK2801 accelerated FRAP half-recovery time to the same extent as observed for the mutant construct indicating that the compound was able to displace BAZ2A from chromatin. Conversely, the inactive GSK-8573 did not have any effect on the half-recovery time of GFP-BAZ2A.
In order to determine the suitability of GSK2801 for in vivo experiments we measured pharmacokinetic parameters after intraperitoneal and oral dosing to male CD1 mice. This pharmacokinetics study showed that GSK2801 has reasonable in vivo exposure after oral dosing, modest clearance, and reasonable plasma stability (Figure 9). These properties should allow GSK2801 to be used as a BAZ2A/B bromodomain inhibitor in vivo.

■ CONCLUSION
In this article, we started from an unselective micromolar BAZ/BET inhibitor and optimized it through iterative medicinal chemistry to a potent inhibitor of BAZ2A/B with >50-fold selectivity over BRD4. X-ray crystallography has provided a convincing rationalization of the steric features resulting in the BET selectivity. The resulting molecule, GSK2801, is a potent and selective ligand for the BAZ2A and BAZ2B bromodomains, suitable for use as a small molecule chemical probe. GSK2801 has excellent selectivity for BAZ2A and BAZ2B, with only low micromolar residual activity at BRD9 and TAF1L. The compound displaces full length GFP-BAZ2A from chromatin at concentrations below 1 μM. We have also identified a closely related inactive control compound to help inform cellular screening results. In order to accelerate discoveries on the roles of BAZ2B and BAZ2A in physiology and pathophysiology, we are sharing these compounds with the research community.
■ EXPERIMENTAL SECTION Chemistry. All solvents and commercial chemicals were reagent grade and used without purification. The purity of final compounds was determined by HPLC/MS and 1 H NMR and determined to be >95%. 1 H NMR spectra were recorded on Bruker BioSpin spectrometer operating at 400 mHz in CDCl 3 , methanol-d 4 , or DMSO-d 6 . The chemical shifts (δ) reported are given in parts per million (ppm), and the coupling constants (J) are in hertz (Hz). The spin multiplicities are reported as s = singlet, br s = broad singlet, d = doublet, t = triplet, q = quartet, m = multiplet, and br = broad.
The LC/MS analysis was performed on a Waters SQD with a Phenomenex Kinetex 1.7 μm XB-C18 column at 40°C, using water + 0.2% v/v formic acid and acetonitrile + 0.15% v/v formic acid as mobile phase, and performed on Waters Acquity BEH C18 2 × 50 mm 1.7 m column at 50°C, using water + 0.20% v/v formic acid, water + 0.20% v/v formic acid, and acetonitrile + 0.15% v/v formic acid. Microwave irradiated reactions were carried out in sealed glass vessels in a Biotage Initiator. Flash chromatography purifications were carried out on a Biotage SP1 instrument.  73 g, 204 mmol). The mixture was stirred at 90°C for 1 h. The solid was filtered off over Celite and washed with acetone. The combined filtrate was concentrated and purified by silica gel chromatography eluting with EtOAc in hexanes (5 to 50%) to give methyl 3-acetyl-7methoxyindolizine-1-carboxylate (4.4 g, 17.80 mmol, 69.8% yield) as a yellow solid.
Synthesis of Compound 19 (GSK2801) Intermediate C. To a solution of methyl 3-acetyl-7-methoxyindolizine-1-carboxylate (4.4 g, 17.80 mmol) in methanol (60 mL) was added NaOH (44.5 mL, 178 mmol). The mixture was stirred at 80°C for 30 min. The solution was concentrated and acidified with 6 N HCl to pH = 1. The solid product was filtered, washed with water, and dried to give 3-acetyl-7methoxyindolizine-1-carboxylic acid (2.4 g, 10.29 mmol, 57.8% yield) as a yellow solid. The product was used in the next step without further purification.

Journal of Medicinal Chemistry
Article cool to room temperature, treated with water, and extracted with EtOAc. The organics were washed with water and brine and then dried over Na 2 SO 4 . The filtrate was concentrated and dried to give 4-propoxypyridine (8.3 g, 60.5 mmol, 91% yield) as a yellow oil.
Synthesis of Compound 19 (GSK2801) Intermediate H. To a solution of 4-propoxypyridine (8.2 g, 59.8 mmol) in ethyl acetate (20 mL) was added 1-chloropropan-2-one (14.28 mL, 179 mmol). The mixture was stirred at ambient temperature overnight. The precipitated solid was filtered, washed with diethyl ether, and dried to give the indicated product quantitatively as a colorless solid.
Synthesis of Compound 19 (GSK2801) Intermediate K. To a solution of 3-acetyl-7-propoxyindolizine-1-carboxylic acid (0.95 g, 3.64 mmol) in DMF (15 mL) was added sodium bicarbonate (0.916 g, 10.91 mmol) and then NBS (0.712 g, 4.00 mmol) portionwise at 0°C. The cooling bath was removed, and the mixture was allowed to stir at ambient temperature for 1 h. The solution was treated with water and extracted with EtOAc. The combined organics were washed with water, then brine, and dried over Na 2 SO 4 and filtered. The filtrate was concentrated to dryness to give 1-(1-bromo-7-propox-yindolizin-3-yl)ethanone (1.05 g, 3.55 mmol, 98% yield) as an off-white solid. The product was used in the next step without further purification.

Journal of Medicinal Chemistry
Article