9H-Purine Scaffold Reveals Induced-Fit Pocket Plasticity of the BRD9 Bromodomain

The 2-amine-9H-purine scaffold was identified as a weak bromodomain template and was developed via iterative structure based design into a potent nanomolar ligand for the bromodomain of human BRD9 with small residual micromolar affinity toward the bromodomain of BRD4. Binding of the lead compound 11 to the bromodomain of BRD9 results in an unprecedented rearrangement of residues forming the acetyllysine recognition site, affecting plasticity of the protein in an induced-fit pocket. The compound does not exhibit any cytotoxic effect in HEK293 cells and displaces the BRD9 bromodomain from chromatin in bioluminescence proximity assays without affecting the BRD4/histone complex. The 2-amine-9H-purine scaffold represents a novel template that can be further modified to yield highly potent and selective tool compounds to interrogate the biological role of BRD9 in diverse cellular systems.


■ INTRODUCTION
The term "epigenome" describes the array of chemical modifications on DNA and histone proteins that dynamically regulate gene expression. 1 The complex set of reversible posttranslational modifications (PTMs) that decorate histones ultimately determines the overall state of chromatin, giving rise to the so-called "histone code", a cellular language involving specific proteins that introduce (writers), remove (erasers), or recognize (readers) PTMs. 2 ε-N-acetylation of lysine residues is a widespread post-translational mark, 3 deposited on proteins throughout the entire proteome by lysine acetyl transferases (KATs), removed by lysine deacetylases (KDACs), and recognized by evolutionary conserved protein modules such as bromodomains 4 (named after the Drosophila melanogaster brahma gene 5 ) as well as the more recently discovered YEATS domains. 6 The human proteome encodes 61 bromodomains (BRDs) found on 42 diverse proteins, including histone acetyl transferases (HATs) and HAT-associated proteins such as GCN5, PCAF, and bromodomain 9 (BRD9), 7,8 transcriptional coactivators such as the TAF and TRIM/TIF proteins, 9,10 ATPdependent chromatin remodeling complexes such as BAZ1B, 11 helicases such as SMARCA, 12 nuclear scaffolding proteins such as the polybromo PB1 13 as well as transcriptional regulators, such as the bromo and extra terminal (BET) proteins. 14 The family of human BRD modules has almost been completely structurally characterized resulting in a recent classification of eight distinct structural subfamilies. 15 The BET subfamily of BRDs has attracted a lot of attention, as its members (BRD2, BRD3, BRD4, and the testis specific BRDT) play a central role in cell cycle progression, cellular proliferation, and apoptosis. 16 BETs contain two N-terminal BRD modules that interact with acetylated histones, 15 transcription factors 17,18 or other acetylated transcriptional regulators, 19,20 an extra terminal (ET) recruitment domain, 21 and a C-terminal motif (in the case of BRD4 and BRDT) responsible for the recruitment of the positive transcription elongation factor B (P-TEFb), 22 ultimately controlling transcriptional elongation. Importantly, BET BRDs have been successfully targeted by small molecule inhibitors, such as the triazolothienodiazepine (+)-JQ1 23 and the triazolobenzodiazepine IBET762, 24 which were identified employing phenotypic  Alternative binding of compound 1 into the bromodomain of BRD4(1) with the 6-chloro substituent adopting an acetyllysine mimetic pose. The compound and protein are colored as in (C). PDB code 4MEN was used as the starting model for docking. (E) Docking of compound 2a positions the 2-amine-9H-purine ring system of the ligand within the Kac cavity, sterically packing between V87 and I146, suggesting that this scaffold topology can be further utilized to target bromodomains. The same color scheme as in (C) and (D) is used with the ligand shown as orange sticks. PDB code 4MEN was used as the starting model for docking. (F) Fragments were tested in a thermal shift assay against bromodomains of the BET subfamily as well as representative members from other families. Thermal shifts are color-coded as indicated in the inset. 9H-Purines showed weak binding across the panel.
screening and have in the past few years consolidated the emerging role of BRDs as viable therapeutic targets. Indeed, BET inhibition suppresses tumor growth in diverse mouse models of cancer, such as NUT midline carcinoma, acute myeloid and mixed lineage leukemia, multiple myeloma, glioblastoma, melanoma, Burkitt's lymphoma, neuroblastoma and prostate cancer, leading to a number of clinical trials seeking to modulate BET function in diverse tumor settings. 25 The initial success targeting BET proteins and the availability of robust recombinant systems of expression as well as biophysical assays to probe BET−ligand interactions have spawned a number of medicinal chemistry efforts seeking to identify novel scaffolds that can block binding of acetylated lysines to these protein interaction modules. Phenotypic screening, molecular docking, and fragment-based approaches have emerged as successful tools for discovering other Kac mimetics, leading to the identification of a number of new chemotypes, including 3,4-dimethylisoxazoles, 26,27 3-methyl-3,4-dihydroquinazolinones, 28 indolizinethanones, N-phenylacetamides, and N-acety-2-methyltetrahydroquinolines 29 triazolopyrimidines, methylquinoline, and chloropyridones, 30 thiazolidinones, 31 4-acylpyrroles, 32 and triazolophtalazines 33 (summerized in Chart 1). Starting from fragment hits, highly potent and selective BET inhibitors have also emerged, 34,35 suggesting that it is possible to access new chemical space for inhibitor development via initial fragment screening. Druggability analysis based on the available structural information 36 suggests that it is possible to target most BRD structural classes. 15 In fact, BRDs annotated as "hard" to target, such as BAZ2B, have recently been successfully targeted by highly selective and potent inhibitors 37 and fragment based approaches have now yielded chemotypes targeting BRDs outside the BET family, including CREBBP/p300, 38 ATAD2, 39,40 BAZ2B, 41 and BRPF1. 42 An unexpected but interesting finding that recently attracted attention was the identification of clinically approved kinase inhibitors and tool compounds that exhibited binding with high affinity, and selectivity across the human BRD family, to BET BRDs. 43 Crystal structures with the first bromodomain of BRD4 revealed acetyllysine mimetic binding of the PLK inhibitor BI-2536 and the JAK inhibitor fedratinib without any significant distortion of the inhibitors when compared to kinase complexes, suggesting the potential to develop polypharmacology targeting both BRD and kinases at the same time. 43 Interestingly, the cyclin-dependent kinase inhibitor dinaciclib was also identified as a binder of BRD4, 44 suggesting that other inhibitor classes might be good starting points for developing inhibitors for BRDs.
In light of the successful fragment-based programs and the reliability for discovering BRDs inhibitors, we started to investigate the purine scaffold as a putative Kac mimetic. Purine is a privileged chemical template, as it is one of the most abundant N-based heterocycles in nature, 45 and a number of purine-based drugs are currently approved and used for the treatment of cancer (6-mercaptopurine, 6-thioguanine), viral infections such as AIDS and herpes (carbovir, abacavir, acyclovir, ganciclovir), hairy cell leukemia (cladribine), and organ rejection (azathioprine). 46 Moreover, purine based compounds have emerged as reliable chemical−biology tools, since they can interact with a variety of biological targets involved in a number of diseases. Some such examples include their activity as microtubule (myoseverin), 90-heat shock protein (PU3), sulfotransferase (NG38), adenosine receptor (KW-6002), and cyclin-dependent kinase (olomoucine, Figure  1A; roscovitine) inhibitors. 47,48 Computational analyses followed by in vitro evaluation of purine-based fragments identified this scaffold as a novel effective Kac mimetic. Interestingly, initial purine fragment hits also demonstrated affinity outside the BET subfamily of BRDs, toward the less explored BRDs of PB1, CREBBP, and BRD9. To our knowledge, the only known BRD9 inhibitors today show cross-reactivity toward BET BRDs and CREBBP employing a triazolophthalazine template. 33 The precise biological function of BRD9 remains elusive, although it has been identified as a component of the SWI/SNF complex 49 and has been associated with a number of different cancer types, including non-small-cell lung cancer, 50 cervical, 51 and hepatocellular carcinoma. 52 Notably, its BRD reader module has been frequently found mutated in lung squamous cell carcinoma, prostate adenocarcinoma, and uterine corpus endometrial carcinoma. 53− 55 We chose a small purine fragment, compound 2a (Chart 2), from our fragment hits for structural optimization and found that some of its 6-aryl derivatives exhibited nanomolar affinity toward BRD9, with lower activity toward BRD4. Importantly, the developed inhibitors induced an appreciable change in the three-dimensional shape of the receptor binding cavity, indicating that their binding occurs through an "induced-fit" mechanism. 56 Previously an induced fit binding was observed when a dihydroquinoxalinone was shown to insert under an arginine residue of the CREBBP BRD, resulting in restructuring of the Kac binding site of this bromodomain. 57 In our case the rearrangement of the binding site of BRD9 was more extensive, with several side chains rotating and shifting to accommodate the small purine ligand. The optimized compound 11 (Chart 2) exhibited nanomolar affinity for BRD9 with weaker micromolar affinity for BRD4 in vitro, displaced the BRD9 BRD from chromatin without affecting BRD4 binding to histones, and did not show any cytotoxicity toward human embryonic kidney cells. Our work establishes the proof-of-concept of using 2amino-9H-purines as a starting point to develop Kac mimetic compounds targeting BRDs outside the BET family, with compound 11 representing a promising low nanomolar starting point toward the discovery of selective BRD9 ligands to be used as chemical probes for deep biological and pharmacological investigation of BRD9.  61,62 were calculated on the BRD4(1) and BRD9 structures, considering the centroid of the cocrystallized ligand (N,5-dimethyl-N-(4-methylbenzyl) [1,2,4]triazolo[1,5-a]pyrimidin-7-amine for BRD4(1), from PDB code 4MEN), or cocrystallized ligand 7d (from the BRD9/7d complex), or Tyr106 (for BRD9, PDB code 3HME) for grid generation. In all cases, cubic inner boxes with dimensions of 10 Å were applied to the proteins, and outer boxes were automatically detected. Ring conformations of the investigated compounds were sampled using an energy window of 2.5 kcal/mol; conformations featuring nonplanar conformations of amide bonds were penalized. Side chains of residues close to the docking outputs (within 8.0 Å of ligand poses) were reoriented using Prime (version 3.4, Schrodinger, LLC, New York, NY, 2013), and ligands were redocked into their corresponding low energy protein structures (Glide Standard Precision Mode), considering inner boxes dimensions of 5.0 Å (outer boxes automatically detected), with resulting complexes ranked according to GlideScore.
Rigid Docking. Calculations were performed using the Glide software package (SP mode, version 6.1, Schrodinger package) 58−60 in order to determine the binding mode of compound 11 into the acetyllysine cavity of BRD9. First the receptor grid was generated focused on the BRD9 binding site taking as a reference structure the experimentally determined complex structure of compound 7d with BRD9, with inner-and outer-box dimensions of 10 × 10 × 10 and 21.71 × 21.71 × 21.71, respectively. Standard Precision (SP) Glide mode was employed accounting for compound flexibility in the sampling of compound 11. The sampling step was set to expanded sampling mode (4 times), keeping 10 000 ligand poses for the initial phase of docking, followed by 800 ligand poses selection for energy minimization. A maximum number of 50 output structures were saved for each ligand, with a scaling factor of 0.8 related to van der Waals radii with a partial charge cutoff of 0.15. A postdocking optimization of the obtained docking outputs was performed, accounting for a maximum of 50 poses based on a 0.5 kcal/mol rejection cutoff for the obtained minimized poses.
General Synthetic Information. All commercially available starting materials were purchased from Sigma-Aldrich and were used as received. Solvents used for the synthesis were of HPLC grade and were purchased from Sigma-Aldrich or Carlo Erba Reagenti. NMR spectra were recorded on Bruker Avance 600 or 300 MHz instruments. Compounds were dissolved in 0.5 mL of MeOD, CDCl 3 , or DMSOd 6 . Coupling constants (J) are reported in hertz, and chemical shifts are expressed in parts per million (ppm) on the δ scale relative to the solvent peak as internal reference. 13 C spectra are reported for representative compounds. Electrospray mass spectrometry (ESI-MS) was performed on a LCQ DECA TermoQuest (San Jose, CA, USA) mass spectrometer. Chemical reactions were monitored on silica gel 60 F 254 plates (Merck), and spots were visualized under UV light. Analytical and semipreparative reversed-phase HPLC were performed on an Agilent Technologies 1200 series high performance liquid chromatography system using Jupiter Proteo C 18 reversed-phase columns ((a) 250 mm × 4.60 mm, 4 μm, 90 Å, flow rate = 1 mL/min; (b) 250 mm × 10.00 mm, 10 μm, 90 Å, flow rate = 4 mL/min respectively, Phenomenex). The binary solvent system (A/B) was as follows: 0.1% TFA in water (A) and 0.1% TFA in CH 3 CN (B). Absorbance was detected at 240 nm. The purity of all tested compound (>98%) was determined by HPLC analysis. Microwave irradiation reactions were carried out in a dedicated CEM-Discover focused microwave synthesis apparatus, operating with continuous irradiation power from 0 to 300 W utilizing the standard absorbance level of 300 W maximum power. Reactions were carried out in 10 mL sealed microwave glass vials. The Discover system also included controllable ramp time, hold time (reaction time), and uniform stirring. After the irradiation period, reaction vessels were cooled rapidly (60−120 s) to ambient temperature by air jet cooling. Compounds synthesized are shown in Chart 2.
General Procedure a for the Suzuki−Miyaura Cross-Coupling of Free Halopurines (3a,b, 3d, 3f−h, 4a−d, 5a, 6a− c, 7a−e, 8a, 9a,b, 10, 11). 2-Amino-6-bromopurine (50.0 mg, 0.23 mmol), commercially available boronic acids (0.29 mmol), Pd(OAc) 2 (2.70 mg, 0.012 mmol), P(C 6 H 4 SO 3 Na) 3 (34.0 mg, 0.06 mmol), and Cs 2 CO 3 (228.0 mg, 0.70 mmol) were added to a 10 mL microwave vial equipped with a magnetic stirrer. The vial was evacuated and backfilled with nitrogen three times. Degassed acetonitrile (0.5 mL) and degassed water (1.0 mL) were added by means of an airtight syringe. The mixture was heated under microwave irradiation at 150°C for 5−15 min. After irradiation, the vial was cooled to ambient temperature by air jet cooling and a mixture of cold water and HCl (1.5 M) was added (5.0 and 2.0 mL, respectively). The mixture was subsequently poured into crushed ice and then left at 4°C overnight. The resulting precipitate was filtered and purified by HPLC to give the desired product in good yields (53−90%). HPLC purification was performed by semipreparative reversed-phase HPLC using the gradient conditions reported below for each compound. The final products were obtained with high purity (>95%) detected by HPLC analysis and were fully characterized by ESI-MS and NMR spectroscopy.
General Procedure b for TBAF-Assisted N9-Alkylation of Purine Rings (2b, 3c, 3e, 4e, 5b, 8b, 8c). 2-Amino-6-arylpurine (0.1 mmol) was dissolved in 0.4 mL of THF at room temperature. To this mixture 0.2 mL (0.2 mmol) of TBAF (1.0 M solution in THF, Aldrich) and iodomethane (12.5 μL, 0.2 mmol) or chloroacetone (16.0 μL, 0.2 mmol) were added. The reaction was stirred at room temperature for 10 min. Water was added, and the aqueous layer was extracted three times with dichloromethane. The combined organic layers were washed with water, dried with anhydrous Na 2 SO 4 and concentrated under vacuum. The crude mixture was purified by semipreparative reversed-phase HPLC using the gradient conditions reported below for each compound. Compounds were obtained in good yields (50−88%) and high purity (>98%) and were fully characterized by ESI-MS and NMR spectroscopy.
General Procedure c for the Synthesis of 2-Hydroxyl-6arylpurines (8d−f). A three-necked flask was charged with the 2amino-6-arylpurine derivative (7b−d) (0.5 mmol) and 50% H 2 SO 4 (2.0 mL). The mixture was stirred at room temperature for 30 min and then cooled to −5°C. A solution of NaNO 2 (48.3 mg, 0.7 mmol) in H 2 O (200 μL) was added dropwise, and the release of nitrogen gas was immediately observed. The reaction mixture was then stirred at −10°C for 2 h, and urea (24.0 mg, 0.4 mmol) was added to decompose the excess of NaNO 2 . The mixture was then stirred at 50°C for 1 h and neutralized with 50% NaOH solution, diluted with water, and extracted three times with EtOAc. The combined organic layers were dried with anhydrous Na 2 SO 4 and concentrated under vacuum. The crude mixture was purified by semipreparative reversedphase HPLC to get the pure products in good yields (47−63%). Compounds were fully characterized by ESI-MS and NMR spectroscopy.
Isothermal Titration Calorimetry. Experiments were carried out on an ITC200 titration microcalorimeter from MicroCal, LLC (GE Healthcare) equipped with a Washing module, with a cell volume of 0.2003 mL, and a 40 μL microsyringe. Experiments were carried out at 15°C while stirring at 1000 rpm, in ITC buffer (50 mM HEPES, pH 7.5 (at 25°C), 150 mM NaCl). The microsyringe was loaded with a solution of the protein sample (300−740 μM protein for BRD9 and BRD4(1), respectively, in ITC buffer) and was carefully inserted into the calorimetric cell which was filled with an amount of the ligand (0.2 mL, 20−25 μM in ITC buffer). The system was first allowed to equilibrate until the cell temperature reached 15°C, and an additional delay of 60 s was applied. All titrations were conducted using an initial control injection of 0.3 μL followed by 38 identical injections of 1 μL with a duration of 2 s (per injection) and a spacing of 120 s between injections. The titration experiments were designed in such a fashion as to ensure complete saturation of the proteins before the final injection. The heat of dilution for the proteins was independent of their concentration and corresponded to the heat observed from the last injection, following saturation of ligand binding, thus facilitating the estimation of the baseline of each titration from the last injection. The collected data were corrected for protein heats of dilution (measured on separate experiments by titrating the proteins into ITC buffer) and deconvoluted using the MicroCal Origin software supplied with the instrument to yield enthalpies of binding (ΔH) and binding constants (K B ) in the same fashion as that previously described in detail by Wiseman and co-workers. 66 Tables 1, 2, and 3.
Bioluminescence Resonance Energy Transfer Assay (BRET). HEK293 cells (8 × 10 5 ) were plated in each well of a 6-well plate and co-transfected with histone H3.3-HaloTag (NM_002107) and Nano-Luc-BRD9 (Q9H8M2-BRD amino acids 120−240) or NanoLuc-BRD4 full-length (060885). Twenty hours post-transfection 2 × 10 4 cells were trypsinized, washed with PBS, and exchanged into media containing phenol red-free DMEM, 10% FBS in the absence (control sample), or the presence (experimental sample) of 100 nM NanoBRET 618 fluorescent ligand (Promega). Cell density was adjusted to 2 × 10 5 cells/mL and then replated in a 96-well assay white plate (Corning Costar no. 3917). At the time of replating compound 7d or 11 were added at a final concentration spanning from 0.005 to 33 μM and the plates were incubated for 18 h at 37°C in the presence of 5% CO 2 . NanoBRET substrate (Promega) was added to both control and experimental samples at a final concentration of 10 μM. Readings were performed within 5 min using the CLARIOstar reader (BMG Lifetechnologies) equipped with 450/80 nm bandpass and 610 nm long-pass filters. A corrected BRET ratio was calculated and is defined as the ratio of the emission at 610 nm/450 nm for experimental samples (i.e., those treated with NanoBRET fluorescent ligand) subtracted by and the emission at 610 nm/450 nm for control samples (not treated with NanoBRET fluorescent). BRET ratios are expressed as milliBRET units (mBU), where 1 mBU corresponds to the corrected BRET ratio multiplied by 1000.
Confocal Imaging. HEK293 cells were transfected with histone H3.3-HaloTag using FuGENE HD (Promega). Twenty-four hours post-transfection cells were labeled with 5 μM HaloTag TMR ligand (Promega) in complete medium (DMEM and 10% FBS) for 15 min at 37°C and 5% CO 2 . Medium containing HaloTag-TMR ligand was then washed twice with fresh complete medium. Cells were placed back at 37°C and 5% CO 2 for 30 min and then imaged. Images were acquired on an Olympus Fluoview FV500 confocal microscope (Olympus, Center Valley, PA, USA) containing a 37°C and CO 2 environmental chamber (Solent Scientific Ltd., Segensworth, U.K.) using appropriate filter sets.
In Vitro Cytotoxicity. Following NanoBRET substrate addition and NanoBRET measurements, to the same wells an equal volume of CellTiter-Glo reagent (Promega) was added and the plates were incubated for 30 min at room temperature. Total luminescence was measured, and relative compound toxicity was determined by comparing the RLUs (relative light units) of a sample containing DMSO vehicle (in the absence of compound 11) to the RLUs of the samples containing 0.005−33 μM compound 11.
Crystallization. Aliquots of the purified proteins were set up for crystallization using a mosquito crystallization robot (TTP Labtech, Royston, U.K.). Coarse screens were typically set up onto Greiner 3well plates using three different drop ratios of precipitant to protein per condition (100 + 50 nL, 75 + 75 nL, and 50 + 100 nL). Initial hits were optimized further scaling up the drop sizes. All crystallizations were carried out using the sitting drop vapor diffusion method at 4°C.   ). Initial phases were calculated by molecular replacement with PHASER 71 using the known models of BRD9 and BRD4(1) (PDB codes 3HME and 2OSS, respectively). Initial models were built by ARP/wARP 72 followed by manual building in COOT. 73 Refinement was carried out in REFMAC5. 74 In all cases thermal motions were analyzed using TLSMD 75 and hydrogen atoms were included in late refinement cycles. Data collection and refinement statistics can be found in Table 4. The models and structure factors have been deposited in the PDB with accession codes 4XY8 (BRD9/7d), 4XY9 (BRD4(1)/7d), and 4XYA (BRD4(1)/11).

■ RESULTS AND DISCUSSION
In order to assess the binding of purine fragments on human BRDs, we first performed molecular docking experiments employing the previously determined crystal structure of the complex of BRD4(1) with a 5-methyltriazolopyrimidine ligand (PDB code 4MEN 30 ). To this end we investigated binding of purine fragments 1, 2a, and 2b ( Figure 1B), seeking to determine acetyllysine competitive binding modes within the BRD cavity with promising predicted binding affinities, ideally establishing favorable interactions with residues implicated in acetyllysine peptide recognition. In order to account for putative conformational changes of the receptor's binding site cavity upon ligand binding, we employed the Induced Fit docking protocol 61,62 (as implemented in the Schrodinger software package). Molecular modeling resulted in good accommodation of the investigated purine fragments within the Kac binding site of BRD4(1), mainly packing between the ZA-loop hydrophobic residues (Val87, Leu92, Leu94) and Ile146 from helix C, in a groove that is capped on one end by Tyr97 and Tyr139 and Trp81 on the other end ( Figure 1C). We observed different conformations of compound 1 within the BRD4(1) cavity, with the two chloro functions pointing to the top of the pocket ( Figure 1C) or adopting a Kac mimetic pose with one chlorine inserting deep into the pocket ( Figure  1D). Compound 2a was also found in two different states in our calculations, either orienting its primary amine function away from the conserved asparagine (Asn140, Figure 1E) or

Journal of Medicinal Chemistry
Article directly engaging this residue while orienting its 6-Br substituent toward the ZA-loop (Supporting Information Figure 1A,B). In all cases the ligand poses resulted in promising predicted binding energy values (−9.13 kcal/mol for 1, −9.95 kcal/mol for 2a, and −9.13 kcal/mol for 2b). We identified in our calculations, in the case of compound 2a, poses that exposed the halogen in position 6 so that the purine core scaffold may be further optimized based on its topology within the BRD binding site. The opposite was true in the case of the 2-Cl substituent or the methyl substituent at N9 (1 or 2b, respectively), which resulted in steric clashes and topologies that would not allow for subsequent optimizations. To better understand the binding mode of the purine scaffold to BRDs, given the multiple docking conformations observed, we decided to systematically probe the topology of these fragments employing synthetic chemistry and structure−activity relationships.
We purchased 2,6-dichloro-9H-purine 1 and 2-amino-6bromo-9H-purine 2a and synthesized compound 2b employing a TBAF-assisted N-9 methylation on the purine ring of 2a. To confirm binding of these fragments to human bromodomains, we employed a thermal shift assay (ΔT m assay) which we have previously used successfully with fragments and various human bromodomains. 23,26,28,76,77 Typically we perform the assay using 100 μM compounds in the case of fragments and very weak ligands, but in the case of the purine analogues tested we were surprised to see binding at 10 μM compound to BET BRDs and in particular to BRD4(1); however, the optimized CDK inhibitor olomoucine did not show any significant stabilization toward any proteins in the panel ( Figure 1F, Supporting Information Table S1). Encouraged by this result, we tested these compounds at the same concentration against five other diverse BRDs in an effort to cover most of the human BRD phylogenetic tree (Supporting Information Figure 1C) and found that despite their structural diversity, the BRDs of  (5), and BRD9 exhibited weak binding. We were particularly intrigued by the interaction of 2a with BRD9, since to our knowledge only a small subset of compounds has been previously shown to bind to this module. 33 To further evaluate binding of 2a onto BRD9, we performed docking experiments, using the recently crystallized apo structure of the protein (PDB code 3HME 15 ) accounting for the flexibility of key residues after ligand binding. 61,62 Similar to our BRD4(1) docking experiment, we obtained two main binding poses for this compound, with the most energetically favored one (predicted binding affinity of −9.06 kcal/mol) exhibiting an extended hydrogen bond network with the conserved Asn100 and π−π interactions with the ZA-loop Tyr57 and Tyr106 from helix C, while the 6-Br substituent was found oriented toward the external part of the binding site toward the ZA-loop Phe47, suggesting that modifications on this position for subsequent optimization would not be tolerated without affecting the ligand orientation in the acetyllysine cavity (Supporting Information Figure 1D). Conversely, we observed subtle rearrangement of the side chains of Phe44 and Tyr106 accompanied by almost a 90°rotation of the side chain of Phe47, suggesting an induced-fit binding mode. We also observed an alternative binding pose (predicted binding energy of −8.62 kcal/mol) in which the ligand maintained the hydrogen bond to Asn100, albeit from its primary amine function which inserted toward the conserved asparagine, as well as the π−π interaction with Tyr106 from helix C, while orienting the modifiable 6-Br substituent toward the top of the BRD cavity (Supporting Information Figure 1E), offering a promising vector for subsequent modifications. Using both

Journal of Medicinal Chemistry
Article possible orientations as starting points, we decided to further interrogate how the purine core scaffold binds to this BRD.
Encouraged by our initial findings, we synthesized a number of 2-amino-9H-purine analogues and tested their ability to interact with human BRDs, primarily of the BET subfamily (subfamily II), while systematically testing binding to representative BRDs from other structural subfamilies (family I, PCAF; family III, CREBBP; family IV, BRD9; family V, BAZ2B; family VIII, PB1(5); see Supporting Information Figure 1C) in order to probe structurally diverse proteins against this core purine scaffold. We utilized an aqueous-phase Suzuki−Miyaura cross-coupling reaction to synthesize 2-amino-6-aryl-9H-purine derivatives, yielding highly C-6 decorated 9Hpurines in a one-step procedure and performed a subsequent TBAF-assisted N-9 alkylation to access N-9 substituted analogues (Scheme 1 and Chart 2). We accomplished the coupling step under microwave irradiation with Pd(OAc) 2 and triphenylphosphine-3,3′,3″-trisulfonic acid trisodium salt as the catalytic system, with Cs 2 CO 3 as base, in a water−acetonitrile reaction solvent. This approach allowed synthesizing 2-amino-6-aryl-9H-purines with very short reaction times (5−15 min) at high yields and purity.
First we introduced a phenyl substituent at position 6 of the core purine scaffold (compound 3a) only to find that BRD2(1), BRD4(1), and PCAF were stabilized in thermal melt assays by this compound (Figure 2A,B). We confirmed binding to BRD4(1) by isothermal titration calorimetry and measured a dissociation constant of 11.99 μM ( Figure 2C and Table 1). The interaction of 3a with BRD4(1) was mainly driven by enthalpic contributions (ΔH = −9.45 kcal/mol), opposed by negative entropy (TΔS = −2.97 kcal/mol). We were also intrigued by the lack of affinity toward BRD9 in the ΔT m assay, since our initial fragment hit, compound 2a, had exhibited a thermal shift of 1.6°C toward that domain. We therefore tested this scaffold by isothermal titration calorimetry against BRD9 and measured a weak dissociation constant of 8.5 μM ( Figure  2D, Table 2) suggesting that our primary assay (ΔT m ) may not be very robust in the case of BRD9 when applied to weak ligands. This is important, since the first bromodomain of BRD4 has been repeatedly shown to bind to weak compounds employing this assay, 76 while it has been noted that other BRDs do not always exhibit high temperature shifts although they bind to several compounds very potently. 37 Additionally, in the case of BET proteins it has been demonstrated that thermal melt data correlate well with in vitro dissociation constants. 23 Importantly, similar to the BRD4(1)/3a interaction, the compound interacted with BRD9 mainly driven by enthalpic contributions (ΔH = −9.11 kcal/mol) opposed by negative entropy (TΔS = −2.42 kcal/mol).
In order to better explore the structure−activity relationship of 6-phenyl substituted 9H-purines, we synthesized a range of compounds carrying different patterns of functions, including para-substitutions (compounds 3b−h), meta,para substitutions (compounds 4a−e), meta,meta′ substitutions (compounds 5a,b), ortho,ortho′ substitutions (compounds 6a−c), and ortho,meta′ substitutions (compounds 7a−e) (Figure 2A and Chart 2). We tested binding of these analogues to the eight BET BRDs, as well as the five more diverse BRDs previously mentioned, employing the same thermal shift assay as above. Interestingly, compounds synthesized that carried an additional methyl modification at N9 (compounds 3c, 3e, 4e, 5b) exhibited very weak or no binding toward most BRDs while showing small thermal shits (1.0−1.3°C) for the bromodomain of CREBBP. Para substitution of the 6-phenyl-9H-purines (compounds 3b−h) resulted in lower stabilization of all BRDs; however, compound 3f showed binding toward all BRDs without any hints of selectivity toward BRD9. Meta substitutions (compounds 4a−e) were also very weak across BRDs with no affinity for BRD9, suggesting that modifications on that vector were not tolerated. Interestingly meta,meta′ substitution (compound 5a) resulted in binding to most BRDs albeit weak, with ΔT m values between 1.1 and 1.8°C. As expected, this binding event was abolished when the compound was methylated at N9 (compound 5b). Binding was not improved with ortho,ortho′ substitutions of the 6-phenyl 9Hpurine scaffold (compounds 6a−c) ( Figure 2B and Supporting Information Table S1). Since methyl substitution at N9 could not be tolerated in BRD4(1) or BRD9 binding, we concluded at this stage that the five-membered ring is probably not oriented toward the top of the BRD cavity but points toward the bottom of the acetyllysine binding cavity, as predicted in our docking model (Supporting Information Figure 1A,B), with the 6-substituted position toward the front of the pocket in order to accommodate the larger phenyl-substituted functions.
We decided to further test combinations in ortho,meta′ substituted compounds by first maintaining a methoxy functionality at the ortho position while changing the steric bulk at the meta′ position (compounds 7a−d). 2-Methoxyphenyl substitution (compound 7a) yielded thermal shifts between 1.4 and 2.5°C for BET BRDs as well as 1.5°C for the BRD of CREBBP while significantly stabilizing BRD9 compared to all previous compounds tested (2.9°C). We confirmed this binding by ITC and measured a dissociation constant of 641 nM against BRD9 ( Figure 2D, Table 2). Notably the change in affinity was accompanied by negative entropic contributions (TΔS = −4.55 kcal/mol). Intrigued by this step change in affinity, we tested halide analogues at the meta′ position while retaining the ortho-methoxy functionality (compounds 7b−d) and observed improved thermal shifts for all compounds tested against BRD9, while BET affinity seemed to be variable. Notably, we observed ΔT m values following the order H < F < Cl > Br, suggesting that steric bulk and charge at the meta′ position is important, with compound 7c showing a ΔT m of 3.8°C against BRD9. We validated binding using ITC as an orthogonal method and measured dissociation constants of 351, 297, and 397 nM against BRD9 following the same ranking as the thermal shift assay for compounds 7b, 7c, and 7d, respectively ( Figure 2D, Table 2). Importantly, we observed a gradual increase in entropic contributions with increasing bulk of the 5-halide-2-methoxyphenyl substitution, with compound 7a exhibiting the lowest entropic term (TΔS = −4.55 kcal/ mol) and compound 7d the highest (TΔS = −1.57 kcal/mol). Interestingly, BRD4(1), which exhibited ΔT m values of 1.1 and 3.2°C for compounds 7c and 7d, was found to bind weakly to these scaffolds by ITC, and the dissociation constants that we measured were 2.04 and 4.7 μM, respectively, following again the same trend seen in the thermal melt assay ( Figure 2C, Table 1). Affinity for BRD9 was lost when we synthesized compound 7e which carried a bromine function at the meta′ position and an ethoxy substituent at the ortho position, suggesting that the longer and bulkier group probably affects rotation of the 6-aryl ring with respect to the core 9H-purine-2amine fold.
In order to verify the mode of interaction of the 9H-purine core within the BRD acetyllysine binding cavity, we crystallized and determined the complexes of compound 7d with the

Journal of Medicinal Chemistry
Article bromodomain of BRD9 and the first bromodomain of BRD4. The ligand was found in both cases to occupy the acetyllysine recognition pocket ( Figure 3A and Supporting Information Figure 2A) and was clearly defined in the electron density map ( Figure 3B and Supporting Information Figure 2B). Compound 7d directly engaged, via the primary amine function as well as the nitrogen at position 3, the conserved asparagine in both structures (Asn100 in BRD9; Asn140 in BRD4(1)) and was additionally held in place via a number of hydrogen bonds to the protein backbone and the network of conserved water molecules previously described 76 ( Figure 3C and Supporting Information Figure 2C). Notably, compound 7d initiated hydrogen bonds to a water molecule located on top of the BRD cavity, which in turn linked the ligand to the top of the ZAloop, either to the carbonyl of Ile53 (in the case of BRD9) or Asn93 (in the case of BRD4 (1)). This tight network of interactions explains the potency of this compound for BRDs. The mode of interaction is similar to that observed before for triazolothienodiazepine complexes such as (+)-JQ1, with the five-membered ring of the purine system acting as the acetyllysine mimetic moiety, superimposing well with the methyltriazolo ring of (+)-JQ1 (Supporting Information Figure  2D) while the bromomethoxyphenyl substituent of 7d stacks well between the ZA-loop Leu92 and the ZA-channel's Phe81/ Pro82 of BRD4(1) (Supporting Information Figure 2E). Consistent with our induced fit computational models of binding of 2a to BRD9, superimposition of the BRD9/7d complex to the apo structure of BRD9 (PDB code 3HME) 15 revealed rotations of the side chains of Phe47 and Phe44 while the top of the ZA loop collapsed toward the ligand ( Figure  3D,E), resulting in an unprecedented induced fit pocket of BRD9. Particularly the side chain of Phe47 rotated 120°, thus blocking the ZA-channel of the protein resulting in a steric bulk around compound 7d. On the basis of our thermodynamic measurements showing high entropic contributions to binding (Table 2), we concluded that all compounds in this subseries (7a−d) should be inducing a rearrangement of the binding cavity residues of BRD9 upon binding to the acetyllysine site. Intriguingly as the substituent size increases, the affinity also increases as exemplified by the determined dissociation constants, with the chloro analogue (7c) exhibiting a dissociation constant of 297 nM against BRD9 while the bromo analogue (7d) is slightly weaker (397 nM against BRD9; 4.65 μM against BRD4(1)). However, the structural rearrangement that we observed was unique to BRD9; the structure of compound 7d in complex with the first bromodomain of BRD4 did not reveal any rearrangements of the acetyllysine binding cavity as the inhibitor packed between Trp81 and Leu92 of the ZA-loop (Supporting Information Figure 2D,E). It is tempting to speculate that the observed increase in entropic contributions within the series of compounds 7a−d measured by ITC, following the increase in the halide substituent size, is associated with the side chain rearrangement observed by docking as well as the crystal structure of compound 7d with the bromodomain of BRD9. However, entropy changes may be due to a number of factors, such as release of water molecules, therefore making further interpretation of the small data set obtained very difficult.
We next questioned whether the primary amine function at position 2 of the 9H-purine core scaffold is necessary for binding to bromodomains. First we substituted the amine with a chlorine group (compound 8a, Figure 4A) while retaining the 6-(5-bromo-2-methoxyphenyl) substitution, resulting in loss of affinity toward all BRDs in our panel ( Figure 4B). In the case of BRD9 we confirmed this observation by performing an isothermal titration calorimetry measurement which yielded a K d of 7.9 μM ( Figure 4C). As with compounds from previous series, methyl substitution at position 9 completely abolished bromodomain binding (compound 8b, Figure 4A), measured by both thermal melt ( Figure 4B) and ITC assays in the case of BRD9 ( Figure 4C); larger substituents (compound 8c) could not be tolerated suggesting that the core scaffold retained its pose within the bromodomain binding cavity. Hydroxy substitution at position 2 while retaining a 6-(5-halide-2methoxyphenyl) substituent (compounds 8d−f) had variable effects on the 9H-purine affinity toward BRDs. Moreover, fluoro (8d) and bromo (8f) substituted compounds lost affinity across the panel, while the chloro-substituted compound (8e) promiscuously bound to most bromodomains in the ΔT m assay, albeit weaker than its primary amine analogue 7c ( Figure  4A,B), thus suggesting that the interactions initiated by the hydroxyl group and the conserved asparagine (Asn100 in BRD9; Asn140 in BRD4(1)) are not favored over the primary amine.
The 9H-purine crystal structure complexes that we obtained with the bromodomains of BRD9 and BRD4(1) highlighted the failure of the ligand to insert deep inside the bromodomain cavity, thus not replacing the conserved network of water molecules. In an attempt to reach deeper within the cavity, we synthesized compounds 9a and 9b, introducing a methyl group at position 8 of the 9H-purine core. The low solubility of compound 9a did not allow for any measurements, but compound 9b exhibited very low affinity for all bromodomains in our panel with the exception of BRDT(2) (ΔT m of 2.9°C), suggesting that substitution at this position of the 9H-purine core in the absence of a 2-amine function cannot be tolerated and these compounds could not displace the conserved water molecule network deep inside the bromodomain pocket. Our synthetic efforts to insert a fluorine atom at position 8 through a C-8 electrophilic fluorination on the bis(tetrahydropyran-2yl)-protected derivative of 2a, following a reported metalation− fluorination reaction with N-fluorobenzenesulfonimide, 78 were also unsuccessful; we observed formation of the corresponding 8-phenylsulfonyl product instead of the 8-fluoro derivative, similar to reported work done by Roy and co-workers, 79 (1) showing raw injection heats for titrations of protein into compound. Data have been corrected and displayed as described in (C). All ITC titrations were carried out in 50 mM HEPES, pH 7.5 (at 25°C), 150 mM NaCl, and 15°C while stirring at 1000 rpm.
Our structural insight suggested that augmentation of the 6-(5-halide-2-methoxyphenyl) substituent would take a toll on BRD9, as it would force the structure toward an apo-like conformation, sterically pushing Phe47 toward the apoconformation of BRD9, while it should not affect binding to BET bromodomains which contained an open channel between Trp81 and Leu92 ( Figure 3E and Supporting Information Figure 2E) . To test this hypothesis, we synthesized compound 10 ( Figure 4A) and tested its ability to bind to human BRDs. Unfortunately its bright yellow color and low solubility did not allow for further validation. We therefore decided to increase the size of the substituent found in 7d by cyclizing the 2methoxyphenyl ring into a 2,3-dihydrobenzofuran-7-yl while retaining the bromo function at position 5. We tested the resulting compound 11 in the ΔT m assay against the panel of bromodomains and observed increased temperature shifts among BET BRDs (between 1.7 and 5.4°C) together with a remarkable increase in the case of BRD9 (6.5°C). To verify this observation, we performed isothermal titration calorimetry measurements and obtained a dissociation constant of 278 nM for BRD9 ( Figure 4C) while BRD4(1) binding resulted in a much weaker affinity (1.4 μM) ( Figure 4D). Notably the entropic contribution of compound 11 binding to BRD9 was comparable to compound 7d for both BRD9 and BRD4(1), and thus we speculated that a similar rearrangement should be taking place upon binding to the bromodomain of BRD9.
In order to test our hypothesis that compound 11 binding would also result in structural rearrangement of the bromodomain binding cavity of BRD9, in a similar way to compound 7d, while not affecting the cavity of BRD4(1), we attempted to determine the crystal structures of this compound in complex with these two bromodomains. Compound 11 readily crystallized with BRD4(1) and was found to occupy the acetyllysine binding cavity of the bromodomain ( Figure 5A) and was very well-defined in the density ( Figure 5B) despite its weaker binding affinity of 1.37 μM toward BRD4(1) ( Table 1). The ligand was found to directly interact with the conserved asparagine (Asn140) as well as with the conserved network of water molecules while packing between the ZA-channel tryptophan (Trp81) and the ZA-loop leucine (Leu92) ( Figure  5C). Our efforts however to obtain a BRD9/11 complex did not yield diffracting quality crystals suitable for structure determination, and as such, we employed computational methods to account for its binding to BRD9. Rigid docking into the BRD9/7d complex structure resulted in a conformation similar to that observed with compound 7d, with the ligand engaging the conserved asparagine via its primary amine function and the 6-aryl-substituted ring packing betweent the ZA-loop Ile53 and Phe44 (Supporting Information Figure 3A). We then performed induced-fit docking employing the algorithms previously described for purine fragments using the complex of BRD9/7d as starting point and obtained a pose whereby the 2-amine function inverted and inserted in the BRD pocket, without any changes in the surrounding side chains of Phe44, Phe47, Ile53, and Tyr106 (Supporting Information Figure 3B). Intrigued by this finding, we performed another induced fit docking experiment, starting with the BRD9 apo structure and allowing residues to freely move in the presence of the ligand. We observed a similar set of side chain rearrangement within the BRD9 acetyllysine cavity, including  (11), resulting in loss of signal due to separation of the BRD9-histone complex. (E) Titration of compounds 7d and 11 into HEK293 cells transfected with NanoLuc-fused full length BRD4 (UniProt code O60885) and Halo-tagged histone H3.3. Although the ligands gradually disrupt the BRD4-FL/H3.3 interaction, they fail to elicit the same effect as in the case of BRD9 in the concentration range tested. a rotation of Phe47 resulting in capping of the binding groove, accompanied by repositioning of Phe44 from helix C and Ile53 from the ZA-loop (Supporting Information Figure 3C). We therefore concluded that the 2-amine-9H-purine scaffolds that we developed can induce a closed pocket within the bromodomain of BRD9 resulting in tight binding, without at the same time exhibiting high affinity for BRD4(1) or other BET family members. Indeed, we determined the dissociation constants for binding to BET domains employing ITC and found that both compounds 7d and 11 exhibited low micromolar affinities (1.4−4.6 μM) correlating well with the higher thermal shifts observed (Supporting Information Table  S1), while retaining higher affinity against the bromodomain of BRD9 (Supporting Information Figure 4 and Tables 2 and 3).
Having obtained a small window of selectivity over BRD4, we wanted to verify that the 2-amine-9H-purine scaffolds that we developed are in fact active in a cellular environment and can perturb the interaction of BRD9 with acetylated histones. Previous studies have established that potent compounds that competitively bind to the acetyllysine binding cavity of BET proteins can displace the entire protein from chromatin in fluorescence recovery after photobleaching (FRAP) assays, 23,35,77 while larger BRD-containing proteins that contain in addition multiple reader domains are harder to fully displace; in the case of CREBBP this was solved by creating an artificial GFP-fusion construct that contained three BRD modules of CREBBP and a nuclear localization signal. 38 BRD9 is part of the large SWI/SNF complex, 49 and its bromodomain has been shown to bind to acetylated histone H3 peptides. 15 We speculated that 9H-purines should be able to competitively displace the bromodomain of BRD9 from chromatin and constructed a bioluminescence resonance energy transfer (BRET) system that combined NanoLuc luciferase fusions of the BRD9 bromodomain (Supporting Information Figure 5A) or full length BRD4 and Halo-tagged histone H3.3 as BRET pairs. This assay is an excellent tool to quantify protein−ligand interactions in a cellular system 80 and has recently been used to determine cellular IC 50 values for the inhibition of the histone− bromodomain interaction in the case of BRPF1 using a 1,3dimethylbenzimidazolone scaffold. 42 First we established by fluorescence microscopy that Halo-tagged histone H3.3 is readily incorporated into chromatin (Supporting Information Figure 5B). We then performed dose−response experiments and demonstrated the NanoLuc-BRD9 bromodomain was readily displaced from chromatin by compounds 7d and 11 with cellular IC 50 values of 3.5 ± 0.11 μM and 477 ± 194 nM, respectively ( Figure 5D), in agreement with the in vitro affinities determined by ITC (Table 2). In contrast, full-length BRD4 was not completely displaced in this assay up to concentrations of 33 μM of both compounds ( Figure 5E), suggesting that the compounds retained the in vitro selectivity toward BRD9 in this cellular system. We further examined the toxicity of compound 11 toward HEK293 cells, using cell viability in the presence of the compound in the concentration regime of our BRET experiments as a readout and did not observe any cytotoxicty (Supporting Information Figure 5C), suggesting that this compound can be used in cellular systems to target BRD9/Kac interactions without affecting BRD4/Kac interactions or causing any cytotoxic responses.

■ CONCLUSION
We have described here a structure-guided approach to identify inhibitors for diverse BRDs starting from small fragment-like 9H-purine scaffolds. Through structure−activity relationship we established compounds 7d and 11 and characterized them as nanomolar binders for the BRD of BRD9 while retaining weak in vitro activity against the first bromodomain of BRD4. These compounds are not cytotoxic at concentrations up to 33 μM in a HEK293 system and can competitively displace the BRD9 bromodomain from chromatin while failing to displace full length human BRD4 at the same concentration range in bioluminescence proximity assays, despite the fact that the in vitro affinity difference for these two proteins is not large. Importantly, compound 7d induces structural rearrangement on the acetyllysine binding cavity of BRD9 resulting in an unprecedented cavity shape which accommodates this scaffold, explaining the higher affinity toward this protein, while docking studies suggest that compound 11 elicits the same type of structural rearrangement. The 9H-purine scaffold therefore offers a simple template that can be used to generate initial tools that will no doubt prove useful in interrogating the biology of bromodomains beyond the BET subfamily, such as BRD9, which have not attracted attention until now, avoiding cross-reactivity with BET bromodomains.