Structure-Based Identification of Inhibitory Fragments Targeting the p300/CBP-Associated Factor Bromodomain

The P300/CBP-associated factor plays a central role in retroviral infection and cancer development, and the C-terminal bromodomain provides an opportunity for selective targeting. Here, we report several new classes of acetyl-lysine mimetic ligands ranging from mM to low micromolar affinity that were identified using fragment screening approaches. The binding modes of the most attractive fragments were determined using high resolution crystal structures providing chemical starting points and structural models for the development of potent and selective PCAF inhibitors.


■ INTRODUCTION
Bromodomains are epigenetic readers that specifically recognize acetyl-lysine (Kac) marks on proteins. Targeting this protein interaction module with small molecules has recently emerged as a potential therapeutic strategy for the treatment of several diseases including cancer and inflammation. 1,2 To date, most inhibitor development efforts have been focused on the BET family of bromodomain proteins for which several inhibitors have now entered clinical testing. Bromodomains have good predicted druggability, 3 and selective chemical tool compounds have been developed even for less attractive binding sites that possess open or highly charged acetyl-lysine binding pockets such as BAZ2 4,5 and ATAD2. 6 In addition, potent inhibitors have been developed for highly druggable bromodomains present in BRPF, 7,8 CBP, 9 and BRD9. 10 Interestingly, several recent inhibitor development projects have highlighted the success of fragment-based approaches identifying inhibitors, in particular for poorly druggable bromodomains. 5,6,8,11,12 P300/CBP-associated factor (PCAF, also known as histone acetyltransferase KAT2B) is a multidomain protein that harbors an acetyltransferase (HAT) and E3 ubiquitin ligase domains as well as a C-terminal bromodomain that may associate with the HATs P300 and CBP. 13−15 While the roles of the acetyltransferase and the E3 ubiquitin ligase activities have been shown to be required for cell proliferation and apoptosis, 14−18 little is known about the regulatory function of the PCAF bromodomain in cellular processes. Selective inhibitors, so-called chemical probes, would therefore be interesting reagents to unravel the functions of the PCAF bromodomain and to assess its therapeutic potential as a targeting site for drug development.
An important role of the PCAF acetyl-lysine recognition module has already been demonstrated for the replication of AIDS viruses. The PCAF bromodomain targets the HIV TAT protein acetylated at K50, an essential association that activates HIV-1 transcription and promotes the integrated proviral replication. 19,20 Development of PCAF bromo-domain inhibitors has therefore been proposed as a potential strategy for the treatment of AIDS, 19,20 and this strategy has been confirmed by early N1-aryl-propane-1,3-diamine-based compounds that have been shown to prevent PCAF/TAT association. 21 This compound class has recently been followed up by the development of a series of 2-(3-aminopropylamino) pyridine 1oxide derivatives that however did not lead to improvement of the micromolar potency of current PCAF bromodomain inhibitors. 22 This prompted us to identify more diverse chemical starting points using fragment screening by thermal shifts assays and NMR as well as structure-based approaches.

■ RESULTS AND DISCUSSION
To extend structural knowledge from our previous apo crystal structure, 23 we initially determined the crystal structure of the PCAF-acetyl-lysine complex to provide insights into acetyllysine recognition. The binding mode of the acetyl-lysine in PCAF strongly resembled that observed in other bromodomains with the Kac carbonyl forming two canonical hydrogen bonds to a conserved N803 and a conserved water molecule. This water is part of a conserved intricate network of structural waters located at the bottom of the pocket that bridge interaction with the ZA loop Y760 and the backbone carbonyl of E750 (Figure 1a). Despite sharing the conserved interactions, the Kac binding groove in PCAF was fenced on one side by the bulky residue Y809, resulting in a tight and restricted pocket (Figure 1b). Large aromatic residues in this position are not common for human bromodomains, present in less than a third of proteins in the family. 23 Another interesting structural feature was the extension of the binding pocket by a shelf created by the ZA loop harboring a mixture of hydrophobic and polar surface residues ( Figure  1b). The wide ZA-loop pocket poses a challenge for inhibitor development. However, recently the development of Baz2 inhibitors demonstrated that such large binding pockets can be efficiently occupied for instance by aromatic ring systems of the BAZ2-ICR probe that are aligned by intramolecular aromatic πstacking. 4 The low affinities of the previously developed compounds prompted us to search for other Kac mimetics that would be compatible with the constraint nature of the central PCAF pocket. We employed different techniques to screen two fragment libraries. In one set of experiments, we used a thermal stability shift assay to screen a small fragment set at 1 mM and identified several fragments that resulted in positive melting temperature (T m ) shifts in the range of ∼1−2°C ( Figure 2). These hits were chemically diverse and contained different potential Kac mimetic scaffolds. The highest T m shifts were recorded for 1-methylquinolin-2(1H)-one) (1, BR004), 1Hindole-5-carboxamide (2, BR005), 2-methylisoindolin-1-one (3, BR013), 4-methoxybenzo[d]isoxazol-3-amine (4, MB360), 2,3dihydrobenzo[b] [1,4]dioxine-5-carboxamide (5, MB364), and (4-(1,2,3-thiadiazol-4-yl)phenyl)methanamine (7, MB093), most of which comprised known Kac mimetic scaffolds. In parallel, we used target immobilized NMR screening (TINS) to screen a diverse collection of approximately 950 commercially available fragments. 24 We first confirmed that binding of Kac to immobilized PCAF could be observed in 2% DMSO (Supporting Information, Figure S1). Subsequently mixtures of fragments were screened with a maximum of four compounds per mix such that the DMSO concentration was never more than 2%. Consistent with earlier predictions, we observed a TINS profile indicating that PCAF is highly ligandable (Figure 2c). On the basis of the profile, 63 hits were selected, a 6.6% hit rate. In contrast to TINS, binding of Kac to immobilized PCAF in the presence of DMSO could not be detected by SPR. Consequently, only those hits that were readily soluble in an aqueous buffer were orthogonally validated by SPR. The binding of 17 hits could be fully characterized by SPR, and three were selected for structural studies based on novel features (Supporting Information, Figure S2 and Table  S1). These fragments (9, 10, and 11) also resulted in substantial T m shifts ( Figure 2).   Table S1.

Brief Article
We next attempted to verify the binding modes of the identified fragments and successfully determined the complex crystal structures for seven Kac mimetic fragments. As expected from their chemical structure, all of these fragments occupied the Kac binding site through groups that mimicked the hydrogen bond interaction of acetyl-lysine. Because the cocrystallized compounds were small, the contacts with the bromodomain were limited only to the canonical hydrogen bond with N803 and the typical water-mediated contact with Y760 ( Figure 3a). However, additional interactions were also observed for Fr11, of which the 1-ethanol decoration was oriented toward the open ZA cavity and formed both direct and water-mediated hydrogen bonds to the backbones of the ZA loop V752 and P751. Superimposition of all structures revealed that the co-crystallized fragments fit tightly into the narrow Kac pocket, and most fragments formed aromatic interactions with Y809 that lines the central acetyl-lysine binding groove of the PCAF bromomdomain (Figure 3b). No significant structural alterations were observed when comparing all complexes, suggesting that the PCAF bromodomain contains a rigid acetyllysine binding pocket. Some Kac mimetic groups of the identified fragments were not specific for PCAF and have been previously shown to bind to other bromodomains, for example, 1 also interacts with ATAD2 11 and the isoxazole 8 with BRD4 and CREBBP 25 with highly conserved binding modes.
The binding of these three compounds was initially confirmed by DSF with 12 and 13, showing a similar degree of stabilization with ΔT m shifts of ∼2.6−2.9°C at 1 mM concentration (Figure 4b). Because of signal interference, 14 was used at half the concentration of the others but nonetheless produced a similar ΔT m shift of ∼3.3°C. As expected, the increase of the ΔT m shift values of these three ligands suggested stronger binding to PCAF bromodomain. We used isothermal titration calorimetry (ITC) to determine binding affinities in solution.
In agreement with the DSF results, the measured K d of 13 was observed to be ∼21 μM, which was 3-fold higher than that of the more decorated 14 (K d of ∼7 μM) (Figure 4c). The increase in the affinity of the latter was due to significantly larger negative binding enthalpy, suggesting more favorable polar interaction of this inhibitor. However, the ligand efficiency (LE) decreased to 0.24 for 14 compared to LE of 0.34 for 13. To gain insight into the selectivity of the most potent fragment hits, we performed a temperature shift screen using a panel of 48 recombinant bromodomains (Supporting Information, Figure S3). This selectivity screen revealed strong interaction of 14 to bromodomains present in CECR2 (7.1°C ), BRD7 (6.5°C), suggesting that this fragment can also be optimized targeting other bromodomains. 12 and 13 were also screened using developed Alphascreen assays. At 25 μM, both inhibitors showed strong inhibition for BRD9 (Supporting Information, Table S2). This is not surprising because 1-

Journal of Medicinal Chemistry
Brief Article methylquinoline-2-ones have been developed into highly specific BRD9 inhibitors. 10 The crystal structures of the ligand−PCAF complex allowed us to evaluate the interactions of the introduced decorations. All compounds could be unambiguously positioned based on the well-defined electron density map ( Figure 5). As expected in complexes with 12 and 13, the 1-methylquinoline-2-one group occupied the Kac binding groove and formed similar canonical interactions as observed in the PCAF complex with 1, while the decoration of both compounds oriented toward the ZA pockets. However, no contact between the 12 acetamide and protein was observed. In contrast, the 13 N-linked sulfonamide was observed to replace a water molecule that bridged interactions between E756 and K753 in the Kac complex (Figure 1a), enabling the formation of two direct hydrogen bonds to these two residues as well as a watermediated contact to the carbonyl backbone of W746 ( Figure  5). Interestingly, a sulfonamide at this analogous position also engaged within the Kac site of BAZ2B in the GSK2801 inhibitor complex. 5 The binding mode of 14 was slightly different potentially due to its larger decorations and its different acetyl-lysine mimetic moiety ( Figure 5). The acetamide interacted with PCAF in a typical Kac mimetic manner, with the benzene core stacking against the Y809 in a similar fashion observed also for the other fragments. The tetrahydropyran expanded into the open ZA pocket, while the N-phenylacetamide extension occupied the BC pocket with the phenyl ring located above N803 and oriented perpendicularly to Y802 and P804 within a distance for π-stacking. However, no hydrogen bond contact was formed between these substitutions and PCAF. Considering its affinity, such limited interactions were rather surprising when compared to the binding mode of the weaker 13, of which the extended group engaged in more hydrogen bond interactions. We hypothesize therefore that 14 may gain its increased binding potency through a larger contact surface. Hence, an extension of the Kac-mimetic fragments resulted in improved binding affinities despite the lack of direct polar interaction with the protein. In addition, the large surrounding space adjacent to the ZA and BC loops provides few spatial constraints, enabling the binding of diverse chemical moieties, albeit with little shape complementarity. It is likely that elongated, flexible, and uncharged substituents that could span the interface as well as engage a few interactions may be beneficial for increasing ligand potency through targeting the ZA and BC pockets in the PCAF bromodomain.

Journal of Medicinal Chemistry
Brief Article

■ CONCLUSION
The fragment-based approach coupled with structural information presented here offers a platform expanding the knowledge on PCAF Kac-mimetic scaffolds and binding modes to aid development of potent and selective PCAF bromodomain inhibitors. The structural models provide valuable data for the rational design of inhibitors to the rigid and poorly enclosed PCAF Kac binding pocket. Comparison of inhibitor binding modes highlights the importance of aromatic interactions with Y809 and provides several diverse acetyl-lysine mimetic chemotypes. These chemotypes may serve as attractive starting points for inhibitor development efforts targeting this interesting bromodomain. All fragments that were cocrystallized formed an acetyl-lysine mimetic hydrogen bond to N803. Thus, for future development of these fragments into more potent hits the fragment need to be grown rather than linked in order to increase potency.

■ EXPERIMENTAL SECTION
Recombinant Protein Production. The PCAF bromodomain (aa 715−831) containing an N-terminal His 6 -tag was recombinantly expressed in Escherichia coli Rosetta. The protein was initially purified by Ni 2+ -affinity chromatography and subsequently cleaved the tag with TEV protease. The cleaved protein was further purified using size exclusion chromatography and stored in 25 mM HEPES, pH 7.5, 150 mM NaCl, and 0.5 mM TCEP.
Identified Fragments and Compounds. Fragments were purchased from chemical vendors and were at least 95% pure according vendor specifications. The purity and correct structure of each compound was validated by NMR. The purity of the synthesized compounds 12, 13, and 14 was assessed by analytical HPLC using an Agilent 1100 equipped with photodiode array detector (DAD), quaternary gradient pump, and micro plate sampler (Agilent 220). Separation was performed upon Centurysil C18-AQ + 5 μm, 50 mm × 4.6 mm (Johnson). The flow rate of the mobile phase was kept at 3.5 mL/min. Mobile phases B and C were acetonitrile with 0.35% CF 3 CO 2 H and water with 0.35% CF 3 CO 2 H, respectively. The gradient conditions were as follows: 0−0.5 min 1% B and 99% C, 3.7 min 90% B and 10% C, 5 min 99% B and 1% C. The injection volume was 10 μL. 12 and 13 were ≥90% pure by HPLC at 254 nm and ≥99% by evaporative light scattering detection (ELSD). 14 was ≥99% pure by both methods.
Thermal Stability Assays. The protein at 2 μM in 10 mM HEPES, pH 7.5 and 250 mM NaCl was mixed with the fragments/ compounds at 1 mM concentration (unless stated). The assays, data evaluation, and melting temperature (T m ) calculation were performed using a Real-Time PCR Mx3005p machine (Stratagene) with the protocols described previously. 26 Isothermal Calorimetry. All isothermal calorimetry (ITC) experiments were carried out on NanoITC (TA Instruments) at 15°C in the buffer containing 25 mM HEPES, pH 7.5, 150 mM NaCl, and 0.5 mM TCEP. Titration was performed by injecting PCAF (250 μM) into the reaction cell containing the compounds (15−20 μM). The corrected data of the integrated heat of titration were fitted to an independent single binding site model based on the manufactor protocol, from which the thermodynamics parameters (ΔH and TΔS), equilibrium association and dissociation constants (K a and K D ), and stoichiometry (n) were calculated. All experiments were repeated twice.
SPR. Recombinant PCAF was biotinylated on the N-terminal AVI tag in vivo. Purified, biotinylated PCAF was captured onto a NeutraAvidin-coated CM5 sensor chip surface in 20 mM sodium phosphate pH7.5, 150 mM NaCl, 2 mM DTT, and 0.05% P20 Tween. This buffer was subsequently used for all studies. The immobilization level was 10150 RU. All SPR experiments were performed using a Biacore T200 instrument at 20°C. Fragment hits from TINS were directly solubilized in the running buffer. Each fragment was titrated at six different concentrations for K D determination. An acetyl-Lys 16 peptide derived from H4 was used to monitor the ligand binding capacity of the immobilized PCAF bromodomain at regular intervals.
Crystallization and Structure Determination. Apo crystals of PCAF (∼14−18 mg/mL) were obtained using the vapor diffusion method at 4°C and either single-PEG-based (21−35% PEG 3350, 0.1 M Bis-Tris pH 5.5−7.0) or PEG-smear-based (21−40% mediummolecular-weight PEG smears buffered either with 0.1 M Bis-Tris pH 6.0−7.5 or 0.1 M Tris pH 7.5−8.8) 27 reservoir solutions. To produce the PCAF−ligand complex crystals, soaking was performed using the mother liquor supplemented with the fragments/compounds at ∼10− 20 mM concentration and 20% ethylene glycol. Diffraction data were collected either in-house or at the Diamond Light Source. Initial structure solution was achieved using molecular replacement with Phaser 28 from CCP4 suite 29 and the published coordinates of PCAF. 23 Manual model building alternated with refinement were performed in COOT 30 and Refmac, 31 respectively. Data collection and refinement statistics are summarized in Supporting Information, Table S3.
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.5b01719.
Crystallographic data collection and refinement statistics, compound syntheses, selectivity screening data (PDF)