Fragment-Based Discovery of BRD4-BD2 Selective Binders: Biophysical Validation and Structural Insights for Domain-Selective Inhibitor Design

Abstract

BRD4, a member of the bromodomain and extra-terminal (BET) family of epigenetic readers, plays a central role in transcriptional regulation and is implicated in a wide spectrum of diseases. Although BRD4 inhibitors targeting both bromodomains have advanced into clinical studies, their development and therapeutic use are limited by dose-limiting toxicities associated with nonselective BET inhibition. These limitations underscore the need for domain-selective targeting strategies. In this study, we performed fragment-based screening of a 200-compound diversity library against BRD4-BD2, identifying three chemically distinct hits: CCBT-T-226 (methyl 2-(2-(3,5-dimethylisoxazol-4-yl)­acetamido)-5-propylthiazole-4-carboxylate), CCBT-F-70 (2-(benzylamino)­pyrimidin-4­(3H)-one), and CCBT-F-182 (3-(8-methyl-3,8a-dihydroimidazo­[1,2-a]­pyridin-2-yl)­aniline). These fragments were validated as binders using TSA, ITC, STD-NMR, and ¹H-¹⁵N HSQC NMR. Among them, CCBT-T-226 and CCBT-F-182 exhibited measurable BD2-over-BD1 selectivity, with CCBT-F-182 showing a pronounced domain preference, and CCBT-T-226 demonstrating approximately 7-fold selectivity for BD2 over BD1. X-ray crystallography of the BRD4-BD2–CCBT-F-70 complex, together with cofolding models for CCBT-T-226 and CCBT-F-182, provides a structural framework for the rational, structure-guided development of potent and selective BRD4-BD2 inhibitors.

1. Introduction

Epigenetic regulation controls gene expression without altering the DNA sequence, playing a fundamental role in cellular identity, development, and disease pathogenesis. Among these mechanisms, histone lysine acetylation modulates chromatin structure and transcriptional activity by weakening histone-DNA interactions and promoting an open chromatin conformation that is permissive to transcription. − These modifications are dynamically written and interpreted by specialized reader proteins that recognize acetyl-lysine residues to regulate downstream transcriptional events. Bromodomains represent one of the most well-characterized families of such reader modules wherein these conserved protein domains, typically ∼110 amino acids in length, adopt a canonical four-helix bundle architecture (αZ, αA, αB, and αC helices) connected by the ZA and BC loops forming a hydrophobic binding cavity that accommodates acetylated lysine residues, via a conserved asparagine-mediated hydrogen bond essential for ligand recognition. − Among bromodomain-containing proteins, the bromodomain and extra-terminal (BET) family has emerged as a central regulator of transcriptional programs and disease pathogenesis, comprising four members: BRD2, BRD3, BRD4, and the testis-specific protein BRDTeach sharing a conserved domain architecture consisting of two tandem N-terminal bromodomains (BD1 and BD2), followed by an extra-terminal (ET) domain, and in some cases, a C-terminal domain involved in transcriptional regulation, enabling BET proteins to function as epigenetic scaffolds that coordinate gene expression through recognition of acetylated histones and transcription factors. −

BRD4 plays a pivotal role in transcriptional elongation and chromatin signaling by binding acetylated chromatin via its bromodomains and recruiting P-TEFb to promote RNA polymerase II phosphorylation, thereby regulating key oncogenic genes, including MYC, BCL2, and CDK6. − Its dysregulation has been implicated in a wide range of diseases, including hematological malignancies, solid tumors, inflammatory disorders, and cardiovascular diseases, establishing it as a compelling target. − The first demonstration of this potential came with small-molecule inhibitors such as JQ1, followed by clinical-stage pan-BET inhibitors, including I-BET762, OTX015, and CPI-0610, which competitively displace BET proteins from chromatin by targeting both the BD1 and BD2 domains across all BET family members with comparable affinity. − However, this nonselective inhibition has been associated with dose-limiting toxicities, including thrombocytopenia, gastrointestinal toxicity, and fatigue, arising from broad transcriptional disruption in normal cells, thereby driving interest in domain-selective BET inhibitors with improved therapeutic windows. −

Although BD1 and BD2 share the canonical bromodomain fold, subtle differences in amino acid composition, including the Asp144/His437 switch and Ile146/Val439 substitution (BRD4 BD1/BD2 numbering), create distinct binding pocket geometries that underlie divergent biological functions and opportunities for selective targeting. − Functionally, BD1 primarily mediates stable chromatin binding and basal transcription, whereas BD2 governs stimulus-dependent gene regulation and P-TEFb recruitment, with BD1 preferentially recognizing acetylated histone H4 and BD2 exhibiting higher affinity for acetylated histone H3 peptides. − These findings indicate that selective targeting of BD2 can preferentially disrupt disease-associated transcriptional programs while sparing essential housekeeping functions, a distinction that has driven the development of BD2-selective inhibitors such as ABBV-744, GSK046, and XY153, which exhibit improved safety profiles compared to pan-BET inhibitors. ,− Selective targeting of BRD4-BD2 represents a particularly attractive strategy in autoimmune and inflammatory diseases and may prevent radiation-induced fibrosis. ,, Nevertheless, most current BD2-selective inhibitors act as pan-BD2 agents with comparable affinity across BRD4, BRD3, BRD2, and BDRT, given the high degree of similarity among BET-BD2 domains, which makes this inherently challenging to achieve. ,

Recent advances in fragment-based drug discovery, combined with structure-guided optimization, have demonstrated the feasibility of selective BRD4-BD2 inhibitors, offering the potential to improve therapeutic index. Building on this, we implemented a fragment-based screening strategy, combined with orthogonal biophysical validation and high-resolution structural characterization, to identify chemical scaffolds that preferentially bind BRD4-BD2 over BRD4-BD1. This integrated approach enabled efficient exploration of diverse chemical space within a rational, structure-guided framework, leading to the discovery of distinct fragments that serve as early chemical tools for structure-guided optimization of BRD4-BD2 inhibitors or degraders with enhanced specificity and reduced off-target effects.

2. Materials and Methods

2.1. Protein Expression and Purification

The BRD4 bromodomains BD1 and BD2 were cloned into the pET28a expression vector containing an N-terminal 6×His tag followed by a TEV protease cleavage site for recombinant expression in E. coli. Constructs encoding BRD4-BD1 and BRD4-BD2 were transformed into BL21 (DE3) E. coli cells and grown in LB medium at 37 °C to an OD₆₀₀ of 0.8. Protein expression was induced with 0.5 mM IPTG, followed by overnight incubation at 18 °C. Cells were harvested by centrifugation (6000 rpm, 30 min, 4 °C) and resuspended in lysis buffer containing 50 mM HEPES (pH 7.5), 500 mM NaCl, 5% glycerol, and 0.5 mM TCEP, supplemented with Protease Inhibitor Cocktail III (1:200 v/v). Cells were lysed by sonication at 4 °C, and the lysate was clarified by centrifugation (16,000 rpm, 45 min, 4 °C). The clarified supernatant was incubated with Ni-NTA resin for 1 h at 4 °C, and the bound protein was eluted using an imidazole gradient ranging from 20 mM to 500 mM. Protein-containing fractions were pooled and subjected to TEV protease cleavage overnight at 4 °C to remove the 6×His tag, and purity was assessed by SDS–PAGE. The cleaved protein was concentrated and further purified by size-exclusion chromatography using a HiLoad 16/60 Superdex 75 column (GE Healthcare, Cytiva). The protein was eluted in 25 mM HEPES (pH 7.5), 150 mM NaCl, 5% glycerol, and 0.5 mM TCEP, and purified fractions were stored at −80 °C.

2.2. Thermal Shift Assay (TSA)

TSA for BRD4-BD2 was performed using a QuantStudio 6 Flex Real-Time PCR System equipped with a 96-well heat block (Applied Biosystems). For primary screening, 15 μM BRD4-BD2 protein and 500 μM fragments were added to a final reaction volume of 25 μL in 25 mM HEPES buffer (pH 7.5), with DMSO maintained at 1.5% across all samples. SYPRO Orange dye (Thermo Fisher Scientific; Cat. No. S6650; 5000× stock) was added at a final 10× concentration (1:500 dilution). Control reactions contained 15 μM BRD4-BD2 and 1.5% DMSO without fragments. Reaction mixtures were prepared in MicroAmp Fast Optical 96-well plates (Applied Biosystems), incubated on ice for 15 min, and sealed with optical adhesive film. Thermal unfolding profiles were obtained by ramping the temperature from 25 to 95 °C at 1 °C min^–1 with continuous fluorescence monitoring. Raw fluorescence data were exported to Protein Thermal Shift software (Thermo Fisher Scientific) for curve fitting and Tm determination using a Boltzmann sigmoid function. ΔTm values were calculated relative to the averaged control Tm values. The initial screen of 200 fragments was performed in singlets. Hits were subsequently retested in triplicate against BRD4-BD2 to confirm activity and evaluated against BRD4-BD1 under identical conditions to assess selectivity.

2.3. Saturation Transfer Difference NMR (STD-NMR)

STD-NMR spectra were recorded on a Bruker Avance-III HD 600 MHz spectrometer at 298 K. Bruker TopSpin software was used to acquire and process the NMR data. A solution (500 μL) containing ligand (400 μM) and protein (10 μM) was prepared in D₂O with 1% D₆-DMSO, 10 mM phosphate buffer (pH 7.4), and 50 mM NaCl. For STD NMR measurements, the standard Bruker pulse sequence stddiffesgp.3 was used. The on-resonance frequency was set to 0 ppm and the off-resonance frequency to −40 ppm. The number of scans was typically 64. Both the saturation pulse and the relaxation delay were set to 5 s, and the power of the shaped pulse for saturation was set to 40 dB.

2.4. Isothermal Titration Calorimetry (ITC)

The thermodynamic parameters governing the interaction between BRD4-BD1 and BRD4-BD2 with CCBT-T-226 were determined using ITC on a MicroCal PEAQ-ITC instrument (Malvern Panalytical). The same experimental conditions were used for both proteins. All samples were prepared in identical buffer conditions (25 mM HEPES, 150 mM NaCl; pH 7.5, 0.1% DMSO) and degassed prior to use. The sample cell was loaded with 30 μM of CCBT-T-226, while the syringe contained 250 μM BRD4-BD2 protein. Titrations were performed at 25 °C using 13 injections: an initial 0.4 μL injection, followed by 12 injections of 3 μL each, with a 150 s interval between injections. The reference power was set to 10 μcal/s, and the stirring speed was maintained at 750 rpm. Control experiments were performed by titrating protein into the buffer in the cell, and the resulting heats of dilution were subtracted from the raw data prior to analysis. Data were processed and analyzed using MicroCal PEAQ-ITC Analysis Software (Malvern Panalytical), and the binding isotherm was fitted using a one-set-of-sites model to extract thermodynamic parameters.

2.5. Nuclear Magnetic Resonance Spectroscopy

For ¹H–¹⁵N HSQC NMR analysis, ¹⁵N-labeled proteins were expressed in E. coli BL21 (DE3) cells grown in M9 minimal medium supplemented with 0.5 g/L ¹⁵NH₄Cl. All ¹⁵N-labeled constructs were purified using the same protocol as described for the unlabeled proteins, with the final NMR buffer consisting of 10 mM Tris (pH 7.2), 100 mM NaCl, and 1 mM DTT. All NMR experiments were performed on a 600 MHz Bruker Avance III HD spectrometer equipped with a cryoprobe. Data were processed using Bruker TopSpin software and analyzed with NMRFAM-SPARKY. Typical ¹H–¹⁵N TROSY-HSQC spectra were acquired at 298 K with 2048 complex points in the direct dimension and 128 increments in the indirect dimension, using 64 scans per increment and a recycle delay of 1.0 s. The spectral widths were 9615 Hz for ¹H and 2128 Hz for ¹⁵N. Backbone amide resonance assignments were transferred from previously reported data for BRD4-BD2 (BMRB ID: 50146) and BRD4-BD1 (BMRB ID: 50145). Titration experiments were performed using 50 μM protein in the same buffer supplemented with 0.5% DMSO, with fragments added to a final concentration of 200 μM.

2.6. Crystallization, X-ray Data Collection, and Structure Determination

The BD2 domain of BRD4 was crystallized in the apo form using previously reported conditions. Initial trials used 0.1 M MIB buffer (pH 7.0) and 30% PEG 1000 via sitting-drop vapor diffusion at 4 °C. Conditions were optimized by varying PEG concentration and buffer pH using both sitting- and hanging-drop methods at 4 °C, with drops prepared by mixing equal volumes (2 μL) of protein (10 mg mL^–1) and reservoir solutions. Diffraction-quality crystals were typically obtained within 6–8 days at 25–30% PEG 2K and 0.1 M MIB buffer (pH 7.0). For structural characterization of fragment hits, two approaches were employed: soaking and cocrystallization. For soaking, hit fragments were prepared as 20 mM stock solutions in crystallization buffer and incubated with apo BRD4-BD2 crystals for 4 h at room temperature, with periodic monitoring for crystal integrity. The structure of the BRD4-BD2 complex with CCBT-F-70 was determined by soaking apo crystals. Crystals were harvested, cryoprotected, and flash-frozen in liquid nitrogen prior to data collection. Diffraction data were collected to 2.4 Å at 100 K (λ = 1.540562 Å, Rigaku MicroMax-007 integrated with a HyPix 6000 detector, processed using CrysAlisPro and scaled using AIMLESS within the CCP4 suite. , The structure was solved by molecular replacement in Phaser-MR using BRD4-BD2 (PDB ID: 5UEX, chain A) as the search model, followed by iterative model building in Coot and refinement in REFMAC5 and Phenix. Atomic coordinates and structure factors have been deposited in the RCSB PDB (accession code: 26FF)

Co-crystallization trials for CCBT-T-226 and CCBT-F-182 were conducted by screening protein–fragment mixtures against Hampton and Jena Bioscience crystallization kits using the sitting-drop method at 16 °C. Diffraction-quality crystals of CCBT-T-226 obtained following further optimization by the hanging-drop method were subsequently determined to be in the apo form.

2.7. Computational Prediction of protein–ligand Binding Modes

Protein–ligand complex structures for CCBT-T-226 and CCBT-F-182 in complex with BRD4-BD2 were predicted using Boltz-2 cofolding via the Neurosnap cloud platform (Neurosnap Inc.; https://neurosnap.ai). Boltz-2 is an open-source biomolecular foundation model (MIT/Recursion) that extends the diffusion-based AlphaFold 3 architecture with a binding affinity prediction module trained on ∼3 million protein–ligand measurements from ChEMBL, BindingDB, and PDBbind, enabling FEP-comparable accuracy at over 1,000-fold greater speed. Protein sequences and ligand SMILES were provided as YAML inputs, and multiple independent models were generated per complex; only predictions with aggregated confidence scores exceeding 0.95 were retained.

2.8. Molecular Dynamics (MD) Simulations

MD simulations were performed using the Desmond Molecular Dynamics System (Desmond v6.5; D. E. Shaw Research & Schrödinger Release 2021–1). , The initial coordinates of the protein–ligand complex structure generated via Boltz-2 cofolding, were prepared with the Protein Preparation Wizard to optimize hydrogen-bonding networks and assign protonation states at pH 7.0. The system was explicitly solvated in an orthorhombic box with the TIP3P water model, ensuring a minimum 20 Å buffer. Neutralization was achieved with 0.15 M NaCl, with a 10 Å exclusion zone applied around the ligand. System relaxation and equilibration followed Desmond’s default multistep protocol. A 100 ns production run was conducted under the NPT ensemble at 300 K and 1.01325 bar using the OPLS_2005 force field. Temperature and pressure were controlled with the Nosé–Hoover thermostat and Martyna–Tobias–Klein barostat. Trajectory frames were recorded every 100 ps, and RMSD and RMSF analyses were performed using the Desmond Simulation Interaction Diagram tool.

3. Results and Discussion

3.1. Fragment Screening Identifies Three Distinct Scaffolds Targeting BRD4-BD2

To identify fragment-sized ligands targeting the BRD4-BD2 bromodomain, a fragment-based screening campaign was conducted using thermal shift assay (TSA) as the primary biophysical filter. A library of 200 chemically diverse fragments was screened, with ligand-induced thermal stabilization (ΔTm) serving as the binding readout and a threshold of ΔTm ≥ 1 °C applied to define primary hits. Six fragments exceeded this threshold in the initial screen. Upon confirmatory retesting, three fragments, CCBT-T-226, CCBT-F-70, and CCBT-F-182, demonstrated reproducible thermal stabilization of BRD4-BD2, with ΔTm values of 2.4 °C, 1.2 °C, and 2.1 °C, respectively (Figure A), confirming consistent, ligand-induced stabilization of the target domain. Structurally, the three confirmed hits represent chemically distinct scaffolds (Figure B). CCBT-T-226 incorporates a 3,5-dimethylisoxazole moiety, − CCBT-F-70 features a dihydropyrimidinone moiety, , and CCBT-F-182 contains an imidazo­[1,2-a]­pyridine scaffold, − motifs with established precedent in reported bromodomain inhibitor chemotypes, supporting their capacity to engage the acetyl-lysine (KAc) recognition pocket.

1.

Fragment-based screening of BRD4-BD2. (A) Thermal shift assay showing binding of three fragment hits to BRD4-BD2. (B) Chemical structure of the hits identified through fragment screening. (C) Reconfirmation of the hits binding by STD-NMR. One-dimensional (1D) spectra are shown with reference spectra in blue and difference spectra in red; nonligand peaks are masked.

To orthogonally validate binding in solution and confirm direct protein–ligand interaction, selected hits were subjected to STD-NMR. All three fragmentsCCBT-F-70, CCBT-F-182, and CCBT-T-226produced well-resolved, ligand-specific signals in the STD spectra (Figure C), confirming direct, specific binding to BRD4-BD2 in solution. Taken together, convergent evidence from TSA and STD-NMR establishes CCBT-T-226, CCBT-F-70, and CCBT-F-182 as validated fragment hits against BRD4-BD2, collectively representing three structurally distinct chemical scaffolds and providing a diverse starting point for structure-guided optimization of BRD4-BD2 inhibitors.

3.2. CCBT-T-226 Engages the Canonical Acetyl-Lysine Binding Pocket of BRD4-BD2

CCBT-T-226 was prioritized for detailed biophysical characterization based on its consistent thermal stabilization of BRD4-BD2 in TSA, exhibiting a ΔTm of ∼2.4 °C at both 150 μM and 500 μM, indicative of specific, concentration-independent engagement with the target domain. Structurally, CCBT-T-226 incorporates a 3,5-dimethylisoxazole moiety, a well-established (KAc) mimetic prevalent in bromodomain inhibitor chemotypes, providing a structural rationale for its capacity to engage the canonical KAc recognition site. To obtain quantitative binding parameters, ITC was performed, yielding a dissociation constant (Kd) of 1.3 μM (Figure A). This low micromolar affinity is notable for CCBT-T-226 and is consistent with the degree of thermal stabilization observed by TSA, further reinforcing the robustness of the interaction.

2.

Binding characterization of BRD4-BD2 with CCBT-T-226. (A) ITC data for CCBT-T-226 binding to BRD4-BD2. (B) Overlay of HSQC spectra of BRD4-BD2 in the presence of DMSO (blue) and CCBT-T-226 (red), with perturbed residues highlighted by green circles. (C) Mapping of HSQC perturbed residues onto BRD4-BD2 structure (PDB: 7RUH). The perturbed residues are shown in stick representation, the ZA loop (blue, between αA and αZ) and BC loop (orange, between αB and αC) are indicated.

Due to the absence of crystal structures for compounds CCBT-T-226 and CCBT-F-182 and ¹H-¹⁵N HSQC NMR spectroscopy was employed to resolve the binding site at residue-level precision (Figure B). Titration of CCBT-T-226 into isotopically labeled BRD4-BD2 induced significant peak intensity disappearance as well as chemical shift perturbations (CSPs) that mapped exclusively to residues lining the canonical KAc binding pocket, confirming site-specific engagement (Figure B). The conserved Asn433 peak intensity has disappeared, consistent with its well-characterized role in coordinating the acetyl carbonyl of KAc-containing substrates through hydrogen bonding and supporting a KAc-mimetic binding mode for the isoxazole moiety (Figure B). Peak disappearances were observed for residues Ala373, Phe376, Lys378, Asp381, Val382, Leu385, and Gly386 within the ZA loop, and for Asn433, His437, and Glu438 within and flanking the BC loop, together with Met442 on helix αC. These residues collectively define the canonical KAc binding cavity of BRD4-BD2 (Figure C). The disappearance of WPF shelf residue Phe376 and the ZA channel gating residue Leu385 further supports ligand engagement consistent with established bromodomain inhibitors. Beyond this primary anchor point, significant CSPs were observed for Trp374 (WPF shelf tryptophan), Val380, Cys391, Asp392, Ile394, Met400, Thr402 (helix αA), Phe426, Cys429 (helix αB/BC loop boundary), and Lys445 (helix αC), indicating compound engagement of the structural elements that collectively define the compound binding pocket. Of particular significance for selectivity, Asp436 and His437 in BRD4-BD2 correspond to Gly143 and Asp144 in BRD4-BD1, a nonconservative substitution at the BC loop that constitutes a primary determinant of BD2 over BD1 selectivity. Perturbation of these residues suggests that CCBT-T-226 already exploits this divergence, establishing a structural basis for selective BD2 engagement. Collectively, the low micromolar Kd observed in ITC, thermal stabilization in TSA, and well-defined, site-specific NMR perturbations unambiguously establish that CCBT-T-226 binds specifically within the functional KAc recognition site of BRD4-BD2 with a binding mode characteristic of the BET bromodomain inhibitor class. The KAc-mimetic isoxazole scaffold, productive WPF shelf engagement, and perturbation of selectivity-relevant BC and ZA loop residues together identify CCBT-T-226 as a structurally informative and chemically tractable starting point for structure-guided optimization toward potent, selective BRD4-BD2 inhibitors.

3.3. Differential Domain Selectivity of Fragment Hits for BRD4-BD2 Over BRD4-BD1

To evaluate intraprotein domain selectivity, all three BRD4-BD2 fragment hits were assessed against BRD4-BD1 under identical TSA conditions. CCBT-T-226 induced a ΔTm of 1.4 °C for BRD4-BD1, lower than the 2.4 °C stabilization observed for BRD4-BD2, indicating a quantitative preference for the BD2 domain (Figure A). CCBT-F-70 produced comparable stabilization of both domains (ΔTm ∼ 1.3 °C for BD1 versus ∼1.2 °C for BD2), suggesting limited domain discrimination. CCBT-F-182, by contrast, stabilized BRD4-BD1 by only ∼0.4 °C, well below the 2.1 °C shift observed for BRD4-BD2, yielding the largest interdomain ΔTm of the series and indicating a pronounced BD2 selectivity among the three hits (Figure A; Table ).

3.

Specificity of CCBT-T-226 for BRD4-BD2 over BRD4-BD1. (A) Thermal shift assay comparing BRD4-BD2 and BRD4-BD1 in the presence of CCBT-T-226. (B) ITC profile of CCBT-T-226 binding to BRD4-BD1. (C) Overlay of HSQC spectra of BRD4-BD1 in the presence of DMSO (blue) and CCBT-T-226 (red). The residues that showed perturbations are highlighted with green circles, and (D) Mapping of HSQC perturbed residues onto the BRD4-BD1 structure (PDB: 8ZM8) upon addition of CCBT-T-226. The perturbed residues are shown as sticks and the ZA loop (blue, between αA and αZ) and BC loop (orange, between αB and αC) are indicated.

4.

Specificity of CCBT-F-182 for BRD4-BD2 over BRD4-BD1. (A) Thermal shift assay comparing BRD4-BD2 and BRD4-BD1 in the presence of CCBT-F-182. (B) Overlay of HSQC of BRD4-BD2 with DMSO (blue) and CCBT-F-182 (red). (C) Mapping of HSQC perturbed residues on BRD4-BD2 structure (PDB: 7RUH) upon addition of CCBT-F-182. Perturbed residues are marked in stick representation. The ZA loop (blue, between αA and αZ) and BC loop (orange, between αB and αC) are indicated. (D) Overlay of HSQC spectra of BRD4-BD1 with DMSO (blue) and CCBT-F-182 (red).

1. Biophysical Selectivity Profiling against BRD4-BD2 and BRD4-BD1 .

	ΔTm (°C)		Kd (μM)
Compound	BRD4-BD2	BRD4-BD1	BRD4-BD2	BRD4-BD1
CCBT-T-226	2.4	1.4	1.3	9
CCBT-F-70	1.2	1.3	ND	ND
CCBT-F-182	2.1	0.4	ND	ND

^aND: Not determined.

To quantify domain selectivity for the highest-affinity hit, ITC was performed for CCBT-T-226 against BRD4-BD1. This yielded a Kd of ∼9 μM (Figure B), approximately 7-fold weaker than its BRD4-BD2 affinity (Kd ∼1.3 μM; Figure A), confirming BD2 preference (Table ). To determine BD1 binding site, ¹H–¹⁵N HSQC titration of CCBT-T-226 into uniformly ¹⁵N-labeled BRD4-BD1 revealed two distinct classes of spectral responses. Residues Asp88, Ala89, Tyr97, Ile138, Asn140, and Asp144 exhibited complete attenuation of cross-peaks upon ligand addition, consistent with direct involvement in the binding interface (Figure C). These residues map to the ZA and BC loops flanking the KAc recognition site, including the conserved Asn140. In addition, residues Phe83, Val87, Leu94, Tyr98, Lys99, Ile100, Lys102, and Thr103 displayed notable chemical shift perturbations (CSPs) for residues lining the canonical KAc binding pocket (Figure D). This pattern supports direct engagement of the conserved recognition motifs. Notably, the CSP profile closely resembles that observed for BD2, indicating that the reduced affinity for BD1 arises from diminished binding potency within a conserved binding mode.

Titration of CCBT-F-182 into ¹⁵N-labeled BRD4-BD2 produced residue-specific perturbations consistent with engagement at the canonical binding site. Cross-peaks intensities corresponding to ZA loop residues Val380 and Asp389 were significantly reduced upon ligand addition. In addition, a distinct set of residuesAsp381, Val382, Leu385, Gly386, Cys391, Ile393, Thr402, Asn428, Cys429, Glu438, and Val440exhibited measurable chemical shift perturbations (Figure B). Structurally, these residues span the ZA loop (Asp381–Gly386), the αA′ region (Cys391, Ile393), the αB helix (Asn428, Cys429), and the BC loop/αC helix junction (Glu438, Val440) (Figure C). Notably, no perturbation was observed for Asn433, consistent with the absence of a KAc-mimetic group in CCBT-F-182. The coexistence of peak disappearance and progressive CSPs indicates a noncanonical KAc binding mode, suggesting the involvement of direct contacts and conformational exchange at the ZA and BC loop interfaces. Collectively, the perturbed residues delineate a contiguous surface centered on the conserved KAc recognition site (Figure C). In contrast, titration of CCBT-F-182 into BRD4-BD1 yielded markedly attenuated perturbations. The observed CSPs were weak and spatially restricted, confined to Val87 and Ala89 in the ZA loop and Ile146 and Val147 in the BC loop residues (Figure D). No peak disappearance or appreciable attenuation was detected across the titration series. This subdued response, relative to the pronounced perturbations and peak attenuation observed for BRD4-BD2, reflects weaker ligand engagement at the BD1 binding site, consistent with the thermal shift data. Collectively, these results demonstrate that both fragment hits, CCBT-T-226 and CCBT-F-182 exhibit a measurable binding preference for BRD4-BD2 over BRD4-BD1, with CCBT-F-182 displaying a pronounced selectivity.

3.4. Characterization of Fragment Binding Modes to Guide Optimization

To gain atomic-level insight into fragment binding modes, the cocrystal structure of BRD4-BD2 in complex with CCBT-F-70 (2-(benzylamino)­pyrimidin-4­(3H)-one) was determined by X-ray crystallography (Table ). Two protein–ligand complexes present in the asymmetric unit, the two BRD4-BD2 molecules superimpose with an RMSD of 0.2 Å across all Cα atoms, confirming structural equivalence. Clear electron density for the ligand is observed in both chains (Figure A). The 3,4-dihydropyrimidin-4-one core adopts an essentially identical pose in both copies, with the pyrimidinone ring oriented within the KAc binding pocket and the benzyl substituent directed outward toward the pocket entrance (Figure A).

2. Data Collection and Refinement Statistics of the BRD4-BD2: CCBT-F-70 Complex Structure .

Parameters	BRD4-BD2: CCBT-F-70
Wavelength (Å)	1.540562
Space group	P212121
Unit cell parameters
a, b, c (Å)	48.046, 63.820, 75.241
α, β, γ (deg)	90, 90, 90
Resolution (Å)	19.30–2.40 (2.49–2.40)
R merge (%)	11.0 (36.0)
Mean [⟨I⟩/σ(⟨I⟩)]	7.0 (2.2)
CC (1/2) (%)	98.6 (93.8)
Total no. of reflections	37249 (4233)
Unique reflections	9412 (962)
Completeness (%)	99.1 (99.8)
Multiplicity	4.0 (4.4)
Molecules in asymmetric unit	2
Solvent content (%)	45.50
Refinement statistics
R work (%)	21.66
R free (%)	26.13
Mean B-value (Å2)	30.75
Ramachandran favored (%)	99.07
Ramachandran allowed (%)	0.93
Ramachandran outlier (%)	0
rms deviations
Bond length (Å)	0.002
Bond angle (deg)	0.500

^aValues in parentheses correspond to the highest resolution shell.

5.

Crystal structure and predicted binding modes from cofolding. (A) Crystal structure of the BRD4-BD2:CCBT-F-70 complex. CCBT-F70 is shown in yellow (ball and stick representation), and water molecules are shown as semitransparent spheres. The interacting residues are marked in stick representation. Inset: 2Fo-Fc electron density map contoured at 2σ around CCBT-F70, (B) Co-folded structure of BRD4-BD2 in complex with CCBT-T-226. (C) Co-folded structure of BRD4-BD2 in complex with CCBT-F-182. (D) Co-folded structure of BRD4-BD1 in complex with CCBT-T-226. The ZA loop (blue, between αA and αZ) and BC loop (orange, between αB and αC) are indicated.

The pyrimidin-4-one moiety functions as a KAc mimetic, recapitulating the canonical acetyl-lysine recognition interactions by engaging the conserved Asn433 through a direct hydrogen bond, while the carbonyl oxygen forms a water-mediated hydrogen bond interaction with Tyr390. , An additional water-mediated hydrogen bond is observed between the pyrimidin-4-one nitrogen and Asn433. Asn433 also forms a water-mediated hydrogen bond with the linker nitrogen connecting the pyrimidinone and benzyl moieties, further anchoring the core within the pocket (Figure A). Val380 and Phe376 contribute van der Waals contacts to the heterocyclic core, stabilizing its position at the base of the binding site. The benzyl substituent extends toward the hydrophobic subpocket lined by Val439 and the WPF shelf residues Trp374 and Pro375, establishing favorable packing interactions that contribute additional binding affinity (Figure A). CCBT-F-70 does not make direct interactions with residues unique to BD2, which likely explains its poor selectivity between BRD4-BD2 and BD1. Together, the pyrimidinone KAc-mimetic interactions and benzyl engagement with the WPF shelf support a bipartite binding mode for CCBT-F-70 and provide a basis for further optimization.

3.5. Predicted Binding Modes of CCBT-T-226 and CCBT-F-182 by Cofolding

The binding modes of CCBT-T-226 (methyl 2-(2-(3,5-dimethylisoxazol-4-yl)­acetamido)-5-propylthiazole-4-carboxylate) and CCBT-F-182 (3-(8-methyl-3,8a-dihydroimidazo­[1,2-a]­pyridin-2-yl)­aniline) within BRD4-BD2 were predicted using cofolding, revealing markedly distinct but complementary binding orientations within the shared KAc recognition pocket.

The isoxazole core of CCBT-T-226 is oriented deep within the KAc-binding pocket, where it engages Asn433 through a direct KAc-mimetic hydrogen bond (Figure B), a interaction corroborated by¹ H-¹⁵N HSQC NMR. Phe376, Val380, and Ile387 provide supporting van der Waals contacts, while the thiazole ring packs against the WPF shelf residues Trp374 and Pro375. The propyl substituent extends into the BC loop, contacting His437, Glu438, Val439, and Met442 and the methyl ester moiety on the thiazole projects out of the KAc pocket toward the ZA loop region. The linker amide is not positioned to form a direct hydrogen bond but is at a favorable distance to form a water-mediated hydrogen bond interaction with the side chain of His437 and the backbone of Phe376 (Figure B). Co-folding with BRD4-BD1 confirmed a comparable binding mode (Figure D). The BC loop residues (Asp436-His437-Glu438-Val439) of BRD4-BD2 lies close to the thiazole and propyl moiety of CCBT-T-226, potentially enabling direct or water-mediated interactions. These residues are altered in the BC loop of BD1 (Gly144-Asp145-Asp146-Ile147), which may represent one major difference contributing to the binding selectivity between BD2 and BD1. Additionally, the ZA loop residues also show variation, with Tyr377 and Lys378 replaced by Gln41 and Gln42 in BD1. We propose that the Lys378 side chain could orient toward CCBT-T-226, potentially enabling direct hydrogen bonding and water-mediated interaction with the methyl ester group attached to the thiazole. We hypothesize that these nonconservative substitutions alter the electrostatic and steric profile of this subpocket and may be key determinants underlying the selectivity of CCBT-T-226 for BD2 over BD1, warranting further investigation. This analysis identifies the propyl and methyl ester vectors as tractable handles for further exploiting these differences to enhance BD2 selectivity. BD2 selectivity over BD1 was further supported by MD simulation. A 100 ns simulation of the cofolded structure showed a high ligand RMSD (∼8 Å) in the BRD4-BD1 complex, compared to a markedly lower RMSD (∼2 Å) in the BRD4-BD2 complex. Although both complexes showed high RMSD during the initial phase, the BD2 complex stabilized over time and maintained a consistently low RMSD throughout the simulation (Supporting Information, Figure S1). These results support preferential binding of CCBT-T-226 to BRD4-BD2, consistent with the TSA, ITC, and NMR binding data. Co-folding of CCBT-T-226 with the BD2 domains of other BET family members, such as BRD2, BRD3, and BRDT, was also modeled and predicted binding poses similar to that observed for BRD4-BD2 (Supporting Information, Figure S2).

In contrast, CCBT-F-182 occupies the hydrophobic KAc core via its benzenamine moiety, which acts as a non-KAc mimetic and does not form a hydrogen bond with Asn433, as supported by NMR interaction studies. Cys429, Phe376, and V380 create a hydrophobic environment that accommodates the benzeneamine group. The imidazo­[1,2-a]­pyridine ring is oriented outward and stabilized by His437, Val439, Trp374, and Pro375 (Figure C). This hydrophobic-driven binding mode, involving key residues from the BC and ZA loops, likely underlies the pronounced BD2-over-BD1 selectivity of this fragment. Introducing an Asn433-directed polar interaction represents a clear opportunity for affinity maturation. Together, these models, combined with the CCBT-F-70 crystal structure, provide complementary, multivector structural frameworks for the rational advancement of all three scaffolds toward potent and selective BRD4-BD2 inhibitors.

4. Conclusion

BRD4 bromodomains represent validated therapeutic targets, with increasing emphasis on achieving BD2 selectivity over BD1 to maximize therapeutic benefit while minimizing on-target toxicity. Fragment-based screening of a 200 diversity compounds against BRD4-BD2 identified three structurally distinct hits: CCBT-T-226, CCBT-F-70, and CCBT-F-182, representing 3,5-dimethylisoxazole, dihydropyrimidinone, and imidazo­[1,2-a]­pyridine scaffolds, respectively. Residue-level interactions within the BD2 and BD1 binding pockets for CCBT-T-226 and CCBT-F-182 were characterized by NMR spectroscopy, while X-ray crystallography elucidated the binding mode of CCBT-F-70. Comparative domain profiling revealed a measurable preference for BD2 binding in both CCBT-T-226 and CCBT-F-182, with CCBT-F-182 showing pronounced interdomain selectivity, as validated by thermal shift and NMR studies. CCBT-T-226 demonstrated approximately 7-fold selectivity for BD2 over BD1. The crystal structure of the BRD4-BD2–CCBT-F-70 complex revealed a bipartite binding mode comprising KAc-mimetic hydrogen bonding with Asn433 and benzyl-driven hydrophobic interactions with the WPF shelf. For CCBT-T-226, cofolding models indicate that the isoxazole moiety mediates engagement with Asn433, supplemented by ZA and BC loop contacts contributed by the thiazole ring and the propyl substituent, as supported by NMR titration studies. In contrast, CCBT-F-182, which lacks a KAc-mimetic group, adopts a hydrophobic-driven binding mode without direct engagement of Asn433, identifying this as a potential opportunity for affinity improvement. Collectively, this work identifies three distinct fragment hits with validated binding modes and measurable BD2 selectivity, providing a structural foundation for the guided development of potent and selective BRD4-BD2 inhibitors with favorable pharmacokinetic properties.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.6c06516.

• Figure S1 presents the RMSD profiles of the molecular dynamics simulations of CCBT-T-226 with BRD4-BD2 and BRD4-BD1; Figure S2 shows the cofolded structures of CCBT-T-226 in complex with BRD2-BD2, BRD3-BD2 and BRDT-BD2 (PDF)

#.

S.P.B. and Y.V.S. contributed equally to this work. S.P.B. and V.K. performed fragment screening; S.P.B. and T.K.P. determined the crystal structure; S.P.B and Y.V.S. handled protein production; Y.V.S. and S.G.B. conducted NMR and ITC experiments; A.S. and B.P. carried out X-ray data collection; B.P. provided scientific input; K.N. designed the study, supervised the work, and analyzed the NMR, ITC, and crystallographic data; K.N. wrote the manuscript with input from all authors.

The authors declare no competing financial interest.