Development of Cell-Active BRD4-D1 Selective Inhibitors to Decode the Role of BET Proteins in LPS-Mediated Liver Inflammation

Abstract

The endogenously expressed BET proteins (BRD2, BRD3, BRD4) are upstream clinical targets for anti-inflammatory treatments, where inhibition of the tandem bromodomains (D1 and D2) have proven efficacious in vitro and in vivo towards NF-κB-mediated inflammation. Despite their efficacy, dose-limiting toxicities associated with BET inhibition have limited clinical progression. One strategy to circumvent these dose-limiting toxicities has included domain- or protein-selective inhibition of the BET bromodomains. Based on previously reported 1,2,4-substituted imidazole scaffolds, we characterize and report on next-generation BRD4-D1 selective inhibitors, 39 and 41. Compound 39 is both highly potent and selective towards BRD4-D1 (K_i = 2.9 ±1.0 nM, >1700-fold over BRD2-D1 via fluorescence anisotropy) over other BET bromodomains in addition to being cell-active at nanomolar concentrations. We also characterized 39’s solubility and cellular activity in addition to its off-target hERG liability (a common cardiovascular risk for drug candidates). An acetylated analogue, 41, had an 80-fold reduced hERG affinity compared to previous BRD4-D1 selective compounds. In the context of liver inflammation, we screened 39 against an LPS-mediated cellular model of liver inflammation. Upon treatment with 39, pro-inflammatory chemokines CXCL1 and CCL2 transcripts were significantly downregulated compared to the control; however, BRD4-D1 selective inhibition remained insufficient to reproduce the anti-inflammatory activity of pan-BET treatment. On a mechanistic level, these data highlight that more than one bromodomain within the BET family may be contributing to CXCL1 and CCL2 expression, where multi-domain inhibition or other therapeutic modalities may be needed in these contexts to achieve sufficient anti-inflammatory effects.

Graphical Abstract

1. Introduction

The bromodomain and extra-terminal (BET) family of proteins have remained therapeutically important protein targets over the past two decades due to their implications in disease.[1,2] Currently, there are >50 ongoing or completed clinical trials using small-molecule pan-BET inhibitors as monotherapies.[3] The BET family consists of four separate proteins, BRD2, BRD3, BRD4, and testis-specific BRDT, each of which contain a set of tandem bromodomains (D1 and D2). The BET family proteins act as epigenetic regulators by binding to acetylated histones and transcription factors through their bromodomains, recruiting auxiliary transcriptional machinery, and influencing downstream gene expression.[4] The BET protein BRD4 can localize to super-enhancer regions enriched in acetylated K27 within histone H3 (H3K27Ac) and recruit the positive transcription elongation factor b (P-TEFb) through its C-terminal domain.[5] In the context of inflammatory signaling, this binding event induces a tumor necroses factor alpha (TNF-α) mediated pathway that is propagated by the expression of cytokines, such as C-X-C chemokine ligand 1 (CXCL1), C-X-C chemokine ligand 6 (CXCL6), and C-C chemokine ligand 2 (CCL2).[6] In alcoholic hepatitis, both acetylated K27 (H3K27ac) and trimethylated lysine 4 (H3K4me3) on histone H3 serve as activating marks to recruit transcriptional machinery and act as a chemokine super enhancer.[6,7] The BRD4 bromodomains also bind the acetylated RelA subunit of nuclear factors-kappa B (NF-κB), where BRD4 functions in a stabilizing and co-activating role.[8,9] This direct interaction between BRD4 and NF-κB is necessary for mediating immune cell infiltration, where this complex is responsible for the expression of damaging cytokines and chemokines in a wide variety of tissues, including regulating interleukin 6 (IL-6), CXCL1, and CCL2.[9–11] However, a unified mechanism of how BET proteins interact to influence inflammation pathways has remained elusive with regards to the role of individual domains.[12–14] In some cases, such as inflammation from titanium nanoparticle exposure, BRD2 and BRD4 were drivers of primary macrophage inflammation through siRNA knockdown experiments of individual BET proteins.[7,15] In other settings, BRD2 knockdown alone was not sufficient to significantly reduce IL-6 induced JAG1 expression, where BRD4 knockdown or pan-BET inhibition was needed to achieve efficacy.[16] BRD3 has also been implicated in inflammatory models, where BRD3 was found to operate independently from the other BET proteins to drive inflammation.[17] Alternative mechanisms suggest that BRD3 has been shown to cooperate with BRD4 and acetylated chromatin to drive proinflammatory cytokine production.[18] The heterogeneity of outcomes in inflammatory model systems highlight the need to further understand the mechanistic role of the BET bromodomain-containing proteins on a domain- and protein-specific level for each disease indication.

While genetic knockdown experiments have aided in deciphering BET protein function,[16–19] several small molecule inhibition approaches targeting BRD4 and other BET family bromodomains have provided domain-level insights to reduce the effects of BRD4-mediated inflammation. This approach includes pan-BET inhibitors, which inhibit all eight of the BET bromodomains. The pan-BET inhibitor IBET-151 was shown to be effective at reducing the activation of IFN-γ target genes in K562 cells.[12] Recent work by Xu et al. suggests that small molecule inhibitors targeting chromatin regulatory circuits (through pan-BET inhibition with (+)-JQ1) and NF-κB expression pathways have therapeutic potential, highlighting the parallel functions of NF-κB.[10] While these data show that the inhibition of the BET bromodomains is important for reducing NF-κB activation and its effects on inflammation, more selective inhibitors (e.g., pan-D1 and pan-D2) have elucidated domain-dependent mechanisms. Thus, pan-BET inhibitors blur our mechanistic view of this process and contribute their own toxicities, highlighting the importance of developing domain-selective probes.[19] Seminal work by Gilan et al. suggests that pan-D2 inhibition is well tolerated from a safety standpoint and also reduced the activation of IFN-γ in K562 cells.[12] Domain selective approaches, such as pan-D2 inhibitor ABBV-744, reduced LPS-stimulated neuroinflammation through the regulation of the STAT1/3/5 pathway.[20] However, when considering β-cell inflammation in diabetes model systems, pan-D1 inhibition was shown to be more effective than pan-D2 inhibition at reducing IL-1β activation by displacement of the BRD4/NF-κB complex.[13] Examples of domain-specific inhibitors also include those reported against either TLR3-mediated acute airway inflammation or colonic inflammation, such as ZL0454, ZL0516 and ZL0590.[21–23] Notably, a number of reported domain-selective inhibitors have demonstrated either biased selectivity (between 10–30 fold selective) or have not been reproducible in other assay formats, limiting and clouding the amount of mechanistic insight available.[24–26]. Further, work in both LPS-stimulated and acetaminophen-induced liver inflammation also suggests that pan-D1 inhibition is necessary to reduce damaging downstream cytokines, whereas use of iBET-BD2 was less efficacious in these inflammation models.[27] Similarly to the pan-D1-biased approaches, BRD4-D1 selective inhibitors were also effective against these inflammation models (1 and 2, Figure 1A); however, BRD4-D1 selective ligands were limited by their cellular target engagement, thus clouding the effectiveness of BRD4-D1 inhibitors in this context.

Figure 1. Previous Developments Towards BRD4-D1 Selective Inhibitors.

(A) Structures of previous compounds reported for inhibiting BET-bromodomains with differential selectivity profiles. (B) Structure of previously reported BRD4-D1 selective inhibitor, 30, with interactions with BRD4-D1 highlighted. IC₅₀ reported by fluorescence anisotropy (FA).[25] (C) Compound 30 co-crystalized with BRD4-D1 to highlight key interactions of structured-water displacement (red), phenyl ether (magenta), and the solvent exposed region (blue). Water-mediated hydrogen bond with D144 shown with dashed yellow lines. PDB: 7R9C.

When approaching BET-inhibition from a therapeutic perspective, pan-BET inhibitors such as ABBV-075 have shown promise in clinical trials due to their high efficacy in anti-inflammatory models despite the on-target selectivity associated with inhibiting multiple BET bromodomains (Figure 1A). The two most common toxicities include thrombocytopenia and gastrointestinal adverse effects both in clinical studies[28,29] and in pre-clinical assays for thrombocytopenia., e.g. a colony forming megakaryocyte assay (CFU-Mk).[14] To address these toxicities associated with pan-BET inhibition, more selective inhibitors that are pan-D1 or pan-D2 selective have been developed in efforts to reduce BET-associated toxicities. One example includes clinical candidate ABBV-744, a pan-D2 inhibitor, which is 18-fold less toxic compared to ABBV-075 in the CFU-Mk assay (Figure 1A).[14] In the same CFU-Mk assay, 2 showed a 50-fold improvement compared to ABBV-075, highlighting the potential therapeutic benefit of single-domain inhibition within the BET family. Compound 2 exhibited anti-inflammatory activity in an in vivo acetaminophen-induced liver inflammation model but failed to phenocopy pan-BET effects in LPS-stimulated in vivo models. The partial activity can be somewhat attributed to the poor cellular availability of 2, which did not engage cellular BRD4 until single digit micromolar concentrations. This elevated concentration may mask the BRD4-D1 selective profile of 2. Importantly, previously reported BRD4-D1 selective compounds 1 and 2 were also shown to be orally bioavailable.[27] Given the lack of BRD4-D1 selective inhibitors, there is a current need for cell-available inhibitors to determine the exact mechanistic effects of BRD4-D1 selective inhibition in different disease models of inflammation.

Upon further investigation into 1 and 2, we discovered that both of these imidazole-based molecules inhibit the hERG potassium ion channel with <1 μM potencies (hERG IC₅₀ = 0.97 and <0.10 μM, respectively) (Figure 1A). hERG liabilities are one of the most frequent adverse effects leading to the failure of drug candidates, where inhibition of the hERG ion channel is detrimental to clinical development.[30,31] This has been previously noted with other BET inhibitors, as well, where previous efforts have focused on removing the hERG activity of ABBV-074, resulting in a pan-D2 inhibitor that had an affinity of >30 μM.[32,33] For BRD4-D1 selective 1 and 2, there is a need to improve upon the trisubstituted-imidazole scaffold in two important aspects: (i) to eliminate the significant hERG activity that poses a cardiovascular liability, and (ii) to improve cellular efficacy and permeability in an inflammation model to further understand the mechanistic role of BRD4 in inflammation. To address hERG-related safety concerns and to aid development of cell-active inhibitors, we herein report the discovery and characterization of 39 and 41, a potent set of BRD4-D1 selective inhibitors (K_i = 2.4 ±1.0 nM and 23 ±3 nM, respectively). Both 39 and 41 improve activity against BRD4-D1 in a cellular context compared to 1 and 2. Co-crystal structures also reveal that subtle interactions in water-mediated hydrogen bonding networks with BRD4-D1 may be responsible for improved affinity of 39. Further structure-activity relationships led to 41 with improved physicochemical properties, an 8-fold reduction of hERG affinity compared to 1, and an >80-fold reduction of hERG affinity compared to 2. Our results also demonstrate that these probes effectively engage BRD4 in cells through two orthogonal assays. Using these improved BRD4-selective inhibitors, we investigated the mechanism of BRD4-D1 inhibition in LPS-stimulated inflammation in cell models. In these inflammatory models, we assess the impact of these inhibitors on the expression of downstream inflammatory cytokines, CXCL1 and CCL2. While CXCL1 was responsive to BRD4-D1 selective inhibition with our next generation inhibitors, CCL2 had similar responses to 1, suggesting that inflammatory response may be differentially regulated through the BET bromodomains. Future work seeks to understand the role of BRD4 inhibition more completely, and the impact of additional BET proteins on governing this inflammation. Importantly, the partial efficacy we observe with BRD4-D1 selective inhibition may suggest that there are compensatory mechanisms at play, and that multiple bromodomains should be inhibited in the context of inflammation, whether through co-dosing strategies (e.g. with pan-D2 inhibitors) or through less-selective small molecules (e.g. BRD4 D1 + D2 selective inhibitors).

2. Results

2.1. Structure-Activity Relationships Towards BRD4-D1 Selective Inhibitors

The high degree of lipophilicity and poor physiochemical properties derived from the aromatic core of the BRD4-D1 selective scaffold was initially hypothesized to hinder cellular inhibition of BRD4-D1, where the high lipophilicity could result in their sequestration into lysosomes or limited cellular uptake.[26,27] To address these and additional hERG activity concerns, we synthesized an initial set of compounds using the previously described synthesis.[34,35] This initial set of ligands focused on three regions within the binding pocket: (i) the structured-water displacement group, (ii) the solvent-exposed region, and (iii) the phenolic ether, which makes important hydrophobic contacts between the WPF shelf and the BC-loop of the N-terminal bromodomain of BRD4 (Figure 1B and 1C). Previous iterations of inhibitors revealed that bromo- or iodo-substituents served as potent motifs to displaced structured-waters, as did trifluoromethyl substitutions.[34,36] The solvent-exposed region of the trisubstituted imidazole scaffold has been less well-explored, focusing primarily on alkylated piperidine or (S)-pyrrolidine groups which maintain a protonatable nitrogen. The solvent exposed region is also important due to its ability to impart selectivity as demonstrated by iBET-BD1,[12] GSK789,[37] or Olinone[24], as D144 within this region is non-conserved between the D1 and D2 bromodomains. As for the phenyl ether region of earlier analogues, the pyrimidine and phenyl ether fall within a conserved groove created between the WPF shelf, ZA loop, and the αC loop of BRD4-D1. While typically more sterically hindered, phenyl ethers led to a gain in potency for our BRD4-D1 selective scaffold,[35,36] this region has been well-explored by other compounds for its hydrophobic interactions. Molecules that have capitalized on this site include pan-BET inhibitors like (+)-JQ1[1] and early D1-selective inhibitors, such as MS042[38] and compound 3A.[39]

This initial set of analogues was screened through competitive FA assays, using IBET-151 as a positive control against BRD4-D1 (Figure 2). To first address the group that displaces structured water in the binding site reported previously,[34,36] the p-trifluoromethyl group was replaced with a less hydrophobic bromine. This substitution on related analogues led to potent and selective compounds through a halogen-bonding interaction with M105.[36] To aid in strengthening this interaction, we also synthesized two electron-deficient analogues in compounds 4 and 5. However, these analogues significantly reduced affinity towards BRD4-D1. The 4-bromo- group (3, IC₅₀ = 1290 ±290 nM) was the most potent of the series, whereas the electron-deficient analogues eroded affinity (4 and 5, IC₅₀ = 3900 ±1800 nM and 3700 ±1600 nM, respectively). We observed a marked decrease in affinity compared to previously described analogues in Cui et al., who noted 4-bromo and 4-iodo substituents enhanced binding via a halogen bond using compounds with a 3,5-dimethyl phenolic ether.[34,36] The observed decrease in affinity suggests an altered binding mode resulting from changing the phenolic ether region of the inhibitor.

Figure 2. Structure-Activity Relationships of Tri-Substituted Imidazole Scaffold for BRD4-D1.

All IC₅₀ values were collected in experimental triplicates as technical replicates in a competitive FA assay for BRD4-D1, reported as mean ± s.d. IBET-151 data was collected as 12 experimental replicates of technical triplicates. Compounds in red represent IC₅₀ values more potent than 1. (A) Substitution in the water-displacement region (red) of 1. (B) Substitutions to the solvent-exposed region (blue) of 1. (C) Modifications to the piperidine-motif (green) of 1 to access acidic dyad. (D) Modifications to (S)-pyrrolidine (purple) to access interactions with BRD4-D1. (E) Screen of phenolic ethers (pink) in 1. (F) Combinative structure activity relationships and analogues (black).

We next modified the solvent-exposed region of our scaffold, a piperidine motif in 1 and 2. We hypothesized that this region would be amenable to substitutions to improve physiochemical properties while maintaining potency, as this region has been used as an exit vector for PROTAC development.[35] Likewise, replacing the piperidine with an (S)-pyrrolidine led to a micromolar affinity (6, IC₅₀ = 1400 ±600 nM), which was improved through the addition of piperidyl-based ethylamines in previous reports by engagement of an acidic dyad (D144/D145).[36] To engage these residues, we synthesized several cis-cycloakyl amines in compounds 7, 9, and 11. The corresponding cis-stereoisomers were chosen and synthesized to orient the amine to engage D144/D145. Indeed, in minimized structures, these compounds are predicted to place the primary amine between the pyrrolidine nitrogen and the N,N-dimethylethylamine shown in a previously reported BRD4-D1 selective inhibitor (Figure 1B), which showed a distance of 3.1 Å and 5.4 Å from the N1 nitrogen to the pyrrolidine nitrogen and to the N,N-dimethylethyl amine, respectively (Supplemental Figure S2). Compounds 7, 9, and 11 showed a predicted distance of 4.8 Å, 5.0 Å, and 4.8 Å, respectively, placing the amine between the pyrrolidine in previous 1,2,4-imidazole-containing compounds and the water-engaging N,N-dimethylethylamine. Consistent with these predictions, analogues 7 and 9 improved potency ~2-fold compared to 1 (IC₅₀ of 64 ±26 nM and 67 ±18 nM, respectively), whereas 11 displayed comparable binding affinity. Given our previous work on (S)-pyrrolidine motifs, we also synthesized two compounds that aimed to modulate hERG potency by incorporating electron-withdrawing β-hydroxy and β-hydroxymethyl moieties, 8 and 10. Remarkably, 8 was characterized to have a greater than 5-fold increase in BRD4-D1 potency compared to the parent 6. To further probe the solvent-exposed region, we functionalized 1 through the addition of alkylation, acylation, or sulfonylation of the amine (Figure 2). Several of these analogues contained protonatable heterocycles or primary amines to engage the aspartate dyad, such as compounds 21–23. While amines 22 and 23, showed improvements in affinity (IC₅₀ of 70 ±30 nM and 98 ±21 nM, respectively), the significant increase in hydrophobicity and the inclusion of tertiary amines left these compounds undesirable due to concerns of toxicity and hERG liabilities.[30]

To address the hERG activity of 1 and 2, we predicted that electron-withdrawing substituents or acyl-motifs would reduce hERG activity through pKa modulation of the protonatable amine.[32] Despite compound 2’s higher activity towards hERG, compound 2 has a slightly lower calculated pKa of 9.0 due to inductive effects (Supplemental Table S4). Thus, the observed increase in hERG activity is likely due to the change of a secondary amine in 1 to a tertiary amine, resulting in compound 2 behaving as a Pearlstein pharmacophore.[40,41] This preliminary data suggests that analogues that modulate the protonatable amines must significantly lower the pKa and avoid tertiary amines. These factors led to the synthesis of compounds with electron-withdrawing amide substituents (13, 15, and 24). Compound 13 showed minimal inhibitory activity towards BRD4-D1, whereas 15 showed an IC₅₀ of 290 ±180 nM. Notably, the acetylated analogue (16, IC₅₀ = 260 ±50 nM) showed comparable affinity comparable to 2 (Figure 2). This result inspired us to synthesize additional compounds with similar motifs, where the urea-based analogue (17) had less affinity with an IC₅₀ of 680 ±70 nM. Further, the sulfonyl analogues of these compounds (18 and 19, respectively) showed IC₅₀ values of 1670±280 nM and 310 ±14 nM. Incorporation of sulfamide-motifs onto the 7 to add hydrogen-bond donors resulted in 12 (IC₅₀ = 580 ±60 nM) but did not result in additional potency compared to that of 7. The presence of hydrogen-bond donors on 17 and 19 had an inverted relationship on activity compared to 18 and 16, respectively. The addition of sulfamide substituents also improved affinity in the pyrrolidine series (26, IC₅₀ = 1030±160 nM) compared to 6, albeit still yielding micromolar activity. The difference in BRD4-D1 affinity between compounds 12, 19, and 26 suggests that the positioning of the sulfamide plays a key factor in inhibitory activity. This observation was supported by compounds 25 and 27, each containing an amide substituent, having activities of 310±190 nM and 320±150 nM, respectively. Given that the IC₅₀ of compounds 25 and 27 are comparable to the affinity of compound 15, in addition to the trends for the sulfamide series, these data suggest that the piperidine and (S)-pyrrolidine motifs have unique exit vectors that are overcome by flexible alkylation substituents to engage in additional interactions with BRD4 .[34,36,42]

The importance of a hydrophobic phenolic ether has been previously investigated with this tri-substituted imidazole series.[42] While we recognized that the carvacrol-based phenolic ether shown in 1 and 2 was responsible for generating highly potent compounds,[35,36] its high hydrophobicity and potential metabolic liability represented potential challenges for cellular uptake. To this end, we synthesized a series of substituted phenolic ether analogues (compounds 30, 31, and 32) to maintain hydrophobic interactions near the WPF shelf while reducing overall hydrophobicity (Figure 1C). These compounds series resulted in little change in activity, with 30 and 31 having similar activity to 1, whereas 32 had an IC₅₀ of 270 ±40 nM (Figure 2E). However, incorporating a tert-butyl group with 33 resulted in a two- and three-fold loss of activity compared to 1 and 31, respectively (Figure 2F). Additional modification through the addition of an acetamide resulted in an approximately two-fold loss in activity, similar to the relationship seen between 1 and 15. However, mesylating the tert-butyl analogue (analogue 34) resulted in a near complete loss of activity. These data demonstrate a non-additive structure-activity relationships in these compounds, which is dependent on the solvent-exposed region and the phenyl ethers in concert with one another.

2.2. Co-Crystallization of BRD4-D1 with compounds 8 and 9

To examine how changes in the solvent-exposed region further perturb the binding angle and pose of our inhibitors, protein co-crystal structures were obtained with 8 and 9 and BRD4-D1. These crystal structures revealed that, the core binding pose remained intact with the trifluoromethyl motif for displacing structured waters and the direct hydrogen bond between N140 and the imidazole (Figure 3A). The structures of 8 and 9 overlayed with a previously reported BRD4-D1 selective inhibitor showed no significant difference in the core binding pose.[36] This suggests that the most flexible regions of the molecule are likely the solvent-exposed and the phenolic ether regions. Indeed, when comparing the pose of the phenolic ether in 9 and compounds reported in Cui et al.,[36] the carvacrol-based phenolic ether is rotated 180 degrees, whereas the water-mediated hydrogen bonding network is conserved between both crystal structures. Further comparison of 9 to previous BRD4-D1 selective compounds containing a 3,5-dimethyl phenolic ether[36] reveals that the phenolic ether of 9 is offset by 33 degrees, while the solvent-exposed region conserved the hydrogen-bonding network with D144 (Supplemental Figure S3). However, compounds containing an iodo- group as a water-displacement motif deeper within the binding pocket, which then leads to a 0.9 Å shift previous inhibitors and 9.[34,36] The collected crystal structures for this set of compounds suggest that changes in the solvent-exposed region or the phenolic ether leads to subtle changes in binding angles due to functionality of the CF₃ water displacement motif.

Figure 3. Co-Crystal Structure of 9 with BRD4-D1.

(A) Structured-water displacement region and key acetyl-lysine mimic when bound to 9 (PDB: 9P34). (B) Solvent-exposed region of 9 with a two-water bridge to D144, mimicking the interactions seen by previously reported BRD4-D1 selective inhibitors. Hydrogen bond shown by yellow dashed lines, distance measurement shown in Å.

In the co-crystal structure of 8, the pyrrolidine ring is unexpectedly directed towards the L92 backbone and away from D144 (Supplemental Figure S3). This novel interaction in the β-hydroxy pyrrolidine series could explain the increased affinity compared to 6, despite the lack of N,N-dimethyl substituents that contribute to affinity in previous reports.[36] To further probe the distance dependence of the interaction with D144, we examined the crystal structure of compound 30 with BRD4-D1. The pyrrolidine nitrogen distance closely aligned with structure-minimized models, with a predicted distance of 3.7 Å, whereas the N,N-dimethylethylamine showed a higher deviation (predicted distance of 6.4 Å), suggesting that these pendant amines remain flexible compared to the pyrrolidine-based core. Together, these structures reveal that the solvent exposed region is highly amenable to changes that maintain the water-mediated hydrogen bonding network to D144, whereas the phenolic ether mobility may be largely dependent on hydrophobic interactions of the isopropyl- and isopropyl-like functionalities.

2.3. Expanding Structure-Activity Relationships for Cycloalkyl Analogues

To further evaluate the relationship between the solvent exposed region and the flexible phenolic ether region, we repeated a similar phenolic ether screen using 3,5-biscyclopropyl, 3-cyclopropyl, and 3,5-dimethyl phenyl ethers to yield analogues 38, 39, and 40 for the cyclopentyl series and analogues 42, 43, and 44 for the cyclobutyl series. The cyclohexyl analogues were excluded due to poor physiochemical properties. To better predict cellular availability and to better reflect the drug-like properties of our analogues, we also calculated the lipophilic efficiency (LipE) of our compounds. The phenolic ethers were seemingly interchangeable in previous piperidine-based series, as with 1, The seemingly interchangeable phenols in the previous piperidine-based series, shown in 1. However, this trend does not translate to the cyclopentyl and cyclobutyl series as the addition of cyclopropyl-containing phenolic ethers in the cyclobutyl series (42 IC₅₀ = 290 ±70 nM and 43, IC₅₀ = 570 ±140 nM) did not yield greater potency (Table 1). Compound 44, however, displayed the highest LipE of the series despite weaker affinity for BRD4-D1 compared to the parent compound 11 (LipE of 2.28 versus 1.55, respectively).

Table 1. Combinative Structure-Activity Relationships Towards BRD4-D1 between solvent-exposed motifs (R₁) and phenolic ether region (R₂).

Compound	R1	R2	IC50 (nM)	Ki (nM)	cLogP	LipE
1			140 ±50	12 ±5	5.95	0.90
30			114 ±12	7.8 ±1.5	5.90	1.04
31			110 ±30	6.3 ±1.9	4.96	2.00
32			270 ±40	18.4±2 7	5.01	1.56
Compound	R1	R2	IC50 (nM)	Ki (nM)	cLogP	LipE
9			67 ±18	4.1 ±1.1	5.81	1.36
38			105 ±11	6.5 ±0.7	5.77	1.21
39			39 ±17	2.4 ±1.0	4.83	2.58
40			170 ±50	10.3 ±2.9	4.88	1.89
41			320 ±120	23.5 ±2.9	4.42	2.08
Compound	R1	R2	IC50 (nM)	Ki (nM)	cLogP	LipE
11			160 ±40	7 ±5	5.25	1.55
42			290 ±70	18 ±4	5.20	1.34
43			570 ±140	35 ±4	4.27	1.97
44			250 ±40	15.4 ±2.5	4.32	2.28

IC₅₀ and K_i values determined for BRD4-D1 in a competitive FA assay. IC₅₀ values were collected as experimental triplicates of technical triplicates, reported as mean ± s.d. K_i values reported as mean ± s.d. cLogP for each compound was calculated by ChemDraw 23.1.1. LipE values were calculated as LipE = pIC₅₀ – cLogP. Blue values represent compounds with LipE > 2.0.

When examining the relationship between the phenolic ether substituents and the cyclopentyl-based series, the structure-activity relationships diverged compared to that of the cyclobutyl-based or piperidine-based series. In this case, both the 3,5-cyclopropyl and 3,5-dimethyl substituents (compounds 38 and 40, respectively) showed slightly weaker binding than the parent compound, with IC₅₀ values of 105 ±11 nM and 170 ±50 nM, respectively (Table 1). However, the 3-substittued monocyclopropyl analogue, 39, showed an improvement in affinity to 1 with an IC₅₀ of 39 ±17 nM (K_i = 2.4 ±1.0 nM) while having the highest LipE of the molecules screened (LipE = 2.58). To investigate this trend further and to probe the effects of stereochemistry on our cyclopentyl scaffold, we docked all four stereoisomers of 9 against BRD4-D1 using the Schrodinger Maestro program using the co-crystal structure of 9 bound to BRD4-D1 as a reference. These studies revealed that the (S,R)-configuration (present in compounds 9 and 39) had the best predicted binding affinity out of the four isomers, with a docking score of –12.39 kcal/mol, whereas the other stereoisomers ranged from −11.0 to −11.8 kcal/mol (Supplemental Figure S4). The lower-energy binding poses predicted from this stereoisomer justified use of the cis-isomer, where the (S,R)-isomer best complemented the surface of the BRD4-D1 protein upon docking. When looking at predicted physiochemical properties of the cyclopentyl series, the (S,R)-configuration was also predicted to be the most soluble (QPLogS = –6.95), the most membrane permeable (QPPCaco = 3.74 ×10⁻⁶ cm s⁻¹), and was predicted to be the least potent for hERG (QPLogHERG = –6.2) out of all four of the stereoisomers. To address potential hERG concerns with 39, the acetylated analogue was also synthesized (compound 41, Table 1). Through FA, 41 had a 10-fold loss in activity towards BRD4-D1; despite this, the lower hydrophobicity of this scaffold afforded a LipE of 2.08, comparable to 44. Indeed, these three improved analogues (39, 41, 44) were the only compounds observed to have LipE ≥2.0 (Table 1, Supplemental Figure S5). Recent work by Hirst et al.[43] has demonstrated pan-BET inhibitors with much higher lipophilic efficiencies for clinical candidates (LipE >4.5). Despite the limitations of the hydrophobic 1,2,4-imidazole core, 39 shows significant improvements in both physiochemical properties and BRD4-D1 activity compared to 1.

2.4. Compounds 39 and 41 are BRD4-D1 Selective Inhibitors over Other BET Bromodomains.

Whilst the modifications made in both the solvent-exposed and the phenolic ether regions of compounds 39 and 41 compared to 1 should not impact selectivity, we next tested if these changes within the trisubstituted-imidazole scaffold could impact the selectivity of these probes towards the family of BET bromodomains. In our previous work, we established that the addition of an ethanolic-moiety to the solvent-exposed region did not significantly alter the selectivity of compound 2.[27] To address this concern for 39 and 41, we turned to differential scanning fluorimetry (DSF) as an initial platform to assess selectivity across the BET bromodomains. Here, we selected the pan-BET inhibitor, IBET-151, which showed >6 °C stabilization in melting temperature for the majority of BET bromodomains, with a 3.96 ±0.09 °C and a 3.70 ±0.12 °C stabilization for BRD3-D2 and BRD2-D2, respectively. The observed stabilization of the BET bromodomains is consistent with a nanomolar inhibitor for the BET family, as well as with the well-characterized pan-BET selectivity profile for IBET-151. Both of our previously reported compounds, 1 and 2, showed thermal shifts consistent with reported AlphaScreen data. For example, compound 1 showed the largest stabilization was observed for BRD4-D1 (ΔT_m = 4.9±0.3 °C), with more moderate stabilization towards BRD3-D1 (ΔT_m = 2.9±0.4 °C) and BRD2-D1 (ΔT_m = 1.3±0.2 °C), and negligible stabilization towards the D2 domains (Figure 4). Compound 2 had a similar stabilization profile, with changes in melting temperature for BRD4-D1, BRD3-D1, and BRD2-D1 of 7.1±0.3 °C, 3.0±0.3 °C, and 1.9 ±0.2 °C, respectively. Given the one-to-one stoichiometry present in our assay conditions, these data are also consistent with previous reports of 1, which showed BRD2-D1 and BRD3-D1 as the closest off-targets, with a 23-fold and an 83-fold selectivity window.[35]

Figure 4. Selectivity Analysis of 39 and 41 Towards BET Bromodomains.

(A) Differential Scanning Fluorimetry of BRD4 (black), BRD3 (red), and BRD2 (green) bromodomains with BRD4-D1 selective inhibitors. (B) AlphaScreen selectivity of BRD4 (black), BRD3 (red), and BRD2 (blue) bromodomains of 39 from ReactionBiology. Refer to Supplemental Tables S6 and S7 for additional information.

As we observed similar selectivity profiles for our compounds through DSF compared to previous reports,[27,35,42] we also investigated the selectivity of our two lead analogues: 39 and 41. Compound 39 showed a significant stabilization for BRD4-D1 (ΔT_m = 5.6±0.3 °C) with reduced stabilization for BRD2-D1 (ΔT_m = 1.3±0.2 °C) and BRD3-D1 (ΔT_m = 1.5±0.4 °C) compared to that of 1, indicating improved selectivity. Compound 41 had a similar selectivity profile, albeit with a lower stabilization of BRD4-D1 (ΔT_m = 4.5±0.8 °C). These differences highlight the semi-quantitative nature of DSF rather than the absolute affinity of 41.[44] Moreover, the improved apparent selectivity shown by 39, and to some extent 41, suggests that either the 3-cyclopropyl phenyl ether, the cis-cyclopentyl incorporations, or both, into the scaffold aid in the selectivity towards BRD4-D1. Finally, the selectivity observed by DSF was also confirmed through AlphaScreen for 39, which was analyzed against the six ubiquitously expressed BET bromodomains (Figure 4B). Here, the potency of 39 was observed to be higher than what we observed in our fluorescence anisotropy assay (IC₅₀ = 78 nM), though the overall selectivity for BRD4-D1 was conserved, with the closest off-target being BRD2-D2 (IC₅₀ = 1700 nM). Compound 41 was also assayed against BRD3-D1 and BRD4-D2, which gave IC₅₀ values of 6700 and 4000 nM, respectively (Supplemental Table S5). Compound 39 was shown to have a higher selectivity for BRD4-D1 than that reported in AlphaScreen when screened against BRD2-D1 and BRD2-D2. Under these conditions, 39 was more selective for BRD4-D1 than of 1, where it was 1700-fold selective over BRD2-D1 and over 4100-fold selective over BRD2-D2. Compound 41 displayed a similar level of BRD4-D1 selectivity in both assay formats (Supplemental Table S7). This discrepancy between the two biophysical assays is likely due to the lower affinity of 39 observed in the AlphaScreen format (IC₅₀ = 78 nM versus 2.4 nM for AlphaScreen and fluorescence anisotropy, respectively) but nevertheless supports that these molecules are both adequately potent and selective to serve as BRD4-D1 selective probes for investigating mechanisms of bromodomain inhibition.

2.5. Cellular Activity of BRD4-D1 Selective Inhibitors

Compounds 1 and 2 were previously reported as potent and selective against BRD4-D1 in vitro, though both of these compounds were limited as chemical probes due to their poor cellular availability until micromolar concentrations.[27] This resulted in higher concentrations needed to engage cellular BRD4-D1 and potentially leading to a pan-D1 inhibition profile for these probes through eroding the selectivity profile. To address this shortcoming of these compounds, we tested the cellular activity of a subset of compounds against BRD4-D1 using a live-cell nano-luciferase bioluminescence resonance energy transfer (NanoBRET) as a screening tool. Here, we screened an active set of compounds based on LipE and potency at 5, 1.25, and 0.3 μM to replicate previous cellular assays.[27] The NanoBRET competition-based assay uses a fluorescent tracer in HEK293T cells transfected with a nanoluciferase-tagged BRD4-D1 fusion construct to measure competitive binding to BRD4-D1. Pan-BET inhibitor IBET-151 served as a positive control. As demonstrated in our previous report, compounds 1 and 2 bind to BRD4-D1 with nanomolar potency in biophysical settings, but still fail to engage BRD4 in a cellular setting until micromolar concentrations.[27,45] Here, 1 (2 ±5% inhibition at 1.25 μM) and 2 (45 ±4% inhibition at 1.25 μM) retain low cellular activity against BRD4-D1, though 2 was observed to be slightly more active in our three-point assay screen than previously reported. IBET-151 was sufficiently cell active with 85 ±6% inhibition of BRD4-D1 at the same concentration range (Figure 5A). Although the 3-cyclopropyl phenyl ether had a reduced hydrophobicity, incorporation of the moiety did not significantly improve cellular availability in this context (compound 31). As expected, the incorporation of hydrogen-bond donors such as 8 also showed minimal cellular activity compared to 2. Of the acyl-substituted piperidine motifs (16, 17, and 19), the urea moiety was observed to have the highest cellular availability (17, 35 ±12% inhibition at 1.25 μM) despite the addition of two hydrogen-bond donors. While these acyl-substituted compounds only marginally improve cellular inhibition of BRD4-D1, changing the solvent-exposed region with the cyclopentyl- series was observed to have significantly improved activity. The carvacrol-based phenolic ether (9, 24 ±3% at 1.25 μM) showed mild improvement over 1; however, incorporation of less-hydrophobic ethers showed improved activity with 39 reaching significant activity at micromolar concentrations (68% inhibition at 5 μM, 37% at 1.25 μM). Indeed, 27 and 39 showed the greatest activity in this assay at nanomolar concentrations (39, 25 ±6% at 0.31 μM). Both of the synthesized cyclobutyl analogues, despite the promising LipE (11 LipE = 1.55, 44 LipE = 2.28), showed little change in activity at lower concentrations compared to 1. As for the amide-containing compounds (15, 25, 27), the morpholinamide-containing group (which removes all hydrogen-bond donors) showed the highest activity, with inhibition of 45 ±3% at 1.25 μM.

Figure 5. Cellular Target Engagement of BRD4-D1 Selective Inhibitors.

(A) 3-point scan of BRD4-D1 selective inhibitors against NLuc-BRD4-D1 in HEK293T. Compounds were incubated for 2 h at 37 °C. Error bars represent s.d. as biological triplicates. Blue: Acyl-motif containing analogues White: cyclopentyl analogues Orange: cyclobutyl analogues Green: Amide-containing analogues (B) Binding isotherms for IBET-151 (black), 39 (red), and 41 (blue). Error bars represent s.d. as biological triplicate of technical triplicates for 3-fold dilutions. (C) Table of EC₅₀ and K_i,app values for competition experiment shown in (B). EC₅₀ and K_i,app are shown as mean ± s.e.m. from biological triplicates from technical replicates. ^aValues for 1 and 2 are from previous reports.[21]

We then screened a subset of these compounds in an orthogonal cellular thermal shift assay in order to cross-validate the findings of our NanoBRET assay (Supplemental Figure S6).[27] Our previous work has demonstrated the reliability of the CETSA assay in two distinct cell lines to monitor cellular BRD4 engagement.[27,35] In this context, both 9 and 19 stabilized BRD4 at 1.25 μM to a level comparable to IBET-151, demonstrating the cellular efficacy of these compounds. Notably, stabilization to this extent was not observed with 1 or 2 until 5 μM concentrations.[27] Moreover, 39 stabilized BRD4 to a level similar to the 25 °C control, where it was the only compound observed to do so. In conjunction with the NanoBRET competition experiments, these data demonstrate that 39 is cell active while highlighting discrepancies in the magnitude of activity observed in the CETSA and NanoBRET assays. Both assays show the same general trends in cellular activity, where the quantitative difference may be due to the competitive nature of NanoBRET compared to the native binding observed through CETSA. To further quantify the activity observed in NanoBRET, we fully characterized both 39 and 41 in an 11-point titration with our NanoBRET assay, which showed EC₅₀ values of 1.90 ±0.08 μM and 1.64 ±0.51 μM, respectively (Figure 5B-C). While these are comparable to 1 and 2, (EC₅₀ = 1.31±0.23 μM and 2.45±0.24 μM respectively), the total inhibition at the 10 μM reveals that 41 is approximately two-fold more active in this assay compared to 1 (87.5±0.1% versus 39±18%). Further, at this concentration, 41 is more active than 39 despite the 10-fold weaker binding to BRD4. These data are consistent with an increased cell permeability profile for 39 and 41. Given the substantial increase in activity of the cyclopentyl- series in both cell-based assays, we moved forward with 39 and 41 for further characterization.

2.6. Physiochemical Properties of Next Generation BRD4-D1 Selective Inhibitors.

As for the potential of 39 and 41 as therapeutic candidates, we prioritized balancing cellular efficacy, selectivity, solubility, and hydrophobicity. To address concerns of solubility with previous hydrophobic scaffolds, we implemented an ¹⁹F NMR-based assay to measure the solubility of these compounds. Both compounds 1 and 39 were exceedingly soluble in this assay, where both compounds maintained a linear range up to 1000 μM in D₂O due to their synthesis as trifluoroacetate salts (Supplemental Figure S7). When we moved to performing this experiment in a buffered system (PBS, pH 7.4), there was a drastic decrease in the solubility for both compounds, with solubilities of 262 ±25 μM and 50 ±35 μM for compound 1 and 39, respectively. The observed decrease in solubility between the D2O and PBS systems indicates that the solubility of our scaffold relies heavily on protonation state of the imidazole core. However, this suggests that for our scaffold, analogues containing fewer basic amines (such as those present in 39) should be formulated in acidic environments to enhance solubility.

In addition to solubility, we also investigated the lipophilicity of each of our lead compounds through the determination of LogD_7.4 and their property forecast index (PFI). Both LogD_7.4 and PFI have previously served as useful metrics for describing the lipophilicity of BET inhibitors for orally-available inhibitors.[43] We proceeded to establish the LogD_7.4 for 1, 2, 39, and 41 using a previously established HPLC-based protocol.[46] Here, both 1 and 2 had measured LogD_7.4 of 4.33 ±0.04 and 4.71 ±0.09, respectively, representative of their highly hydrophobic composition (Supplemental Table S8). Compound 39 was expected and observed to be much less hydrophobic than the two previous scaffolds, having a LogD_7.4 of 2.70 ±0.10. This could contribute to the compound’s higher efficacy in our cell-based BRD4-engagement assays. Compound 41 had a LogD_7.4 much higher than both compounds of 5.26 ±0.06 owing to its lack of a protonatable amine. Despite this, compound 41 (cLogP = 4.41) remains much more polar compared to 1 or 2, contributing to its improved cellular activity through NanoBRET. Given the similarity of these four scaffolds in their aromaticity, it is therefore unsurprising that the trends between the PFIs are the same, with 39 having the lowest PFI (PFI = 6.7). Though previous work has defined orally bioavailable compounds with having a PFI < 6, other work suggests that compounds with PFI >7 are more likely to suffer from challenges involving solubility, promiscuity, and overall development.[43,47] In our case, both 39 and 41 (PFI = 6.7 and 9.2, respectively) were well above the threshold for oral bioavailability, despite previous generations of compounds showing adequate oral bioavailability in mice with substantial metabolic stability (shown with compound 1).[27] As both 39 and 41 show improved cellular engagement to BRD4-D1, we posit that these compounds serve as useful chemical probes and potential starting points for PROTACs for both cellular and in vivo work to investigate underlying mechanisms within the BET family.

2.7. Efficacy of BRD4-D1 Selective Inhibition in Reducing CXCL1 and CCL2 in LPS-Induced TSEC Models.

To determine the anti-inflammatory activity of our compounds, we chose to investigate our compounds in transformed liver sinusoidal endothelial cells (TSECs), a liver endothelial immortalized cell line. Liver sinusoidal endothelial cells an important source of chemokines and cytokines in vivo during liver injury and disease.[48–51] We have also shown that this model system is epigenetically controlled through the inhibition of BRD4 and other BET bromodomains.[27] As such, treatment of TSECs with pan-BET inhibitor IBET-151 showed significant efficacy in reducing the downstream expression of CXCL1 and CCL2 as biomarkers for LPS-induced inflammation. The pan-D2 inhibitor, iBET-BD2, importantly does not significantly reduce the expression of CXCL1 in prior work.[27] Compounds 1 and 2 also led to a reduction in CXCL1 levels, though not as significant as IBET-151 at 1.25 micromolar concentrations. When looking at the expression of CCL2 mRNA, however, IBET-151 led to a decrease in CCL2 mRNA levels comparable to control levels. As for BRD4-D1 selective compounds 1 and 2, these compounds significantly reduced CCL2 expression, but were not sufficient to reduce expression to comparable levels as IBET-151.[27] Encouragingly, changes in the scaffold of 1 and 2 to next-generation compound 39 and 41 did not impact the toxicity of our compounds towards TSECs through XTT assays until 200 μM (Supplemental Figure S8). Both 39 and 41 were not observed to impact TSEC viability at single-digit micromolar concentrations, similar to the concentrations we observe for optimal cellular engagement of BRD4.

While previous BRD4-D1 selective inhibitors 1 and 2 showed anti-inflammatory activity at micromolar concentrations, we next probed whether the previously reported anti-inflammatory response with BRD4-selective inhibition was due to limited cellular availability or due to the more selective mode of inhibition. In doing so, we paneled a series of improved inhibitors against the reported LPS-stimulated TSEC model.[27] These compounds included cyclopentyl-containing compounds 9, 39, and 40, as well as the more hydrophobic cyclohexyl- analogue 7. Together, these four compounds represent a range of potencies and physiochemical properties to determine differential cell-active effects on LPS-stimulated inflammation (LipE from 0.83 – 2.58, IC₅₀ values from 39 – 170 nM). Specifically, 40 was chosen as a representative compound with higher potency and lower hydrophobicity compared to 41. Compounds containing a carvacrol-based phenolic ether (7 and 9) were active at 1.25 μM despite their high hydrophobicity, where 9 significantly reduced the downstream expression of both CXCL1 and CCL2 (p = 0.026 and p = 0.003, respectively, Figure 6). Compound 40 also reduced the expression of CXCL1 and CCL2 compared to the LPS-treated control but was unable to achieve statistical significance for CCL2. This reduced effect is likely due to the decreased potency towards BRD4-D1 compared to the other analogues screened. The more potent and cell-active compound 39 reduced CXCL1 expression to a similar level to compounds 7 or 9 at 1.25 μM, where a similar trend towards CCL2 was observed. Importantly, despite achieving a significant reduction in cytokine expression, 7, 9, and 39 did not reduce CXCL1 nor CCL2 expression to similar levels as IBET-151. Rather, these results may suggest that BRD4-D1 inhibition alone can significantly reduce the downstream expression of these inflammatory cytokines in models of liver inflammation but are insufficient to drive a similar anti-inflammatory response as pan-BET inhibition. Given the enhanced selectivity and the improved ability of 39 to inhibit cellular BRD4-D1, these data suggest alternate mechanisms between the BET bromodomain-containing proteins in controlling the downstream expression of inflammatory cytokines, where multi-domain inhibition may be necessary to achieve therapeutic effects.

Figure 6. CXCL1 and CCL2 Expression Levels in LPS-Stimulated TSECs.

qPCR expression CXCL1 (left) and CCL2 (right) in LPS-stimulated TSEC cells dosed with 1.2 μM of inhibitor. Data shown as mean ±s.d. of experimental duplicates of technical triplicates. * = p<0.05 ** = p<0.01 *** = p<0.0001 n.s. = non-significant.

2.8. Structure-Activity Relationships for Modulating hERG Activity

While compounds 39 and 41 have yet to reach the potency required to achieve clinical efficacy in the demonstrated LPS-mediated inflammation model, the structure-activity relationships discussed provide important structural insights into mitigating the hERG liability of the trisubstituted-imidazole scaffolds. Here, we took a systematic approach to assess which parts of the trisubstituted imidazole scaffold contributed to the undesired hERG activity through a selected screen of 10 compounds. These compounds were then divided into four categories: molecules that aimed to reduce hydrophobicity, molecules that tuned pKa through inductive effects, molecules with acyl-like motifs, and combinative approaches (such as analogue 41). Generally, compounds with highly hydrophobic properties tend to be hERG inhibitors.[30] Compound 1 and compound 2 had hERG activities of 0.97 and <0.10 μM, respectively. As such, we note that compound 2’s structure was consistent with Pearlstein’s pharmacophore, as well as containing a tertiary alkyl amine, leading to its increased inhibitory activity despite the additional polarity added from the ethanolic group.[40,41] Compound 20 was synthesized as a variant of 2 through a secondary alcohol at the cost of being a tertiary amine. As expected, the additional hydrophobicity and increased pKa of this compound compared to 2 did not alleviate hERG concerns, showing similar activity as 2 (IC₅₀ = <0.10 μM, Figure 6). Rather, incorporation β-hydroxyl groups, such as those found in 8, have been shown to decrease hERG activity.[30] This trend was also observed here, as 8 had an IC₅₀ = 5.9 μM towards the hERG ion channel, an approximately six-fold improvement on 1. The reduced activity is likely aided by the incorporation of hydrogen bond donors, as there was not a significant alteration of the pKa of the protonatable amine. Compound 25 also incorporated hydrogen bond donors through addition of a primary amide, which both reduced cLogP (cLogP of 5.23 for 25, cLogP of 5.68 for 2) and reduced the pKa of the protonatable nitrogen (pKa of 7.6 for 25, pKa of 9.0 for 2). This combination of factors decreased hERG activity to 2.82 μM. Importantly, we observed hydrophobicity alone did not have a significant impact on hERG activity (Figure 7) as 31 and 39 (cLogP of 4.95 and 4.83, respectively) have similar hERG activity to 1 despite a significant reduction in hydrophobicity. It is also possible that the positioning of the protonatable nitrogen has a minor role in hERG activity, as both 8 and 25 contain (S)-pyrrolidine motifs, compared to piperidine-based moieties in structures 1 and 2. However, in our series, scaffolds containing additional hydrogen-bond donors compared to 1 and 2 displayed less hERG inhibition. From these compounds, it was clear to us that inductive effects and the incorporation of hydrogen-bond donors did reduce activity of our scaffold towards the hERG ion channel, but not to a sufficient point for to eliminate therapeutic concern. Rather, the largest decrease in off-target hERG activity came from the incorporation of acyl and sulfamide motifs, shown by molecules 16 and 19, respectively. These compounds had hERG IC₅₀ values of 5.68 and 8.87 μM, respectively. These observed decreases in activity are likely due to the modification of the solvent-exposed amine, which is no longer protonatable in these scaffolds to serve as a hERG pharmacophore. Further, 19 which has two additional hydrogen-bond donors compared to 16, also showed lower activity while having little change on hydrophobicity (cLogP = 5.96). Finally, 41 with an acylated amine had a similar activity compared to 19 (IC₅₀ = 8.43 μM) with while having significantly reduced hydrophobicity and retained hydrogen-bond donors compared to other motifs. While it is clear from our data that the incorporation of hydrogen bond donors and acyl-motifs aided in reducing off-target hERG activity, we chose to evaluate these effects with in silico models to guide future scaffold designs. The ten compounds screened for hERG activity were also evaluated using Schrodinger’s Maestro suite and with ChemAxon hERG predictors. Both of these software packages correctly predicted general hERG trends in our model series, ChemAxon had the best fit to absolute inhibitory activity (Supplemental Table S4). From here, we observe no clear trend between hydrophobicity, pKa through inductive effects, or other properties of our compound due to the lack of matched pairs in our analysis. However, we noticed a distinct limit in our imidazole-based scaffold, where molecules with a total polar surface area (TPSA) >65 Ų were more likely to exhibit favorable hERG properties (Supplemental Figure S9). This metric aids in the development of further selective inhibitors in our work with the trisubstituted imidazole scaffold as a limit for potential hERG risk. This is a limited set of data, however, where more compounds should be screened for further therapeutic development. Given the favorable hERG profile (eight-fold improved over 1) and cell-permeable scaffold of 41, this compound can thus serve an adequate probe for investigating BRD4 biology through the selective inhibition of BRD4-D1.

Figure 7. Structure-Activity Relationships to hERG Inhibitory Activity for BRD4-D1 Selective Inhibitors.

Structures of compounds screened against hERG activity in patch-clamp assay by Pharmaron. Analogues that probed hydrophobicity (orange) and inductive and acylation effects (blue) were screened, along with combinatory scaffolds (i.e., 41).

3. Discussion

In this work, we report the synthesis and characterization of two next-generation BRD4-D1 selective inhibitors, compounds 39 and 41, based on a previously-reported 1,2,4-trisubstituted imidazole scaffold.[27,35,42] Compound 39 improved upon the potency of this scaffold compared to previous inhibitor 1 (K_i = 2.4 ±1.0 and 12 ±5 nM, respectively). These next-generation inhibitors also showed a greater selectivity for BRD4-D1 over 1’s closest off target, BRD2-D1, through an FA assay (1700-fold selective for 39, 320-fold selective for 1). The selectivity of 39 is reduced in the AlphaScreen assay compared to FA, likely due to the decreased affinity of our compounds toward BRD4-D1 resulting from solubility challenges. Importantly, these differences in affinity were also observed with 1.[27] Despite these assay differences in absolute selectivity, we posit that 39 serves as a useful and selective inhibitor for BRD4-D1 in biophysical and cellular assays. Both 39 and 41 address the shortcomings of 1, including the latter’s undesirable physiochemical properties and inability to inhibit cellular BRD4. For cellular engagement of BRD4, we note that 39 and 41 inhibit BRD4-D1 in a live-cell assay 1.7- and 2.3-fold more than 1 at 10 μM concentrations (Figure 5C). The observed improvement in cellular efficacy was then confirmed through CETSA, which is both orthogonal to NanoBRET and captures more native protein conditions. Here, 39 had the greatest stabilization of BRD4 at 0.3 μM concentrations of the inhibitors screened, including IBET-151. As 39 and 41 have a ten-fold difference in potency (41 K_i = 24 ±3 nM), our combined data suggest that the cellular efficacy of these inhibitors is linked to not only the potency of our lead compounds, but also the improved drug-like properties of these compounds. In this context, we quantified the hydrophobic nature of these inhibitors through lipophilic efficiency (LipE) as well as LogD_7.4. Of our lead compounds, 39 had the lowest LogD_7.4 (2.70 ±0.10) and the highest LipE (2.58), consistent with its improved cellular efficacy over 1. When we quantified the solubility, however, 39 showed roughly a four-fold reduction in solubility compared to 1 in PBS-buffered systems. While the reason behind the reduced solubility in unclear, this difference may be due to the hydrophobic motifs that are required to maintain BRD4-D1 potency and intra-BET selectivity within the scaffold. Under more acidic conditions, 39 becomes increasingly soluble, suggesting that solubility issues could be circumvented through optimal formulation.

We next investigated the activity of our BRD4-D1 selective compounds against the hERG potassium ion channel. As we further develop BRD4-D1 selective probes as potential therapeutics to treat liver inflammation,[27] the off-target inhibition of the hERG channel is a determining factor and a common adverse event while developing small-molecule therapies.[30] Here, we paneled ten analogues for their inhibitory activity. This preliminary screen revealed that two preliminary factors that governed hERG activity within this series include the presence of protonatable amines (such as those within 1 and 39) in addition to the total polar surface area of these compounds. In our case, the combination of increasing the total polar surface area >65 Ų along with the acylation of 39 led to the most reduction in off-target hERG activity (41, IC₅₀ = 8.43 μM). While there are still improvements to be made on this front, 41 represents an >80-fold improvement compared to previously explored analogue 2. Further, we primarily focused on 39 and 41 as leads for their anti-inflammatory response through our previously established LPS-stimulated TSEC model system.[27] In this inflammatory system, 39 was less efficacious compared to pan-BET inhibition with IBET-151 in reducing CXCL1 and CCL2 mRNA expression, despite improvements of cell permeability and physiochemical properties in the former. When compared to previous-generation inhibitor, 2, our compounds were moderately more efficacious in reducing the expression of CXCL1 and CCL2 mRNA. Taken together, these data suggest a compensatory mechanism within the BET family to govern LPS-mediated inflammation, where BRD4-D1 inhibition shows partial anti-inflammatory activity compared to pan-BET approaches. Pan-D1 inhibitors, IBET-151, and other less-selective inhibitors of the BET bromodomains[27,42] were significantly more efficacious in reducing inflammatory response. These data suggest that either BRD2 and BRD3 contribute to the inflammatory response observed (and thus the inhibition of the broader BET family is needed), or that both bromodomains of BRD4 should be targeted (i.e., through PROTAC-based approaches).[35,52] BRD2 was recently implicated in NF-κB-mediated alcoholic liver injury, supporting co-regulatory mechanisms between the BET family.[53] Due to these limitations, future studies will focus on the effects of co-dosing pan-D2 inhibitors in combination with 39 along with the selective degradation of BRD4 through PROTACs. Indeed, these experiments will determine whether BRD4-D2 contributes significantly to LPS-mediated inflammatory models, where BRD4-selective degradation[35] may offer an alternative approach for developing therapeutic modalities.

4. Conclusions

In this work, we report the development of BRD4-D1 selective inhibitors, 39 and 41, which improve upon cellular BRD4 engagement, BET bromodomain selectivity, and non-specific toxicity of previously reported BRD4-D1 selective inhibitor, 1 (Table 2). These improved scaffolds shed light onto the design principles of developing BRD4-D1 selective inhibitors, where the resulting inhibitors had significantly improved lipophilic efficiencies and reduced hydrophobicity. These design factors allowed for the cellular engagement of BRD4-D1 with 39 and the preliminary determination of BET mechanisms with liver-inflammation models, where previous reports have proposed that pan-D1[13] inhibition or BRD4/BRD2 combinative approaches[7,15,28] as suitable anti-inflammatory strategies. We demonstrate that the selective inhibition of BRD4-D1 in LPS-mediated inflammation models show partial efficacy in reducing downstream pro-inflammatory cytokines. However, BRD4-D1 selective inhibition remains less efficacious compared to pan-BET inhibition. Through the analysis of BRD4-D1 selective inhibition in LPS-mediated inflammation, we show that BRD4-D1 inhibition only contributes to a partial reduction in inflammatory response (primarily through CXCL1). These data support a compensatory mechanism between the BET proteins contributes to inflammation response through select cytokines, where other less-selective inhibitors targeting the BET family were more efficacious.[27] While future work seeks to address these models and the role of BRD4-D2 compared to D1-targeting approaches, this work informs on the treatment of liver inflammation models through the BET family of bromodomains.

Table 2. Summary of Selectivity, Cellular Activity, and Physiochemical Properties for BRD4-D1 Selective Inhibitors.

	1	2	39	41
BRD4-D1 IC50 (μM)	0.14 ±0.05	0.27 ±0.05	0.039 ±0.017	0.32 ±0.12
BRD4-D1 Ki,app (μM) % Inhibition @ 10 μM	0.503 ±0.089a (39 ±8%)a	0.93 ±0.09a (36 ±8%)a	0.73 ±0.04 (67 ±7%)	0.63 ±0.20 (87.5 ±0.1%)
TSEC Viability, IC50 (μM)	--	194 μM	97 μM	194 μM
hERG IC50 (μM)	0.97	<0.10	0.90	8.43
Kinetic Solubility (μM, pH 6.0)	920 ±20	n.d.	250 ±30	n.d.
LogD7.4	4.33 ±0.04	4.71 ±0.09	2.70 ±0.10	5.26 ±0.06
Selectivity over BET Bromodomains	23–83 fold for D1sb >590 fold for D2sb	n.d.	22–1700 for D1s, > 100-fold for D2s	64-fold for BRD3(D1) 38-fold for BRD4(D2)

BRD4-D1 IC₅₀ determined through fluorescent anisotropy, where K_i,app was determined through NanoBRET. Selectivity of each compound determined through AlphaScreen from ReactionBiology. n.d. = not determined.

^aPreviously reported in Doskey et al.[23] through BRD4-D1 NanoBRET assay.

^bPreviously reported in Divakaran et al.[32]

Highlights:

• Highly selective BRD4-D1 inhibitors are optimized for cellular engagement

• Structure-activity relationships lead to the potent compound 39 (K_i = 2.4 ±1.0 nM)

• Acetylated analogue, 41, has 80-fold less hERG activity than previous scaffolds

• BRD4-D1 inhibition is significant but insufficient for CXCL1 and CCL2 reduction

Glossary

BET

bromodomain and extra-terminal

BRD

Bromodomain-containing protein

CCL2

C-C motif chemokine ligand 2

CETSA

cellular engagement thermal shift assay

CFU-Mk

colony forming megakaryocyte assay

CXCL1

growth regulated oncogene-1

DSF

differential scanning fluorimetry

FA

fluorescence anisotropy

hERG

human ether a-go-go related gene

IL-6

interleukin 6

INF-γ

interferon gamma

JAG1

jagged 1

LipE

lipophilic efficiency

LPS

lipopolysaccharide

NanoBRET

nano-luciferase bioluminescence resonance energy transfer

NF-κB

nuclear factors-kappa B

PFI

property forecast index

PROTAC

proteolysis targeting chimera

PTEF-b

positive transcription elongation factor b

siRNA

small interfering RNA

TLR3

toll-like receptor 3

TNF-α

tumor necrosis factor alpha

TSEC

transformed liver sinusoidal endothelial cells

6. Availability of Data and Material

All data generated or analyzed in this study are included in this published article and its supplementary information file. Co-crystal structures have been uploaded to the Protein Data Bank (PDB: 9P33 and 9P34).