Evaluating the Advantages of Using 3D-Enriched Fragments for Targeting BET Bromodomains

Abstract

Fragment-based ligand discovery has been successful in targeting diverse proteins. Despite drug-like molecules having more 3D character, traditional fragment libraries are largely composed of flat, aromatic fragments. The use of 3D-enriched fragments for enhancing library diversity is underexplored especially against protein–protein interactions. Here, we evaluate using 3D-enriched fragments against bromodomains. Bromodomains are highly ligandable, but selectivity remains challenging, particularly for bromodomain and extraterminal (BET) family bromodomains. We screened a 3D-enriched fragment library against BRD4(D1) via ¹H CPMG NMR with a protein-observed ¹⁹F NMR secondary assay. The screen led to 29% of the hits that are selective over two related bromodomains, BRDT(D1) and BPTF, and the identification of underrepresented chemical bromodomain inhibitor scaffolds. Initial structure–activity relationship studies guided by X-ray crystallography led to a ligand-efficient thiazepane, with good selectivity and affinity for BET bromodomains. These results suggest that the incorporation of 3D-enriched fragments to increase library diversity can benefit bromodomain screening.

Fragment-based ligand discovery (FBLD) programs are well-established in both academic and industrial settings. Three FDA approved drugs vemurafenib, erdafitinib, and venetoclax originate from fragment screens and illustrate that FBLD is useful for targeting various classes of proteins including protein–protein interactions (PPIs).¹ FBLD screening takes advantage of libraries that have less complex compounds (MW < 300 g/mol) to effectively sample chemical space¹ and thus reduce library sizes needed for screens. Fragment screens often have higher hit rates than high-throughput screens and are therefore used for protein ligandability analyses.^2,3

Although chemical probes and drugs have ample 3D character, traditional fragment libraries are composed of flat, aromatic-rich compounds, which we refer to here as 2D fragments.^4,5 Interest in evaluating more diverse, 3D-enriched fragments for increasing diversity within fragment libraries has emerged, although the potential advantages are still debated.⁶⁻⁹ With the increase in complexity, there is concern that 3D-enriched fragments will reduce screening hit rates.¹⁰ In contrast, in a retrospective analysis, AstraZeneca found that roughly half of the hits in three of their PPI campaigns had significant 3D character, and these hits more effectively fill the PPI binding pocket than flatter fragments. On the basis of these findings, they recommended a balance between 3D and 2D fragments to efficiently sample chemical space.¹¹ However, a broad claim of the usefulness of 3D-enriched fragments against PPIs is difficult, given the varied size, plasticity, and shape of PPI interfaces.¹² Here we evaluate the benefits of using 3D-enriched fragments for targeting bromodomains, a PPI class of therapeutic value, with more defined but highly similar binding sites.¹²

Of the 61 human bromodomains, the bromodomain and extraterminal (BET) domain family is the most studied. Each protein (BRD2, BRD3, BRD4, and BRDT) has two N-terminal bromodomains (D1 and D2). Aberrant levels and activity of BET bromodomains have been associated with many diseases including cancer, inflammation, and heart disease.¹³ Achieving selectivity among the BET family of bromodomains has proven challenging. As such, the majority of bromodomain inhibitors are pan-BET inhibitors including those developed from fragment screening.

We describe here the results of a 467-compound 3D-enriched fragment screen against BRD4(D1). BRD4(D1) is a good model system for analysis because it is highly ligandable¹⁴ and is suitable for crystallization to analyze fragment binding modes. Additionally, it represents a selectivity challenge due to the similarity in binding site between BET bromodomains.

We previously conducted an NMR-based fragment screen against BRD4(D1) using a traditional fragment library facilitating a more direct comparison with this work.¹⁵ Prior to screening with our new 3D-enriched fragment library, we analyzed the chemical diversity of our prior library based on plane of best fit (PBF) and fraction sp³ (Fsp³) (Figure 1) to describe fragment shapeliness. Fsp³ is the number of sp³ centers/total number of carbon atoms. Fsp³ can be misleading as compounds with a high Fsp³ character can be flat and vice versa. PBF is the average distance of a non-hydrogen atom from a plane drawn through the compound such as to minimize the average.¹⁶ AstraZeneca defined 3D compounds as compounds with a PBF ≥ 0.25.¹¹ We chose a more stringent cutoff with a PBF ≥ 0.3.

Figure 1.

(A) PBF and Fsp³ comparison of the 3D-enriched library and the proprietary traditional fragment library from Urick et al.¹⁵ and hits. (B) PBF analysis of 3D fragment 1, and a more 2D fragment, 2.

Analysis of our prior fragment screen by Fsp³ and PBF showed our previous library and fragment hits to be relatively flat with an average PBF of 0.26 and 0.17 and Fsp³ 0.182 and 0.143, respectively, consistent with many traditional fragment libraries. The average PBF and Fsp³ of our current library were 0.46 and 0.451, respectively. Thus, our 3D-enriched library is composed of fragments that have more 3D character than traditional fragment libraries but is not solely 3D fragments to ensure library diversity.

To evaluate if a trend toward identifying flat BRD4(D1) ligands in screens versus more 3D-enriched fragments was consistent within a broader context, we compared fragments from our prior screen to fragments reported in the PDB for BRD4. In this case, the PDB fragments possessed more 3D character with an average PBF of 0.36. These analyses provided further motivation for evaluating a 3D-enriched fragment library targeting BRD4.

Our prior screen used ligand-observed ¹H CPMG NMR to identify BRD4(D1) ligands using our proprietary traditional fragment library.¹⁵ Protein-observed fluorine (PrOF) NMR¹⁷ was used as a follow-up assay, with 85% agreement between the two assays. Here we use the same screening methods allowing for an accurate comparison. A competition ¹H CPMG NMR assay using pan-BET inhibitor (+)-JQ1¹⁸ was also conducted to guard against false positives. From our screen, 34 competitive and 11 noncompetitive hits were identified. The noncompetitive hits were tested by the PrOF NMR assay but gave no response indicating they were likely ¹H CPMG NMR false positives and not allosteric binders.

PrOF NMR was subsequently used to determine the affinity of competitive hits to BRD4(D1). An advantage of PrOF NMR over ¹H CPMG is the ability to obtain K_d values for weak binders to rank-order compounds. Using PrOF NMR, we also determined selectivity against the bromodomains BRDT(D1) and BPTF. BRDT(D1) is a BET family bromodomain closely related to BRD4(D1) and thus a stringent test case for selectivity. BPTF is a non-BET family I bromodomain.¹⁹ BRD4(D1), BRDT(D1), and BPTF have a tryptophan near the acetylated lysine binding site in the WPF shelf, which was ¹⁹F-labeled for PrOF NMR (Figure 2).

Figure 2.

(A) Example PrOF NMR titration of 1 with BRD4(D1) (right). (B, C) Representative single point PrOF NMR spectra with 1 against BPTF (left) and BRDT(D1) (right) with corresponding protein structures and locations of the 5FW labels. (D) Fragment 1.

PrOF NMR titrations revealed a range of affinities from mid-μM to mM for competitive hits (Table 1). The K_d for two of the hits was unable to be determined due to poor solubility and protein behavior. Table 1 shows the fragments with the highest affinity (3–9, 1). Hits 4 and 5 have ligand efficiency ≥ 0.3, a benchmark for evaluating fragments as potential leads.²⁰ However, the fragments were tested as racemates. If the affinity was improved two-fold assuming only one enantiomer binds, 1 and 6 also have ligand efficiencies ≥ 0.3.

Table 1. Select NMR Screening Hits and SAR Follow-up.

NB is no binding, ND = not determined. ∗ = hit via ¹H CPMG but not PrOF NMR, & = not soluble at high concentration. The italic value represents the ligand efficiency if only one enantiomer binds. (a) K_d represents the average value of three experiments.

A single point PrOF NMR assay of hits against BRDT(D1) and BPTF was done as a preliminary selectivity assessment. A chemical shift change >0.03 ppm at 400 μM fragment constituted binding as previously established.¹⁵ The majority of hits (23, 62%) bound BRDT(D1) or BPTF (Table 2). However, 10 (29%) only bound BRD4(D1) (Table 1, 7, 10–16, Table S3). The hits that were selective for BRD4(D1) over BRDT(D1) and BPTF were tested via ¹H CPMG NMR against a third BET bromodomain, BRD4(D2). 7 had the highest selectivity with only a minimal effect on BRD4(D2). Together, these fragments represent BRD4 selective leads. Having selectivity at the start of the inhibitor development campaign could be used in the future as scaffolds for lead generation for isoform selective inhibitor development.

Table 2. Distribution of Hit Affinities and Selectivities.

affinity of hits (μM)	number of hits
Kd < 100	5 (15%)
100 < Kd < 300	5 (15%)
300 < Kd < 600	4 (12%)
600 < Kd < 1000	6 (18%)
Kd > 1000	6 (18%)
bind via 1H CPMG but not PrOF NMR	6 (18%)
unable to determine	2 (6%)

proteins hits bind
BRD4(D1)	11 (32%)
BRD4(D1), BRDT(D1)	9 (26%)
BRD4(D1), BPTF	4 (12%)
BRD4(D1), BRDT(D1), BPTF	10 (29%)

The 3D-enriched fragment screen resulted in known bromodomain inhibitor scaffolds but also revealed unexplored chemical matter for bromodomain inhibitors. For example, ureas such as 3, 9, 10, and 13 have been reported to bind bromodomains.²¹ Additionally, the pyrazole motif in 7 and 8 and the pyrimidine in 15 are found in bromodomain chemical probes, BAZ-ICR and CPI-637.²² Alternatively, one underrepresented scaffold in fragment libraries is the seven-membered thiazepane, 1. Giordanetto et al. recently noted a lack of seven-membered ring fragments in screening libraries and recommended their inclusion for increasing diversity.²³

The non-ring-fused thiazepane framework has not been reported as a bromodomain inhibitor scaffold. Thus, 1 is of interest because it bears the same seven-membered ring as benzodiazepines and benzodiazepinones such as ((+)-JQ1)¹⁸ and (CPI-637),²⁴ privileged bromodomain inhibitor scaffolds, but can access wider conformational space. Further supporting the lack of exploration of thiazepanes is the absence of literature reporting a high yielding synthesis for library generation. Thus, we decided to do a structure–activity relationship (SAR) by commerce study with molecules 17–29 to investigate the thiazepane scaffold. An optimized synthetic route will be reported in due course.

In addition to defining the stereochemical importance, there are four main areas for SAR with thiazepane 9 to tune affinity and selectivity: the aromatic group, the oxidation state of the thioether, the acetylated lysine mimic, and the stereocenter at position 7 (Table 1). Using commercial fragments, a range of affinities for BRD4(D1) was determined (K_d = 230–2600 μM). The initial SAR was sensitive to the size and type of N-acyl group. Fragments 28 and 29 indicate that large groups on the acetylated lysine mimic are not tolerated. Intermediate-sized acyl groups were preferred including ethyl carbamates (17, 19), cyclopropyl amides (18), and cyclobutyl amides (20). When the position of the noncarbonyl oxygen is moved from the carbamate 19 to the ether 27, binding is abolished. This may be due to the decrease in hydrogen bond accepting ability of the amide. The N-methyl amide in 21, bound weaker (K_d = 510 μM), but maintained a high ligand efficiency of 0.3 and selectivity. No binding was detected for the BRDT(D1) bromodomain or non-BET BPTF bromodomain by PrOF NMR. The cyclopropyl amide-containing thiazepane, 22 showed similar selectivity, but a reduced ligand efficiency.

We next evaluated the thioether oxidation state by comparing the affinities of 16, 17, and 24. Comparison of 17 and 26 shows oxidization of the thioether to the sulfone precludes binding. At the intermediate oxidation state, sulfoxide (24a, 24b), only one set of diastereomers binds BRD4(D1) (24a, K_d = 2600 μM), albeit weakly.

In contrast to the more robust SAR of the nitrogen acylation state, the effects of different aromatic rings on binding were less significant. When the acetylated lysine mimic is an ethyl carbamate or cyclopropyl amide various aromatic rings, including thiophenes, furans, and phenyl groups (1, 17, 18, and 19) bind with similar affinity (K_d = 230–280 μM.) However, only a limited set of analogs was studied.

To study how the thiazepanes engage BRD4(D1), we obtained a cocrystal structure of 1 and 21 (Figures 3, S3, S4, and S7). For both 1 and 21, the S-enantiomer crystallized (Figures S3 and S4). The carbonyl of both thiazepanes serves as the acetylated lysine mimic, anchoring binding through a hydrogen bond to N140 and a water-mediated hydrogen bond to Y97 (Figure 3a). The carbonyls of 1 and 21 overlay well with the native acetylated histone H4K5acK8ac (Figure S7).²⁵ The aromatic ring of the thiazepanes approaches the WPF shelf making hydrophobic interactions. In comparing the binding pose of 1 to the pan-BET inhibitor (+)-JQ1, the seven-membered rings are in different locations in the binding pocket giving insight into possible vectors for future SAR. The ethyl carbamate of 1 extends deeper into the binding pocket than the methyl of (+)-JQ1 (Figure 3b). The thiophene of the S-enantiomer of 1 has the same vector as the chlorophenyl group of (+)-JQ1 extending toward the WPF shelf. Future analogs can be designed based on this aromatic ring.

Figure 3.

(A) Key interactions of 1 with BRD4(D1). PDB 6UWX. Structured waters are indicated as red spheres. (B) Overlay of 1 (gray) with (+)-JQ1 (cyan). (C) Key interactions of 30 with BRD4(D1). PDB 6UVJ. (D) Overlays of 1 (green) and 30 (purple).

Given the potency of intermediate-sized acyl groups and the selectivity of the methyl amide, 21, we tested a final analog of 1 with a shorter methyl carbamate, 30. This compound was not commercially available but could be obtained through modification of a synthesis by Wamhoff et al.²⁶ to yield the racemic product. We tested the binding of 30 to BRD4(D1) by PrOF NMR and found it to be our most potent and ligand efficient compound with a K_d of 32 μM and 0.39 ligand efficiency. The enantiomers were separated; enantiomer (−)-30 contributed most significantly to affinity with a K_d of 20 μM (LE = 0.40), whereas the other enantiomer (+)-30 bound 16-fold weaker. Co-crystallization of 30 (Figure 3c,d) showed the S-enantiomer to bind, which we tentatively assign to (−)-30a given our binding studies. The overlay of 30 with 1 indicates that the methyl carbamate does not extend as deeply into the hydrophobic binding pocket and may thus not suffer as much of a desolvation penalty. The reduced number of rotatable bonds may also account for the increased affinity.

Finally, we tested the selectivity of 30, (+)-30, and (−)-30 against a broader panel of bromodomains (Table 4). For the most similar bromodomain, BRDT(D1), a moderate four-fold selectivity was measured. No binding was detected for non-BET human bromodomains that contain a WPF shelf including PCAF, CECR2, and BPTF. Weak affinity was measured for a nonhuman bromodomain, which also contains a WPF shelf, PfGCN5, K_d = 310 μM. Together, these results based on selectivity and affinity support the thiazepane as a new scaffold lead targeting BET bromodomains.

Table 4. Binding Affinities of 30, (+)-30, and (−)-30 for a Panel for Bromodomains.

fragment	protein	Kd (μM)
30	BRD4(D1)	32 ± 30a
(−)-30	BRD4(D1)	20 ± 4a
(+)-30	BRD4(D1)	320
(−)-30	BRDT(D1)	83
(+)-30	BRDT(D1)	270
30	BPTF	NBb
30	PCAF	NBc
30	PfGCN5	310
30	CECR2	NBb

^aThis K_d represents the average value of three independent experiments.

^bNo binding observed at 400 μM fragment (Δδ < 0.03 ppm).

^cNo binding observed at 250 μM fragment (Δδ < 0.03 ppm)

Having established a selectivity and affinity analysis, we compared the 3D-enriched screen to our prior fragment screen.¹⁵ Given the concerns of a low hit rate and the unknown necessity of 3D-enriched fragments for targeting bromodomains, a hit rate comparison and 3D analysis of hits was done. Urick et al.¹⁵ used ligand-observed ¹H CPMG to screen 930 fragments against BRD4(D1). To ensure a rigorous hit rate comparison, the screening method, screening conditions, and definition of a hit were adopted for the 3D-enriched fragment screen. To our knowledge, this is the first direct comparison of a traditional to a 3D-enriched library using the same assay and same protein target. As anticipated, increasing the complexity of the fragments via 3D character resulted in a lower hit rate than a highly 2D-enriched library. The overall hit rate for the traditional fragments is 2.5-times that of the 3D-enriched fragments, 25% and 10%, respectively. The ratio of total hits to competitive hits was similar with competitive hit rates of 20% and 7% for the traditional and 3D-enriched screens, respectively. Although lower than the traditional library, the 7% competitive hit rate remains high giving 34 competitive hits to follow-up. The lower hit rate may be affected by the higher average molecular weight of our library (241 g/mol vs 180 g/mol, Table S2). Therefore, this reduction in hit rate is a conservative number. These results indicate that future 3D-enriched fragment screens against bromodomains may not necessarily suffer from low hit rates.

An analysis was further carried out to explore the 3D character of the hits of both the 3D and traditional screens compared to the overall library. Analysis of the 3D-enriched and the proprietary library confirmed a difference in library shape composition (Figure 4). Because of a wide range of values, the average Fsp³ and PBF are within the standard deviation of each other (Table S1). However, as shown in the box plot and calculated via the nonparametric Mann–Whitney–Wilcoxon (MWW) test, the distributions of PBF between the traditional proprietary library and 3D-enriched library are different, with the proprietary library being enriched in 2D character. The 3D-enriched hits are similar in overall 3D character of the entire library (Figure 4b). These results indicate that the hits are not selecting for 3D or 2D fragments but are a representative sample from the overall library. In contrast, the proprietary hits are more 2D-enriched than the overall library. When we compare these results to our analysis of the PDB fragment hits for BRD4, we find that both our 3D-enriched library composition and the resulting fragment hits are more three-dimensional than those in our PDB search (average PBF: 0.36 vs 0.44). However, the PDB fragment hits cover a wider diversity of fragment shapeliness, supporting the argument for increasing the diversity of 3D fragments in libraries for bromodomain targeting but not at the expense of more 2D fragments.

Figure 4.

Summary of PBF distribution of the 3D- and 2D-enriched libraries, and BRD4-binding fragments (MW < 300 g/mol) in the PDB. (A) Boxplot showing the distribution of PBF. Average PBF values are black. (B) Quantitative results of the Mann–Whitney–Wilcoxon (MWW) test showing the differences in PBF.

Finally, the 3D character of the hits with the highest affinity and selectivity was analyzed (Tables 1 and S3). With PBFs ranging from 0 (6) to 0.86 (8), there is no correlation between affinity and 3D character. As with affinity, there is a range of PBF values for the BRD4(D1) selective hits from 0.29 to 0.76. However, only two hits, 14 and 15, have a PBF below 0.3, indicating that most of the selective hits have substantial 3D character.

In conclusion, we have completed a 3D-enriched fragment screen against BRD4(D1). A range of affinities and selectivity among hits was observed. The hit rate was lower than a 2D-enriched fragment screen, along with lower ligand efficiencies. However, there were several novel hits with a significant level of selectivity against similar BET bromodomains. Because the binding site is conserved among bromodomains, the hit rate is likely to be similar for future 3D-enriched fragments among highly ligandable bromodomains. Readily diversified scaffolds with seven-membered rings such as the ligand-efficient thiazepane 30 that are not likely to be present in traditional screening libraries were also identified.²³ Finally, despite our prior screen, which led to flat fragment hits, hits were not enriched for 2D or 3D fragments. Taken together, we recommend including 3D-enriched fragments into bromodomain fragment screening libraries to gain access to underexplored scaffolds for inhibitor development.

Glossary

Abbreviations

FBLD

fragment-based ligand discovery

PPI

protein–protein interaction

BET

bromodomain and extra terminal domain

¹H CPMG NMR

proton Carr–Purcell–Meiboom–Gill NMR

PrOF NMR

protein observed ¹⁹F NMR

SAR

structure–activity relationship

PMI

principle moment of inertia

Fsp³

fraction sp³

MWW

Mann–Whitney–Wilcoxon.

PBF

plane of best fit

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.9b00414.

• NMR and FP experimental data; library and screen analyses; general experimental procedures and analytical characterization; BRD4(D1) cocrystal structural analysis of 1, 18, and 30 (PDF)

National Institutes of Health Biotechnology training Grant No. 5T32GM008347–23. Lilly Research Award Program. NIH (GM118047).

The authors declare no competing financial interest.