A Qualified Success: Discovery of a New Series of ATAD2

Aug 9, 2019 - The bromodomain of ATAD2 has proved to be one of the least-tractable proteins within this target class. Here, we describe the discovery ...
0 downloads 0 Views 6MB Size
Download PDF
Article Cite This: J. Med. Chem. XXXX, XXX, XXX−XXX

pubs.acs.org/jmc

Downloaded via NOTTINGHAM TRENT UNIV on August 10, 2019 at 08:45:17 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

A Qualified Success: Discovery of a New Series of ATAD2 Bromodomain Inhibitors with a Novel Binding Mode Using HighThroughput Screening and Hit Qualification Paul Bamborough,*,† Chun-wa Chung,*,† Emmanuel H. Demont,† Angela M. Bridges,† Peter D. Craggs,† David P. Dixon,† Peter Francis,† Rebecca C. Furze,‡ Paola Grandi,§ Emma J. Jones,† Bhumika Karamshi,† Kelly Locke,† Simon C. C. Lucas,† Anne-Marie Michon,§ Darren J. Mitchell,† Peter Pogány,† Rab K. Prinjha,‡ Christina Rau,§ Ana Maria Roa,∥ Andrew D. Roberts,† Robert J. Sheppard,‡,⊥ and Robert J. Watson† †

Medicinal Science and Technology and ‡Epigenetics Research Unit, GlaxoSmithKline Research and Development, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, United Kingdom § Cellzome, Meyerhofstrasse 1, Heidelberg 69117, Germany ∥ GlaxoSmithKline Tres Cantos, 28760 Tres Cantos, Madrid, Spain S Supporting Information *

ABSTRACT: The bromodomain of ATAD2 has proved to be one of the least-tractable proteins within this target class. Here, we describe the discovery of a new class of inhibitors by high-throughput screening and show how the difficulties encountered in establishing a screening triage capable of finding progressible hits were overcome by data-driven optimization. Despite the prevalence of nonspecific hits and an exceptionally low progressible hit rate (0.001%), our optimized hit qualification strategy employing orthogonal biophysical methods enabled us to identify a single active series. The compounds have a novel ATAD2 binding mode with noncanonical features including the displacement of all conserved water molecules within the active site and a halogen-bonding interaction. In addition to reporting this new series and preliminary structure−activity relationship, we demonstrate the value of diversity screening to complement the knowledge-based approach used in our previous ATAD2 work. We also exemplify tactics that can increase the chance of success when seeking new chemical starting points for novel and lesstractable targets.

■

INTRODUCTION

activity relationship (SAR) models, etc.), may be of limited utility. High-throughput screening (HTS), in which large and well-curated diverse compound libraries are screened, is unbiased by expectations and gives greater opportunities to find novel starting points and modes of action. The quality of chemical starting points is a significant factor that influences the chance of success of a drug discovery project. The importance of the physicochemical properties of leads and of remaining within druglike space during optimization is well

Many drug discovery efforts focus on families or classes that have been studied before and are believed to be tractable to smallmolecule inhibition (for example, kinases and G proteincoupled receptors). Although these are more likely to be druggable targets, it can still be challenging to find progressible hits against individual members of established ligandable protein classes. This is especially true if the aim is to find novel modes of action, perhaps to overcome issues of selectivity within the protein family. When novelty is a key objective, approaches that rely on prior knowledge of important ligand binding features (virtual screening, machine learning/quantitative structure− © XXXX American Chemical Society

Received: April 23, 2019

A

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

established.1 Sometimes, starting points for medicinal chemistry are suboptimal because no good small-molecule binding sites exist or the sites that do exist have a strong preference for chemotypes that are difficult to develop into drugs. These may be inevitable consequences of the choice of target. However, the quality of leads may also be suboptimal for avoidable reasons, for example, by poor choices made in the design and execution of the screening cascade. In such cases, good starting points may be hidden among a sea of assay hits operating through unprogressible mechanisms. Such mechanisms may be highly assay-dependent and include phenomena such as optical interference, polystoichiometry or aggregation, redox recycling, covalent modification, etc.2,3 Substructural motifs associated with an increased probability of assay interference can provide a warning flag, but these only cover a fraction of all possible unprogressible mechanisms of inhibition.4 HTS assays are routinely optimized to ensure reproducibility, to yield the expected pharmacology with known control compounds, and to produce a low enough hit rate to avoid overwhelming downstream assays. As it is well known that all screening assays have artefactual positives, confidence to initiate a medicinal chemistry optimization effort has often been gained through a hit disqualification triage process (Figure 1a), in which compounds with unwanted modes of action are sequentially removed from the hit list by selected computational and experimental counterscreens. The type and proportion of mechanistically unprogressible hits can be very reagent- and assay-dependent, so the most effective choice of counterscreens may be unclear at the start of a project. Furthermore, the number

of hits acting through such mechanisms is often underestimated, and in our experience there are many cases where these greatly outnumber hits where the activity results from desirable mechanisms. When combined with low biological tractability, numerous unprogressible hits must be invalidated before any progressible ones are found. This takes time and effort, can undermine confidence in the target, and may even result in program termination. The fragment-based drug design community often employs an alternative positive selection strategy. Faced with the expensive and time-consuming optimization of weak hits from high-concentration screens, it is common to progress only those that come from the intersection of orthogonal biophysical screening methods. These techniques can now be used to test thousands of compounds, a scale compatible with the number of compounds typically tested in dose−response after an HTS.5 Here, we demonstrate the power of using confirmation (orthogonal) screens after HTS, rather than counterscreens a strategy of hit qualification, as opposed to hit disqualification (Figure 1). Using a positive inclusion process, and deemphasizing assays intended to exclude possible unwanted mechanisms, we avoid unwarranted assumptions about what those mechanisms might be. Since compounds are progressed only if they demonstrate desirable attributes, after each assay, our knowledge of and confidence in the remaining hits increases. For example, direct interaction between the compound and target protein can be probed by label-free biophysical approaches such as NMR, surface plasmon resonance (SPR), affinity selection mass spectroscopy, or thermal shift assays (TSAs).6 Similarly, sensible binding stoichiometry can be confirmed by native mass spectrometry, SPR, or isothermal calorimetry (ITC). This approach is particularly useful for less-tractable targets where the proportion of developable hits is small and the number of potential undesirable mechanisms is large. We also show the value of the routine use of hit qualification approaches before the screen itself to ensure the entire triage is fit for purpose. We illustrate the application of hit qualification here using the bromodomain of ATAD2 as an example. ATAD2 or AAA nuclear coregulator cancer-associated protein (ANCCA) is one of over 40 human bromodomain-containing proteins (BRDs). Highly expressed in a wide variety of unrelated cancers, increased expression levels correlate with poor prognosis and disease recurrence in patients.7−10 Downregulation of its expression using siRNA has implicated ATAD2 in pathways including apoptosis, cell survival, proliferation, and migration.7,9,11−13 Drug discovery efforts for oncology have so far focused on the ATAD2 bromodomain rather than its ATPase domain,14 as bromodomains, especially the bromodomain and extraterminal (BET) subfamily, are generally considered to be tractable to most screening approaches.15−17 Compared to the more extensively studied BET family proteins, relatively few inhibitors of ATAD2 have so far been reported (Figure 2).18−26 Although the KAc binding site of ATAD2 retains the features of typical bromodomains, computational predictions suggest that it is less amenable to smallmolecule inhibition than others in the family.20,27 In our hands, ATAD2 has indeed been less tractable than other bromodomains, resulting in lower qualified hit rates from bromodomaintargeted and diverse fragment sets. We have previously reported the discovery of a series of naphthyridone ATAD2 inhibitors. Screening a library of compounds designed to interact with the conserved acetyllysine (KAc) site of the bromodomain family led to GSK8814

Figure 1. Summary of key messages, and schematic of disqualification (left) vs qualification (right) triage approaches to hit confirmation. B

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Figure 2. Some literature ATAD2 inhibitors and affinities reported using various techniques.18−26

Figure 3. (A) Typical HTS process at GSK. Following the primary screen at a single compound concentration and digital correction for plate patterns, typically ∼20 000 compounds are selected, replated, and the assay repeated in duplicate. If a second assay is available, either an orthogonal or an interference screen may be run in parallel. Following confirmation, a few thousand reproducible hits are normally progressed for full-curve IC50 determination and then into hit qualification. Computational approaches (e.g., clustering, property prediction) may be applied at various stages, although over-reliance on this risks losing progressible hits. If the number of hits greatly exceeds compound-handling capacity (for example, in situations with a high, reproducible hit rate acting through nonprogressible mechanisms), then performing in silico filtering without losing the few progressible hits may be particularly challenging. (B) Refined ATAD2 HTS triage process as described below.

(Figure 2).20−22 This knowledge-based fragment approach drew upon historical and structural understanding of bromodomains and their inhibitors. Here, we will describe our concurrent attempt to find alternative, more unusual classes of ATAD2 bromodomain inhibitor by screening a large, diverse compound collection. We discuss the many challenges we faced and how they were diagnosed and overcome by data-driven choices, including our hit qualification triage. This led to the discovery of a single potent series out of the 1.7 million compounds screened.

we developed.20 Unfortunately, in early assay development trials, this protein showed temperature-dependent but reversible precipitation and batch-to-batch aggregation, which raised concerns about the suitability of this construct for highthroughput screening. Reasoning that neighboring regions of the sequence could be important for domain stability, we expressed and purified a longer sequence, 6H-Flag-tev-ATAD2 950−1148 (construct 2). This indeed provided a more robust reagent for SPR binding studies for both fragments and leadsized compounds.20 When used to screen panels of modified histone peptides for binding partners (AltaBioscience), a 21residue triacetylated histone peptide was identified with an affinity of ∼50 μM (Figure S1a,b, Supporting Information). This interaction was confirmed by isothermal titration calorimetry (Figure S1c, Supporting Information). This peptide was subsequently used to configure an ATAD2 time-resolved fluorescence (TRF) energy transfer (TR-FRET) binding assay. A typical HTS triage is shown in Figure 3a. Before carrying out the ATAD2 screen, the reproducibility and effectiveness of every

■

RESULTS Primary Screening Assay. At the start of the project, no tool compounds for ATAD2 were known, and the expression and purification of full-length ATAD2 had not been reported. A short bromodomain-containing construct (construct 1, ATAD2 981−1108) from the Structural Genomics Consortium formed the basis of the 15N−1H heteronuclear single quantum coherence (HSQC) NMR and X-ray crystallography systems C

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

stage of the triage were assessed by passing a robustness set (a panel of 1408 diverse compounds representative of the wider screening collection, filtered to remove known frequent hitters and pan-assay interference compounds)28,29 through the cascade from top to bottom. This set is large and diverse enough to provide an estimate of the expected number of compounds passing through all stages from the full HTS and to provide confidence that the capacity of the lower-throughput assays will not be overwhelmed. As the robustness set compounds were not designed to interact with ATAD2, our expectation was that most hits were likely to act through unprogressible mechanisms and therefore, perhaps counterintuitively, a low hit rate is highly desirable. In reality, the robustness set gave a high hit rate in the 6HFlag-tev-ATAD2 950−1148/peptide displacement TR-FRET assay. When screened at 10 μM, 4.1% of compounds were classed as active, at a high threshold of 50% inhibition (defined using a cutoff of 3 standard deviations (SDs)). If this translated into HTS, 70 000 hits would be generated, making it highly challenging to select a subset for progression through the lowerthroughput triage stages. It was not viable to reduce the hit rate simply by raising the activity threshold: achieving even a modest reduction to a 2% hit rate would require a cutoff of nearly 80% inhibition, leading to the exclusion of many potentially potent hits. In general, raising the cutoff is not advisable because (a) single-concentration HTS inhibition may not extrapolate well to IC50 curves and (b) the mode of action is undetermined at this stage and it is likely that progressible hits would be lost. A high primary hit rate is especially problematic if the primary screen is reproducible, in which case the confirmation stage is unlikely to deliver significant attrition (Figure 3a). Compound 1 was one representative hit from the robustness set, which was reproducibly active in dose−response in the TRFRET assay with an IC50 of ∼30 nM (Figure 4a). Noting its structural similarity to bromosporine (ATAD2 Kd 44 μM), we expected that 1 was more likely to show confirmed binding in hit qualification assays than other structurally novel hits.26,30 However, like the other hits from the robustness set, it showed no detectable binding to the bromodomain in direct binding assays such as 15N−1H HSQC NMR (Figure 4b), suggesting that its TR-FRET activity must arise through an unwanted mechanism. In this case, chemical intuition was no substitute for direct experimental validation as a reliable way to select hits with desirable modes of action. From the robustness set data, we concluded that although the longer 6H-Flag-tev-ATAD2 950−1148 protein was well behaved in some respects, it produced a high level of unprogressible positives that could not be confirmed by other means. Attempts to remove these unwanted hits by modifying assay conditions were unsuccessful (data not shown); so, we therefore turned to make alternative, more robust bromodomain constructs. Alternative ATAD2 Bromodomain Constructs. Alternative ATAD2 bromodomain-containing constructs were designed using bioinformatics analysis of human and orthologue sequences, informed by the X-ray crystal structure of the 981− 1108 core bromodomain module [Protein Data Bank (PDB) code 3DAI] and limited proteolysis of 6H-Flag-tev-ATAD2 950−1148 (constructs 3−15, Table S1, Supporting Information).31 Following small-scale expression trials and purification, eight constructs yielded soluble proteins (constructs 3−10, Table 1) with good integrity and purity, as confirmed by peptide mass fingerprint and liquid chromatography/mass spectrometry

Figure 4. (A) TR-FRET dose−response curve for 1 in the 6H-Flag-tevATAD2(950−1148) peptide displacement assay. Compound 1 did not give fitted curves up to 10 μM in the optimized flag-6H-tevATAD2(981−1121) assay (data not shown). (B) 15N−1H HSQC NMR spectrum (black in the presence of 100 μM 1 and red without) showing lack of evidence for compound binding.

(LC/MS). These were characterized by a range of approaches. An assessment of helical content and stability was made using thermal melting via circular dichroism (CD). CD spectra consistent with the expected bromodomain four-helical bundle were obtained (Table 1; Figure S2, Supporting Information). Most constructs had similar melting temperatures, while a designed hexamutant (construct 4) was more stable and another (construct 8) was biphasic. We note in passing that the design of the hexamutant construct was partly intended to remove cysteine residues, which we speculated could influence the stability of the bromodomain, including the neighboring pair Cys1057/Cys1079, an idea that has recently been discussed by others.32 In parallel, the ability of each protein to bind unlabeled acetyllysine histone H4 peptide was confirmed by competition with biotinylated peptide in TR-FRET assays (Figure S3, Supporting Information). These peptide displacement assays were then used to screen the 1408 compound robustness set at a 10 μM compound concentration. As mentioned above, this should contain few if any genuine ATAD2 binding molecules, so a low hit rate was desirable. Seven of the eight additional constructs 3−10 showed lower hit rates than construct 2 (6H-Flag-tevD

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Table 1. TR-FRET Peptide Competition Binding Assay Results from Compounds Screened at 10 μM against Constructs 2−10 (See Supporting Information Table S1 for Full Data)a

a

CD summarizes the thermal melting curves in Figure S2, Supporting Information. The robustness cutoff is the % inhibition cutoff required to give a 2% hit rate from a set of 1408 diverse compounds. The bromodomain toolset r2 column reports the correlation between n = 2 replicates from a bromodomain-targeted set of 998 compounds. Green = positive result; amber = warning; red = negative result.

concentration−response up to 100 μM. Compounds were classed as binding, nonbinding, or linear binding (a nonsaturating, polystoichiometric response where the calculated Kd was greater than the highest assay concentration). The ability of SPR to screen multiple proteins simultaneously was also useful, as it enabled us to determine differential interaction with wildtype ATAD2 over a mutant bromodomain that lacks the essential acetyl-lysine binding motif (Y1021A/N1064A). The optimal construct for SPR was found to be construct 2 (6H-Flagtev-ATAD2 950−1148). Construct 6 (6H-Flag-tev-ATAD2 981−1121) was also attempted, but this immobilized protein did not show saturating binding or differences between wildtype and mutant ATAD2 for tool compounds (data not shown). The vast majority of the HTS hits did not show progressible binding behavior in the biophysical assays. Only 26 of the 428 hits showed saturating binding behavior by SPR (Table 2). The others either did not bind at all or were linear binders. When the same 428 hits were classified independently by NMR, only 17 showed evidence of binding at 100 μM. Example SPR and NMR results for a compound showing saturating binding and for a linear binding hit are shown in Supporting Information Figures S7 and S8. Reassuringly, there was very good agreement between the NMR and SPR results, with 401/428 hits classified as either nonbinding or polystoichiometric by both techniques. Only one compound classified as binding by NMR failed to show saturating binding by SPR, and only 10 hits exhibiting saturating binding by SPR failed to show binding by NMR. The discrepant compounds had weak biochemical activity (TR-FRET IC50 > 30 μM), so they may lie on the borderline of detectability within the two assays at the highest concentration tested (100 μM). Furthermore, for every linear SPR hit, the binding response with mutant ATAD2 was similar to that with wild type, suggesting that they did not bind at the acetyl-lysine site (Figure 6). This could mean that these compounds bind in an allosteric site, but more likely they may interact with the SPR chip or with sticky patches on the protein in a nonspecific manner. In contrast, all but one of the compounds classed as binders by NMR showed preferential SPR binding to wild-type over mutant ATAD2 (Figure 6), even if not all of these reached saturation at 100 μM. For additional characterization and to confirm binding to fulllength endogenous ATAD2, we employed the bromosphere assay.21 This is a chemoproteomic assay measuring the competitive inhibition of endogenous ATAD2 from Hut-78

ATAD2 950−1148; Table 1; Figures 5a and S4). Three of the new constructs (3−5) had higher hit rates than the others and so were removed from further consideration. Interestingly, there was no agreement in the hits identified from the robustness set even between the well-behaved constructs 6 and 8−10 (Table 1; Figures 5b and S5), consistent with our assumption that these are unlikely to be specific progressible inhibitors. Constructs 6 and 10 also showed TR-FRET dose−response curves for three presumed tool compounds found in parallel activities and believed to be genuine inhibitors as these were consistently active in biophysical assays. We therefore tested a larger knowledge-based set of 996 compounds resembling known bromodomain inhibitors (the “bromodomain toolset”) at 10 μM against constructs 6 and 10 and found that construct 6, Flag-6H-tev-ATAD2 981−1121, gave more reproducible results than construct 10 (Table 1; Figures 5c and S6). Since construct 6 also had a lower-predicted protein consumption, we opted to use this for our primary TR-FRET screen. The optimized form of this assay was subsequently used to identify our previously reported naphthyridinone series.20 HTS and Hit Qualification. A summary of the outcome of the TR-FRET HTS and triage is shown in Figure 1B. Compounds (1.7 million) were screened, and statistical plate pattern correction was carried out.33 Hits were defined as those with inhibition greater than 3SD from the mean (∼29% inhibition) giving 9441 hits (a hit rate of 0.6%). Because of the small number, all could be progressed to confirmation by TRFRET. The more potent and chemically attractive compounds were taken straight to dose−response, while the more borderline hits went first for duplicate retest at 10 μM. Most of the 3354 compounds screened in dose−response showed little concentration dependence and low maximum inhibition and so did not give fittable curves (IC50 > 30 μM). Compounds (428) showing full or partial inhibition were progressed into hit qualification by orthogonal biophysical binding assays. Two independent assays to monitor the direct binding of compounds to the ATAD2 bromodomain were developed in parallel, as we were unsure whether either would be successful, and, in fact, each proved to be complementary to the other. One was 15N−1H HSQC NMR using tag-cleaved 15N ATAD2 981− 1108, which we have described previously.20 After testing at a single concentration (100 μM), each compound was categorized as either binding or nonbinding, and any effects on protein aggregation were noted. The other technology was surface plasmon resonance (SPR), in which hits were tested in E

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Table 2. Hit Qualification Results from Biophysical Assays Showing the Number of Compounds Classified as Binding/ Nonbinding by 15N−1H HSQC NMR and as Binding/Linear Binding/Nonbinding by SPR

Figure 6. SPR binding responses for all hits for mutant ATAD2 (X axis) vs wild-type (Y axis). Hits were analyzed at 11 μM and signals normalized to correct for differences due to molecular weight. Hits are colored according to their 15N−1H HSQC NMR binding class (red = binding, blue = nonbinding). All hits showing preferential binding to wild-type over mutant ATAD2 are the same as those that show binding by NMR.

binders by both NMR and SPR. All of them belonged to a single series of phenylsulfonamides, which had been purchased several years previously to populate the HTS collection. An example, 2, and its summarized biophysical data are shown in Figure 7. Summary data for the rest of the hits is given in Tables 3 and 4. Structure−Activity Relationships. The most potent qualified hits had single-digit micromolar IC50 (pIC50 5.0− 6.0) in the ATAD2 TR-FRET assay (Table 3). Selectivity against the BET family bromodomains is desirable to avoid the extensive pharmacology associated with their inhibition, and this was measured using TR-FRET fluoroligand displacement assays in mutant tandem BET bromodomains.20 Below 50 μM, none of the phenylsulfonamides showed significant binding to BRD4, which we use as a representative of the rest of the BET family (Table 3; full data in Table S2, Supporting Information). ATAD2 SPR dissociation constants were consistent with the ATAD2 TR-FRET IC50s (Table 4). The hits contained little variation at the R1 position, with only chloride and methyl substituents represented. The chloride analogues mostly had better potency; for example, chloride 2 is about 10-fold more potent than its direct methyl analogue 12. Variants at the R2 position included simple cyclic and

Figure 5. % inhibition responses at 10 μM from the diverse robustness set for ATAD2 construct 6 against (A) construct 2 and (B) construct 10. (C) Reproducibility of the 996 targeted bromodomain compounds screened in duplicate against ATAD2 construct 6.

chromatin lysate using a bead-linked ATAD2 bromodomain partner, in this case, an acetylated lysine histone peptide. Thirtyfour hits were chosen to test, mostly ones that were positive in the TR-FRET assay but nonbinding by biophysics. All compounds tested were inactive in the bromosphere assay, except for three, which were positive by SPR and NMR (Table 4). The three active compounds have lower potency in the bromosphere assay than in TR-FRET or SPR, possibly because of nonspecific binding to other components of the lysate or reduced affinity for full-length ATAD2. In summary, after extensive characterization, 16 hits from the 1.7 million screened were confirmed to be specific, saturating F

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

bromide 35 using either palladium- or copper-catalyzed chemistry. The R1 = CF3 derivative (20) was roughly equipotent to the methyl analogue 12 but less active than chloride 2 (Table 5). Increasing the size of the R1 group to methoxy, 21, abolished measurable binding, as did replacing R1 with hydrogen (22), highlighting the tight size restriction of groups at R1. Only the bromide analogue 23 had improved potency over the chloride, by approximately 5-fold, producing the first sub-micromolar lead compound in this series (pIC50 6.5, IC50 300 nM). All of these R1 analogues had minimal BRD4 potency. 23, the most potent ATAD2 compound for which the selectivity window was most measurable, achieved >100-fold selectivity over BRD4 BD1 and 60-fold over BRD4 BD2. Binding to BRD2, BRD3, and BRDT bromodomains was also negligible (Table S2, Supporting Information). Returning to Table 3, the activity of compounds such as 3, 8, and 16 suggested that the hydantoin R3 substituent may not be critical. We therefore prepared a small number of heavily truncated fragments to assess the baseline activity of the minimal phenylsulfonamide core and simultaneously to test additional small R1 substituents, as shown in Scheme 1. The minimal chloride fragment 27 was significantly less potent than parent 2 but retained some measurable activity (pIC50 ∼ 4.1, Table 6). Fragments could be tested at concentrations up to 1 mM, enabling us to conclude that most chlorine replacements (28− 34) resulted in the loss of activity, consistent with the results from the larger scaffold in Table 5. Compound 27 gave partial responses at the concentrations tested (46−55% response at 100 μM), whereas the others showed little response in the 100 μM to 1 mM range on all test occasions. As before, only the bromide (35) was a better R1 group than the chloride (27) by approximately 5-fold. These results also showed that although our initial hits contained active compounds with various R3 substituents, reducing the size of R3 to ethyl resulted in a significant drop in potency (compare 35 to 23, ∼50-fold, and 27 to 2, ∼60-fold), indicating that the hydantoin R3 group does contribute to ATAD2 binding. 23 was a validated lead molecule with activity against ATAD2 and selectivity over the BET bromodomains. The SAR indicated that bromide was the optimal R1 group and that R3 groups including the hydantoin contributed significantly to ATAD2 activity. In TR-FRET binding assays against other bromodomains, 23 had a promising selectivity profile with submicromolar activity against ATAD2 and ATAD2B, at least 10fold selectivity activity against CECR2, TAF1 and BRPF3, and little binding to bromodomains such as BPTF, BRD7, BRD9, and CREBBP (Figure 8a). It also showed activity against endogenous ATAD2 in the bromosphere assay (Figure 8b). Further optimization of this series against the bromodomain of ATAD2 and also against CECR2, including screening against a broader bromodomain panel, is described in a second publication.36 The remainder of this manuscript focuses on understanding the interactions of the core phenylsulfonamide moiety with ATAD2. X-ray Structural and Computational Studies. The common acetyl-lysine binding motifs of bromodomains are well understood and have been extensively reviewed.15 Acetyllysine and most known bromodomain inhibitors form a pair of hydrogen bonds to the W1 water of a water network lying at the base of the pocket and to a conserved asparagine side chain (Asn1064 in ATAD2). However, the conserved parts of the phenylsulfonamide series resembled no previously reported

Figure 7. Biophysical binding data for compound 2. (A) SPR sensorgram and concentration−response curves for mutant (red) and wild-type (black) ATAD2 bromodomains. (B) 15N−1H HSQC NMR spectrum (black in the presence of 100 μM 2 and red without).

symmetrically disubstituted acyclic amines, with piperidine the most heavily represented. In contrast, a variety of larger chemically distinct substituents were tolerated at the R3 position. As the most potent example, we chose 2 for further investigation. Chiral separation generated two undefined stereoisomers, 18 and 19, whose ATAD2 potency was indistinguishable from racemate 2 (Table 5). For the initial investigation of chemistry, we therefore opted to leave the racemic 4-methyl 4-cyclopropyl hydantoin group at R3, as well as the piperidine at R2 unchanged, and varied the R1 position by introducing small modifications. The chemistry to access the noncommercial compounds discussed in this article is described in Scheme 1. Synthesis of the inhibitors started from sulfonyl chlorides 24, which in most cases were commercially available (X = Cl, Br, Me, F) or easily accessible by chlorosulfonylation of the para-substituted nitrobenzine (X = ethyl, Cl).34 The formation of the sulfonamide followed by reduction of the nitro group with Fe(0) gave aniline 25, which could then be coupled with commercially available acid 26 to give compound 23 or with propionyl chloride to give compounds 29, 34, and 35. It was then possible to obtain the trifluoromethyl derivatives 20 and 28 from bromide 23 or 35, respectively, using methyl fluorosulfonyl difluoroacetate.35 Phenol 30, aniline 31, or nitrile 32 could all be obtained from G

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Table 3. ATAD2 and BRD4 Potencies for Phenylsulfonamides Directly from the HTS

a

Some results below the curve-fitting threshold could not be included in mean and SD. In these cases, the numbers included in the mean/number of times tested are both shown.

piperidine. Refinement statistics and electron density maps are provided in Table S3 and Figure S9, Supporting Information. Consistent with their activity in a peptide-competitive binding assay, 9 and the other compounds from this series bind to the acetyl-lysine pocket of ATAD2. However, they lie deeper than acetyl-lysine and most other bromodomain inhibitors. The four conserved water molecules W1−W4 are all displaced by the

acetyl-lysine mimetic, suggesting the possibility of a novel mode of interaction. We therefore tried to obtain X-ray structures of representatives of this series bound to ATAD2 by soaking compounds into protein crystals. Three structures (chloride 9, methyl-containing 15, and bromide analogue 36) are shown in Figure 9. 36 was prepared by the route shown in Scheme 1 using 3-methyl H

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Table 4. ATAD2 SPR, 15N−1H HSQC NMR, and Bromosphere Confirmation of HTS Hits compound

SPR saturating binding

SPR pKd

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Y Y yes but poor curve fit yes but poor curve fit Y Y Y Y Y Y Y Y yes but poor curve fit Y Y Y

5.7 5.2

4.3 4.8 4.8 5.1 4.8 4.9 4.7 4.3 5.1 4.5 4.3

NMR binding binding binding binding binding binding binding binding binding binding binding binding binding binding binding binding

and in this respect, it acts as a mimetic of the W1 water molecule. The hydrogen-bonding potential of the second sulfonamide oxygen is unsatisfied, with carbon atoms of the side chains of Phe1009, Ala1060, and Ile1074 its closest contacts. The sulfonamide dimethylamine group also makes further hydrophobic contacts with the walls of the site, including Val1008 and Ile1074. The remaining polar interaction visible in this crystal structure is a hydrogen bond between the amide NH of 9 and the backbone carbonyl of Lys1011 (Figure 9b). Density for the whole ligand was clear up to this amide but was sparse for the hydantoin group, which was modeled into the space available to avoid protein contacts. In this position, one of the hydantoin carbonyl oxygens would be well placed to form a hydrogen bond to the Asp1014 backbone NH. The binding mode of 15, shown in Figure 9c, is very similar to that of 9. The azepane ring fills more of the site, making more extensive contacts with the ZA loop residues Tyr1021 and Tyr1063. The methyl group of 15 at the bottom of the pocket superimposes precisely over the chlorine of 9. Again, the hydantoin was not well resolved and could only be modeled. The bromide analogue 36 also binds in a very similar way (Figure 9d) but this structure revealed more density for the hydantoin, which permitted its modeling with confidence. One of the two hydantoin carbonyl groups does indeed form a hydrogen bond to the backbone NH of Asp1014, as modeled for 9 and 15, although due to the symmetrical shape of this ring, it is impossible to say if there is any preference for which carbonyl makes this interaction. In the complex with 36, the side chain of Arg1007 adopts a conformation able to form a hydrogen bond to the hydantoin second carbonyl. However, Arg1007 has previously been seen to be highly mobile in complexes with other inhibitors. The density for Arg1007 is similar in the complexes with 36 and 15, but not with 9, suggesting that this interaction is weak and dependent on crystal forces. Alternatively, its formation may depend on the substituents on the hydantoin, but the hydantoin substituents seem to make little difference to ATAD2 potency (Table 3).

bromosphere pIC50 4.8 ± 0.02 (2) 4.8 ± 0.05 (2)

4.4 ± 0.09 (2)

phenylsulfonamide group, which lies at the deepest point of the pocket (Figure 9a). The phenyl ring of 9 makes predominantly hydrophobic contacts with Val1008, Phe1009, and Val1013 (Figure 9b). The chlorine atom fills a small indentation normally occupied by the W2 water, whose hydrogen-bonding potential is usually satisfied by three other waters, including W1 and W3, and by the backbone carbonyl of Ile1056. In the absence of W1− W4, the chlorine of 9 binds in a mixed environment, making polar contacts with the backbone carbonyl of Ile1056 and the hydroxyl group of Tyr1021, as well as nonpolar contacts with the hydrophobic side chains of Phe1009, Ala1060, and Ile1056. The side-chain NH2 of Asn1064 normally donates a hydrogen bond to inhibitor acceptor groups. In the complex with 9, this group is straddled by the two sulfonamide oxygen atoms (Figure 9b) but the heavy atom distance is over 3.5 Å, which is longer than most hydrogen bonds. One of the sulfonamide oxygens forms a hydrogen bond to the side-chain hydroxyl of Tyr1021, Scheme 1. Synthesis of Phenysulfonamide Inhibitorsa

a Reagents and conditions: (a) piperidine, NEt3 or N,N-diisopropylethylamine (DIPEA), CH2Cl2, 0 °C to room temperature, 37−97%; (b) Fe(0), NH4Cl, EtOH/H2O, 70 °C, 67−74%; (c) 26, 1-[bis(dimethylamino)methylene]-1H-1,2,3-triazolo[4,5-b]pyridinium 3-oxid hexafluorophosphate (HATU), DIPEA, CH2Cl2, room temperature, 63%; (d) C2H5COCl, NEt3, CH2Cl2, 0 °C to room temperature, 39−76%; (e) CH3CF2COOSO2F, CuI, dimethylformamide (DMF), 80 °C, 10−15%; (f) KOH, Pd2(dba)3, tBuXPhos, dioxane/H2O, 100 °C, 21%; (g) L-proline, CuI, K2CO3, 25% w/w aqueous ammonia, 110 °C, 22%; (h) CuCN, N-methyl-2-pyrrolidone (NMP), 160 °C, 62%.

I

DOI: 10.1021/acs.jmedchem.9b00673 J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Table 5. ATAD2 and BRD4 Potencies for Compounds 18−23

compound

R1

stereochemistry

ATAD2 pIC50

BRD4 BD1 pIC50

BRD4 BD2 pIC50

2 18 19 12 20 21 22 23

Cl Cl Cl Me CF3 OMe H Br

racemic enantiomer 1 enantiomer 2 racemic racemic racemic racemic racemic

5.9 ± 0.08 (12) 5.9 ± 0.04 (4) 5.9 ± 0.09 (4) 4.9 ± 0.22 (5/6a) 5.0 ± 0.08 (8)