REVIEWS Targeting bromodomains: epigenetic readers of lysine acetylation Panagis Filippakopoulos1,2 and Stefan Knapp1,3

Abstract | Lysine acetylation is a key mechanism that regulates chromatin structure; aberrant acetylation levels have been linked to the development of several diseases. Acetyl-lysine modifications create docking sites for bromodomains, which are small interaction modules found on diverse proteins, some of which have a key role in the acetylation-dependent assembly of transcriptional regulator complexes. These complexes can then initiate transcriptional programmes that result in phenotypic changes. The recent discovery of potent and highly specific inhibitors for the BET (bromodomain and extra-terminal) family of bromodomains has stimulated intensive research activity in diverse therapeutic areas, particularly in oncology, where BET proteins regulate the expression of key oncogenes and anti-apoptotic proteins. In addition, targeting BET bromodomains could hold potential for the treatment of inflammation and viral infection. Here, we highlight recent progress in the development of bromodomain inhibitors, and their potential applications in drug discovery.

Structural Genomics Consortium, Nuffield Department of Clinical Medicine, University of Oxford, Old Road Campus Research Building, Roosevelt Drive, Oxford OX3 7DQ, UK. 2 Ludwig Institute for Cancer Research, Nuffield Department of Clinical Medicine, University of Oxford, Roosevelt Drive, Oxford, OX3 7DQ, UK. 3 Target Discovery Institute, Nuffield Department of Clinical Medicine, University of Oxford, Roosevelt Drive, Oxford OX3 7BN, UK. e-mails: panagis.filippakopoulos@ sgc.ox.ac.uk; [email protected] doi:10.1038/nrd4286 Published online 22 April 2014 1

Epigenetics has been defined as heritable changes in the pattern of post-translational modifications (also referred to as the epigenetic code), which lead to the development of new phenotypes that are not encoded in the DNA sequence1. It is now widely accepted that changes in this complex epigenetic code are one of the major mechanisms that lead to the development of disease (FIG. 1). The post-translational modification of histone acetylation is a hallmark of chromatin that has an open structure that can be accessed by DNA and RNA polymerases as well as transcription factors, resulting in the activation of gene transcription. Acetylation levels on histones are highly regulated by histone acetyltransferases (HATs; enzymes that produce or write acetylation marks) and histone deacetylases (HDACs; enzymes that erase acetylation marks). These enzymes are often deregulated in diseases through mechanisms that include aberrant expression levels, the occurrence of mutants as well as truncations and chromosomal rearrangements. In cancer, deregulation of HDACs results in upregulation of the transcription of growth-promoting genes, downregulation of tumour suppressor genes as well as the deregulation of microRNA activity 2–7. These discoveries led to the development of clinically successful HDAC inhibitors in oncology 5,8–10. Proteins that read epigenetic marks can also be targeted, as evidenced by the development of bromo­ domain inhibitors that specifically target the BET (bromodomain and extra-terminal) proteins. Gene expression

studies on the transcriptional effects of these inhibitors demonstrated that the inhibition of BET bromodomains selectively interfered with gene expression programmes that mediated cellular growth and evasion of apoptosis in cancer 11–13, as well as the inflammatory response in immune cells14. Given the general role of BET proteins in transcriptional elongation15,16, the discovery that inhibition of these proteins only affects the transcription of a small subset of genes was unexpected and suggested that inhibitors of bromodomains may specifically modulate the expression of some disease-promoting genes. Here, we summarize recent progress in the emerging area of bromodomain drug discovery by reviewing current efforts aiming to target the bromodomain class of acetyl-lysine readers using small molecules. We also highlight early studies that used a variety of chemical templates showing potency and selectivity for certain bromodomains in diverse diseases such as cancer, inflammation and viral infections. Finally, we look at early-stage clinical trials of bromodomain inhibitors and the challenges that could be encountered as these molecules progress through the clinic.

Lysine acetylation: a targeted modification ε‑N‑acetylation of lysine residues on the amino-terminal tails of histones was discovered over 30 years ago and has been generally associated with open chromatin architecture as well as transcriptional activation17. More

NATURE REVIEWS | DRUG DISCOVERY

VOLUME 13 | MAY 2014 | 337 © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS

Reader Writer DNA

H2A H3

H2B H4

Eraser

Ac HATs

Ac Bromodomains, chromodomains

Ac HDACs

Me HMTs

Me PHDs, tudor, PWWPs, MBTs

Me DMTs

P Kinases

P 14-3-3, BRCTs

P Phosphatases

Figure 1 | Histone modifications and their readout in the context of chromatin.  Histone proteins (H2A, H2B, H3 and H4) are found at the core of nucleosomes and are often post-translationally modified by ‘writers’ and ‘erasers’, which are enzymes that deposit and remove a variety of chemical modifications that areNature consequently by Reviewsinterpreted | Drug Discovery effector ‘reader’ modules, which translate them in the context of chromatin reorganization and transcriptional control. Writer enzymes include histone acetyltransferaces (HATs), histone methyltransferases (HMTs) and kinases. Eraser enzymes include histone deacetylases (HDACs), demethylases (DMTs) and phosphatases. Reader proteins include modules such as bromodomains, chromodomains, plant homeodomains (PHDs), tudor domains, PWWP (Pro-Trp-Trp-Pro) domains, malignant brain tumour domains (MBTs), 14-3-3 proteins and BRCT domains (named after the carboxy-terminal domain of a breast cancer susceptibility protein). Chemical modifications include acetylation and methylation of lysine, as well as phosphorylation of serine, threonine and tyrosine, among others. For instance, HATs deposit acetylation marks on lysine residues, which are ‘read’ by bromodomain modules and removed by HDACs.

recently, some histone acetylation marks have been associated with the compaction of chromatin18, DNA repair 19, protein stability as well as the regulation of protein–protein interactions20. Lysine acetylation has been studied as a central component of the so-called histone code — the ensemble of post-translational modifications present in histone proteins21 — and has emerged as a widespread post-translational modification that is found across the entire proteome22,23. Indeed, public repositories such as the PhosphoSitePlus database show that there are over 24,000 lysine acetylation sites in human cells, which suggests that this post-translational modification may have a broad role in signalling networks. The frequent occurrence of this modification in nuclear and non-nuclear cellular compartments suggests that acetylation has an important role in signal transduction24. However, no acetylationdependent signalling pathways have been described to date. Dysfunctional levels of acetylation have been linked to diverse diseases, particularly to the development of cancer, where deregulation of the acetylation patterns of histones often promotes the expression of oncogenes, resulting in cellular proliferation and tumorigenesis. Although HDAC inhibitors are successfully used for the treatment of cancer and are under investigation for other conditions such as central nervous system disorders8,9,25,26, approved HDAC inhibitors are largely non-selective. The pleiotropic effects of the currently approved inhibitors have made it difficult to examine and understand their molecular mechanism of action, thus hampering their broader clinical application. The development of specific chemical tool compounds — also called chemical probes — that target epigenetic posttranslational modifications has emerged as an excellent

approach for validating new treatment strategies for diseases that have complex underlying mechanisms. Several highly selective chemical probes have been developed for bromodomains, and results from studies using these inhibitors have suggested that inhibition of bromodomains could have several potential clinical applications, as outlined below.

Bromodomains: readers of acetyl lysine residues Readers of post-translational modifications are structurally diverse proteins than contain one or more effector modules that recognize (that is, read) covalent modifications of proteins and DNA. The recognition of ε‑N‑acetylation of lysine residues is primarily initiated by bromodomains, a family of evolutionarily conserved protein interaction modules that were identified in the early 1990s in the brahma gene from Drosophila melanogaster 27. The human proteome encodes 61 bromodomains, which are present in 46 diverse nuclear and cytoplasmic proteins. These include HATs and HATassociated proteins (such as general control of amino acid synthesis protein 5‑like 2 (GCN5L2; also known as KAT2A), P300/CBP-associated factor (PCAF; also known as KAT2B) and bromodomain-containing protein 9 (BRD9))28,29, histone methyltransferases (such as ASH1L and mixed lineage leukaemia protein (MLL))30,31, helicases (such as SWI/SNF-related matrix-associated actin-dependent regulators of chromatin subfamily A (SMARCAs))32, ATP-dependent chromatin remodelling complexes (such as bromodomain adjacent to zinc finger domain protein 1B (BAZ1B; also known as Williams syndrome transcription factor))33, transcriptional co‑activators (such as tripartite motif-containing proteins (TRIMs) and TBP-associated factors (TAFs))34

338 | MAY 2014 | VOLUME 13

www.nature.com/reviews/drugdisc © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS and transcriptional mediators (such as TAF1)35, nuclear scaffolding proteins (such as polybromo 1 (PBRM1))36 as well as the BET family of proteins16,37. The broader biological functions of bromodomains have been reviewed elsewhere in the literature38–40. Proteins that contain bromodomains are involved in the regulation of transcriptional programmes and have been identified in oncogenic rearrangements that lead to highly oncogenic fusion proteins, which have a key role in the development of several aggressive types of cancer. In addition, bromodomain-containing proteins regulate nuclear factor‑κB (NF‑κB), which is a key transcription factor that mediates inflammatory responses. They are also implicated in the replication of viral genomes and regulate the transcription of some viral proteins. Taken together, this suggests that targeting these proteins could be beneficial in the development of new treatment strategies for cancer, inflammation and viral infections.

Nuclear factor‑κB (NF-κB). A protein complex that controls gene transcription. It has a key role regulation of the immune response that is initiated owing to infection. Its deregulation has been linked to the pathogenesis of cancer, inflammation and viral infection.

Chemotypes Chemically distinct entities with differences in the composition of their secondary metabolites.

Chemical scaffolds Molecular backbone of a molecule on which functional groups are altered during drug design.

Structure of bromodomains Despite having large sequence variations, bromodomain modules share a conserved fold that comprises a lefthanded bundle of four α-helices (named αZ, αA, αB and αC) that are linked by diverse loop regions of variable charge and length (known as ZA and BC loops) which surround a central acetylated lysine binding site (FIG. 2a). A large amount of structural information has enabled the comprehensive structural analysis of bromodomains, establishing the sequence and structural conservation of these modules. Moreover, structure-based alignments have clustered human bromodomains into eight distinct families41 (FIG. 2b). Structural data have established that acetylated lysine is recognized in a central hydrophobic pocket, where it is anchored to a conserved asparagine residue29,42. More recently, it has been demonstrated that some bromodomains bind to two acetylated lysine histone marks that are simultaneously recognized by the same bromodomain module43. This property is shared by all members of the BET subclass of bromodomains41. High-resolution co‑crystal structures showed that the first acetylated lysine mark of histone H4 docks directly onto the conserved asparagine (Asn140 in the first bromodomain of BRD4). Simultaneously, a network of hydrogen bonds, formed via conserved water molecules found in the bromodomain cavity, link to the second acetylated lysine mark, thus stabilizing the peptide complex (FIG. 2c). The largely hydrophobic nature of the central acetylated lysine binding pocket of the bromodomain (FIG. 2d), which is necessary to accommodate the charge-neutralized acetylated lysine and the comparably weak interaction with its target sequences, makes these modules particularly attractive for the development of inhibitors targeting this protein–protein interaction. Discovery of bromodomain inhibitors Initial fragment-like bromodomain inhibitors were described in the literature in 2005 (REF. 44), but potent inhibitors were not developed until 2009, when tri­ azolothienodiazepines were identified as BET bromodomain inhibitors using an anti-inflammatory phenotypic

assay; triazolothienodiazepines had been described in a 1998 patent as potential therapeutics for the treatment of inflammatory intestinal diseases, including ulcerative colitis and Crohn’s disease45, and in a later 2006 patent they were shown to inhibit CD28 co‑stimulatory signals in T cells46. These inhibitors were further explored in subsequent patents as antitumour agents that targeted BET bromodomains47. Following the disclosure of these chemotypes in the patent literature, several efforts have been initiated in industry and academia to develop and characterize potent and selective molecules that can target bromodomains. To date, a large number of chemical scaffolds have been published, which aim to modulate the epigenetic function of the acetyl-lysine reading process (as discussed below). To present these efforts here, we have categorized known bromodomain inhibitors into two main classes, based on the presence or absence of moieties that act as acetylated lysine mimetics. The non-acetylated lysine mimetic class contains small molecules that engage the bromodomain module within the acetylated lysine binding pocket but without forming a canonical hydrogen bond with the conserved asparagine, which typically anchors acetylated lysine peptides. This type of inhibitor sterically excludes peptide binding and, as such, directly inhibits the acetylated lysine-reading function of the bromodomain module. Weak inhibitors such as NP1 (REF. 44), MS7972 (REF. 48), ischemin49, MS436 (REF. 50) and BID1 (REF. 51) belong to this class. Another class of bromodomain inhibitors includes small molecules that directly engage the protein module by forming hydrogen bonds with the conserved asparagine residue in a way that mimics the binding mode of acetylated lysine; this usually results in the binding of the inhibitor deeper within the acetylated lysine binding site but without displacing the conserved water molecules that are present at the bottom of the acetyl-lysine binding cavity. These inhibitors also competitively inhibit the binding of acetylated lysine -containing peptides to bromodomains. Inhibitors that belong to this class include: the thienodiazepines (+)-JQ1 (REF. 52) and the related MS417 (REF. 53) and CPI‑203 (REF. 54); the benzodiazepines GW841819X, I-BET762 (also known as GSK525762A) and GSK525768A55; the benzotria­ zepine BzT‑7 (REF. 56); the isoxazoles isoxazole‑4d57 and isoxazole‑9 (REF. 58); I-BET151 (REF. 59); the isoxazole azepine compound 3 (REF. 60); compound 15 from REF. 61; compounds 36 and 38 from REF. 62; the isoxazole benzoimidazole SGC‑CBP30 (see the SGC‑CBP30 overview on the SGC website for further information); the dihydroquinazoline-one PFI‑1 (REF. 63); the quinazolone RVX‑208 (REF. 64); and triazolophthalazines65. Further members of this class include the kinase inhibitors BI‑2536 and TG‑101348 (REF. 66), the poisonous alkaloid colchiceine (XD1) and its methoxylated analogue colchicine (XD25)67, the 4‑acyl pyrrole compound XD14 (REF. 67), 2‑thiazolidinones68, triazolopyrimidines69, methylquinolines69, chloropyridinones69 indolizine-ethanones70 and N‑phenylacetamides70, as well as methyltriazoles71.

NATURE REVIEWS | DRUG DISCOVERY

VOLUME 13 | MAY 2014 | 339 © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS b

N ZA loop

aB

III

aC

aA aZ

–10 kT/e

+10 kT/e

Z1 A CR EP3 BRD B EB 00 8B RD BP (1) 8B (2) BRWD1(2) PHIP(2) BRWD3(2) BAZ1B BRD9 IV BRD7 BRPF3 BRPF1B D1 BRPF1A BR AD2B ATAD2 AT TRIM66 TRIM33B TRIM33A α TIF1 0C 1 SP1 110A 00 SP SP1 SP140 V

I

GCN5L2 PCAF

PBRM1(1) ASH1L

VIII

SMARCA2A SMARCA2B PBRM1(6) SMARCA4 BRWD3(1) PHIP(1) BRWD1(1)

VI

SP140L

PRKCBP1 TAF1(1) VII TAF1L(1) TAF1(2)

N140

Y139

D145

d

I146

H4K5ac

Y97

BA

BRD4(2) BRDT(2)

L(2)

C136 D144 L94

BRD2(2) BRD3(2)

1 TAF 1 ND1 ZMY MLL 28 TRIMAZ2A B 2B BAZ

c

BRD3(1) BRD2(1)

PB R PB M1( RM 3) P PB BR 1( RM M1 2) 1(5 (4) )

ZA loop

II

CR FAL2 Z

BC loop

BR D4 (1) BR DT (1)

BC loop

CE

a

Y139

N140

BC loop

L94

L92

H4K8ac

Y97

D144 C136 I146

L92

D145

M149 V87

αB

P82

ZA loop

αC W81

Q85

ZA loop M149

V87 αZ

αA

Q85

P82 W81

BRD4 (1) H4K5acK8ac Conserved asparagine

BRD4 (1) Hydrophobic core Conserved asparagine

Figure 2 | Bromodomain family: structure and acetyl-lysine peptide recognition.  a | The bromodomain fold comprises a left-handed bundle of four α-helices (αZ, αA, αB and αC) that are linked by diverse loop regions of variable Nature Reviews | Drug Discovery charge and length (ZA and BC loops), which line up a central binding site. Major structural elements are highlighted on the structure of the first bromodomain of bromodomain-containing protein 4 (BRD4(1)) in complex with a di‑acetylated histone H4 peptide (Protein Data Bank (PDB) ID: 3UVW). The electrostatic potential of the module is also shown, from –10 to +10 kT/e, highlighting the charged nature of the surface lining the acetyl-lysine recognition site. b | Structure-based phylogeny of the human bromodomain family, which consists of 61 modules that are present in 46 proteins. Roman numerals indicate the eight major structural classes, which result from the structural classification of bromodomains in REF. 41. c | Binding of acetyl-lysine peptide to bromodomains. A di‑acetylated H4 peptide (double acetylation; at H4K5acand H4K8ac) is shown (in blue) bound to BRD4(1). The peptide directly docks onto the conserved asparagine (N140) and engages the protein via a series of hydrogen bonds through conserved water molecules, shown as red spheres (PDB ID: 3UVW). d | The hydrophobic nature of the acetyl-lysine binding site is evident by the aromatic and hydrophobic residues that line the central acetyl-lysine-binding cavity, shown here on BRD4(1) (PDB ID: 3UVW). ASH1L, histone-lysine N-methyltransferase ASH1L; ATAD2, ATPase family AAA domain-containing protein 2; BAZ, bromodomain adjacent to zinc finger domain protein; BRPF3, bromodomain and PHD finger-containing protein 3; BRWD3, bromodomain and WD repeat-containing protein 3; CECR2, cat eye syndrome critical region protein 2; CREBBP, CREB-binding protein; EP300, E1A‑associated protein p300; FALZ, fetal Alzheimer’s antigen; GCN5L2, general control of amino acid synthesis protein 5‑like 2; MLL, mixed lineage leukaemia protein; PBRM1, polybromo 1; PCAF, P300/CBP-associated factor; PHIP, pleckstrin homology domain interacting protein; PRKCBP1, protein kinase C-binding protein 1; SMARCA, SWI/SNF-related matrix-associated actin-dependent regulator of chromatin subfamily A; SP100, 100 kDa speckled protein; SP140L, SP140-like protein; TAF, TBP-associated factor; TIF1α, transcription intermediary factor 1α; TRIM, tripartite motif-containing protein; ZMYND11, zinc finger MYND domain-containing protein 11.

340 | MAY 2014 | VOLUME 13

www.nature.com/reviews/drugdisc © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS Although early studies focused only on the identification of compounds that targeted BET bromodomains, a recent study explored the druggability of the entire family of bromodomains (FIG. 3a). This identified unique amino-acid bromodomain signatures that helped determine that all subfamilies had relatively good druggability scores, which indicates that potent inhibitors can be developed72. It is noteworthy that preliminary data suggest that BAZ2B — a bromodomain that was predicted by the analysis72 to be among the most difficult bromodomains to target — has now been successfully inhibited by acetylated lysine mimetic ligands (see the GSK2801 overview on the SGC website for further information).

Druggability scores Numerical quantities that are calculated from experimental structural data and that assess the druggability of structures by taking into account the contributions from the volume, the level of enclosure and the degree of hydrophobicity of a pocket towards the binding of small molecules.

Histac A technology that relies on conformational changes upon acetylation. This change alters the fluorescence resonance energy transfer (FRET) that is generated when a histone peptide is fused to the bromodomain-containing protein BRDT and further fused in tandem with the donor and acceptor protein.

Lipinski’s rule-of-five guidelines A rule of thumb formulated by Christopher Lipinski in 1997, based on the observation that most medications are relatively small and lipophilic molecules. These guidelines are used to evaluate drug-likeness or determine whether a chemical compound with a certain pharmacological or biological activity has properties that would make it likely to be an orally active drug in humans.

Non-acetyl-lysine mimetic templates. Efforts to target bromodomain modules, focusing on PCAF and CREBbinding protein (CREBBP), were initiated almost 10 years ago and led to the identification of the weak PCAF inhibitor NP1 (also referred to as compound 2 in some publications). NP1 was selective over CREBBP and transcription intermediary factor 1β (TIF1β), and it disrupted the interaction of PCAF with an HIV‑1 Tat peptide (NP1 had a halfmaximal inhibitory concentration (IC50) value of 1.6 μM for PCAF). This interaction was characterized using twodimensional 15N heteronuclear single-quantum coherence NMR (15N-HSQC NMR), a method that allows the detection of correlations between nuclei of two different types. The interaction was further probed in vitro using western blots and a biotinylated version of the HIV‑1 Tat peptide immobilized on streptavidin44. Later, CREBBP inhibitors with low micromolar affinity were identified using NMR screening methods, which exploited the interaction of the CREBBP bromodomain with tumour suppressor p53 acetylated at Lys382, yielding molecules such as MS7972 (dissociation constant (Kd) value = 19.6 μM)48. Although this compound had weak affinity for the bromodomain of CREBBP, it effectively suppressed the recruitment of CREBBP to p53, thus modulating the expression of p53 target genes, such as the cell cycle inhibitor p21. More recently, novel CREBBP binders were identified by measuring perturbations of resonance chemical shifts observed in two-dimensional 1H-HSQC NMR and 15 N-HSQC NMR spectra induced by the addition of small molecules from a chemical library. This effort resulted in hits that shared an azobenzene scaffold, which was further optimized to yield molecules such as ischemin, which had a Kd value of 19 μM against the bromodomain of CREBBP49. Ischemin was shown to bind to the bromodomain in a non-acetyl-lysine-competitive way by occupying the top of the acetylated lysine binding pocket of CREBBP, packing between Leu1120 from the ZA loop and Arg1172 from the BC loop (FIG. 3b,c). Interestingly, despite its modest in vitro affinity, ischemin efficiently protected rat cardiomyocytes against myocardial damage by reducing p53‑induced apoptosis instigated by doxorubicin treatment. However, the specificity of ischemin for the CREBBP bromodomain has not yet been evaluated. The diazobenzene scaffold of ischemin was recently used as a template to develop further potent inhibitors for BET bromodomains, which resulted in MS436, a

compound that had an estimated Ki (inhibition constant) value of 30–50 nM for the first bromodomain of BRD4 and a small window of selectivity (ten‑fold) over the second bromodomain of BRD4. In murine macrophages, MS436 blocked the transcriptional activity of BRD4 in the NF‑κB‑directed production of nitric oxide and the pro-inflammatory cytokine interleukin‑6 (IL‑6)50. Ischemin and several other azobenzene scaffolds were also disclosed in a patent 73 that described targeting bromodomains to treat diseases associated with NF‑κB and p53 activity; despite the submicromolar affinities that these compounds exhibit against BRD4, they still crossreact with the bromodomains of PCAF and CREBBP. Benzoimidazole template (non-acetyl-lysine mimetics targeting BET bromodomains). An interesting and potentially useful fluorescent probe technology, termed histac 74, which enables the visualization of histone H4K5/H4K8 di-acetylation in cells, has been used to identify potential bromodomain inhibitors. Initial docking of a compound library into the acetyllysine cavity of the first bromodomain of human BRD2 (Protein Data Bank (PDB) ID: 1X0J) resulted in 192 virtual screening hit compounds (out of 628,402 tested) that complied with Lipinski’s rule-of-five guidelines75. These compounds were further screened by surface plasmon resonance (SPR) to verify binding to the first bromodomain of BRD2, and four compounds were chosen for further assessment in cells using histac technology. One of the four compounds, BIC1 (Kd value = 28 μM; determined by SPR), suppressed binding to the H4K5 and H4K8 double acetylation mark in a concentrationdependent manner. The crystal structure of the first bromodomain of BRD2 in complex with BIC1 unambiguously verified the binding of this scaffold to the bromodomain of BRD2, demonstrating that the compound engages the protein in a unique mode without forming any hydrogen bonds to the conserved asparagine (Asn156 in the first bromodomain of BRD2); instead, it is bound mainly through hydrophobic interactions (PDB ID: 3AQA)51. Triazolothienodiazepine and benzodiazepine template (acetyl-lysine mimetic targeting BET bromodomains). The first high-affinity and selective inhibitors of bromo­ domains came from the triazolothienodiazepine and triazolobenzodiazepine classes of chemical scaffolds, as previously mentioned. Triazolothienodiazepines were studied in the late 1990s; they inhibited cell adhesion and, as such, were tested in models of inflammatory diseases of the gut (as described in a patent)45. The same class of compounds was later shown to inhibit co‑stimulatory signals from CD28 on T cells, a property that has been linked to rejection during transplantation and the development of autoimmune diseases, as well as allergic reactions46. The structurally related class of tri­ azolobenzodiazepines comprises drug-like small molecules that have been extensively developed as specific modulators of the GABA (γ‑aminobutyric acid) receptor and have received regulatory approval for the treatment of anxiety, muscle spasms, seizures and sleeping disorders76.

NATURE REVIEWS | DRUG DISCOVERY

VOLUME 13 | MAY 2014 | 341 © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS

Druggability score (SiteMap score)

a

1.0 0.8 0.6 0.4

V

VII

Druggable

b

VIII

TRIM 24 BAZ2 B TAF1 (1) TAF1 (2) TAF1 L(2) PBRM 1(1) PBRM 1(2) PBRM 1(5) SMA RCA 4

BRD1

PCAF GCN 5L2 FALZ

IV

BP PHIP (2) BRD9

III

CREB

II CECR 2 BRD2 (1) BRD3 (1) BRD4 (1) BRD2 (2) BRD3 (2) BRD4 (2) EP30 0

I

Intermediate

Hard

c N1168

I1122

BC loop

Y1167

N1168

Y1167

S1171

S1171

I1122

αB

V1173

αC

Y1125

R1172 Y1125

F1177

L1120 ZA loop V1115

K9ac

P1110

V1173 R1172

V1115 L1120

F1177 Q1113

αA

P1110

L1109

αZ

L1109

Q1113 Ischemin CREBBP H3 peptide Conserved asparagine

d

e N140

D144

N140 Y139

αC

αB

D145

Y97 L94

I146

Y97

M149

L94

K5ac

K8ac P82

ZA loop

D145 L92 V87

αZ Q85

I146

W81

L92 V87 αA

D144

Y139

M149 Q85

(+)-JQ1 BRD4 (1) H4 peptide Conserved asparagine

P82 W81

Nature Reviews | Drug Discovery www.nature.com/reviews/drugdisc

342 | MAY 2014 | VOLUME 13 © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS Figure 3 | Predicted druggability and binding of small molecules to bromodomains.  The predicted druggability of human bromodomain modules (part a) was calculated in REF. 72 using publicly available structural information. Most modules showed good scores, and the more difficult ones — such as bromodomain adjacent to zinc finger domain protein 2B (BAZ2B) — have already been successfully targeted. Roman numerals correspond to the structural families shown in FIG. 2b. Numbers in brackets refer to the specific bromodomain, as in FIG. 2b. Parts b and c show the overlay of an acetylated histone H3 peptide complex (P300/CBP-associated factor (PCAF)–H3K9ac; Protein Data Bank (PDB) ID: 2RNW) with the complex of ischemin bound to the CREB-binding protein (CREBBP) bromodomain (PDB ID: 2L84). Ischemin engages the protein surface in a nonacetyl-lysine-competitive fashion without inserting deep into the acetyl-lysine binding pocket, yet it occupies the same shelf as the peptide backbone, packing between the BC loop residue R1172 and the ZA loop residue L1120. Parts d and e show the overlay of a di-acetylated histone H4 peptide complex (the first bromodomain of bromodomaincontaining protein 4 (BRD4(1) in complex with H4K5ac/K8ac; PDB ID: 3UVW) with the complex of (+)-JQ1 bound to BRD4(1) (PDB ID: 3MXF). The inhibitor directly engages the conserved asparagine (N140) via its triazolo moiety, further stabilizing the interaction by inserting its chlorophenyl substituent into the shelf between D145 of the BC loop (which occupies the same vector as the second lysine, K8ac) and between the W81/P82 motif and its dimethyl-substituted thieno ring between the W81/P82 motif and L92 of the ZA loop. Peptides are shown as cartoons (H3 in green, and H4 in blue) with acetylated lysine residues shown as sticks. Inhibitors are shown in ball and stick representation. CECR2, cat eye syndrome critical region protein 2; EP300, E1A‑associated protein p300; FALZ, fetal Alzheimer’s antigen; GCN5L2, general control of amino acid synthesis protein 5‑like 2; PBRM1, polybromo 1; PHIP, pleckstrin homology domain interacting protein; SMARCA4, SWI/SNF-related matrix-associated actin-dependent regulator of chromatin subfamily A, member 4; TAF1, TBP-associated factor; TAF1L, TAF1‑like protein; TRIM24, tripartite motif-containing protein 24. Figure is modified, with permission, from REF. 72 © (2012) American Chemical Society.

Commercially available drugs such as alprazolam (a potent short-acting compound that is used for the treatment of anxiety disorders)77, midazolam (a drug that is active against acute seizures, insomnia or as a sedative)78 and L-655,708 (the first subtype-selective inverse agonist that is designed to preferentially bind to the α5 subtype of the GABAA receptor)79,80 are the products of long and systematic medicinal chemistry efforts that have established a rich body of literature around the properties of this class of compounds81–83 as well as the related benzotriazepine scaffolds84–87. Two of these compounds (alprazolam and midazolam) were recently shown to be weak BET bromodomain inhibitors56. Triazolobenzodiazepines were first reported as potent inhibitors of bromodomains in a patent published in 2009; they targeted the BET protein BRD4 and disrupted its interaction with acetylated histone peptides, as well as suppressing cellular proliferation in several tumour models47. Interestingly, many of the disclosed analogues had double- to single-digit nanomolar GI50 values (the concentration required to achieve 50% growth inhibition) in the mixed lineage leukaemia cell line MV4‑11 carrying an MLL–AF4 rearrangement 88. The later-synthesized inhibitor (+)-JQ1 (see the JQ1 overview on the SGC website for further information), which has the same triazolothienodiazepine core scaffold, was shown to be highly potent and very selective against all BET bromodomains when it was studied against a panel of 46 human bromodomains52. The affinity for each BET module ranged from 49 nM to 190 nM. By contrast, the enantiomer (–)-JQ1 did not bind to BET bromodomains52,89 and was completely inactive against

any human bromodomain module. (–)-JQ1 therefore provided a negative control to validate the effects of the triazolobenzodiazepine (+)-JQ1 on the biological role of BET proteins. High-resolution crystal structures also showed that (+)-JQ1 binds to the bromodomains of BET proteins52,89. The triazolo ring of the molecule inserts deep into the acetylated lysine pocket and occupies the same position as the acetyl head group of acetylated lysine present in histone tail peptides (FIG. 3d), and it initiates a direct hydrogen bond to the conserved asparagine (Asn140 in the first bromodomain module of BRD4). The chloro­ phenyl substituent of the diazepine ring occupies the same location as the second acetylated lysine seen in peptide complexes, engaging the shelf between the BC loop and the WP motif (that is, Trp81 and Pro82 in the first bromodomain module of BRD4) that closes the acetyl­ ated lysine pocket. The dimethyl-substituted thieno ring of the molecule packs between the WP motif and Leu92 of the ZA loop, resulting in complete occupancy of the acetylated lysine binding groove (FIG. 3e). It is important to note that the structural information derived from the complexes of BET bromodomains with triazolothienodiazepines and triazolobenzodiazepines has been successfully used to guide computational approaches, resulting in novel classes of inhibitors67–69, as discussed below. Interestingly, the bulky t‑butyl moiety of the molecule rests next to Leu92 and Leu94 of the ZA loop and may be responsible for the low activity of (+)-JQ1 against a range of cellular receptors, which is in agreement with the structure–activity relationship observed for triazolobenzodiazepines that target the GABA receptor52. Furthermore, (+)-JQ1 can competitively displace an acetylated histone H4 peptide in vitro; additionally, (+)-JQ1 displaced fulllength BRD4 from chromatin in cell-based assays using fluorescence recovery after photobleaching (FRAP). MS417 and its inactive stereoisomer MS566 (REF. 53) are two compounds that are structurally similar to (+)-JQ1 and share the same triazolothienodiazepine core scaffold, but contain a methyl-ester instead of a t‑butyl ester. MS417 binds to the bromodomains of BRD4 with Kd values of 36.1 nM (for the first bromodomain) and 25.4 nM (for the second bromodomain) Its mode of binding onto the acetyl-lysine cavity of the first bromodomain of BRD4 has been determined (PDB ID: 4F3I). Importantly, MS417 suppressed tumour necrosis factor (TNF)-induced transcriptional activation of NF‑κB, which demonstrates that it has potential anti-inflammatory properties. The primary amine analogue of (+)-JQ1, CPI‑203, was also shown to inhibit BRD4 activity in vitro and in cells54. Although clinically approved triazolobenzodiazepines are structurally similar to triazolothienodiazepines, they have only weak activities against BET bromodomains; for example, alprazolam binds weakly to the first bromodomain of BRD4, with an in vitro Kd value of 2.5 μM in an acetyl-lysine-competitive mode56. To further develop diverse inhibitors of BET bromodomains, replacement of the stereocentre in triazolobenzodiazepines resulted in the structurally related triazolobenzotriazepines. The most potent inhibitor of this series, BzT‑7, had a Kd value of 640 nM against the first bromodomain of BRD4 (REF. 56).

NATURE REVIEWS | DRUG DISCOVERY

VOLUME 13 | MAY 2014 | 343 © 2014 Macmillan Publishers Limited. All rights reserved

▶

REVIEWS

LC–MS/MS An analytical chemistry technique that combines the physical separation capabilities of liquid chromatography (LC) with the mass analysis capabilities of tandem mass spectrometry (MS/MS), a process that involves multiple steps of mass spectrometry selection, with some form of fragmentation occurring in between the stages of selection.

Liver microsomes A model system used in in vitro ADME (absorption, distribution, metabolism and excretion) studies. Liver microsomes contain a variety of drug-metabolizing enzymes, and so they are used to examine the potential for first-pass metabolism of orally administered drugs.

Ligand efficiency A metric used to aid the selection of lead compounds with optimal combinations of physicochemical properties and pharmacological properties, relying on the measurement of the binding energy per atom of a ligand to its binding partner, such as a receptor or enzyme.

Interestingly, thienodiazepines — including the compound Ro11‑1464 — were shown to regulate levels of the high-density lipid protein apolipoprotein A1 (APOA1) in assays monitoring APOA1 release in liver cells; this was reported in the patent literature in 1997 (REF. 90). However Ro11‑1464 has only been recently linked to BET inhibition owing to its structural similarity with thienodiazepines91. Phenotypic screening (that was independent of the studies described in REF. 92) to monitor APOA1 expression, followed by medicinal chemistry optimization, established compounds containing triazolobenzodiazepine scaffolds as BET-specific inhibitors; GW841819X, I-BET762 (also known as GSK525762A, which is an (S)-enantiomer) and GSK525768A (which is the (R)‑enantiomer of I-BET762) upregulated APOA1 by interacting with a target that was initially unknown92. Further experiments showed that the target was the BET family of bromodomains. Interestingly, only the (S)-enantiomer (GSK525762A) and GW841819X (a racemic mixture) modulated APOA1 levels, whereas the (R)‑enantiomer (GSK525768A) had no effect. Using a chemoproteomics approach that involved immobili­zation of the active (S)- and inactive (R)-enantio­­ mers on a matrix, followed by affinity purification of interacting proteins from cell extracts, the interacting proteins were isolated and identified by liquid chromatography–tandem mass spectrometry (LC–MS/MS) as BRD2, BRD3 and BRD4. The interaction of the inhibitors was mapped to the bromodomain region of BET proteins using constructs lacking or containing the bromodomain modules. It was later reported that the regulation of APOA1 transcription as well as APOA1 protein expression is linked to chemical inhibition of BET bromodomains14,55,93. These inhibitors had similar binding affinity to (+)-JQ1, as measured by several biophysical techniques, and had similar binding modes. Clinical trials of I-BET762 are discussed below. Following these exciting results that linked triazolo­ thienodiazepines and triazolobenzodiazepines to BET bromodomain inhibition, several chemical series based on the azepine ring system, including thienodiazepines94–99, benzodiazepines100–104 and azepine derivatives105, have been disclosed in patents. These patents disclose a rich body of data and methods demonstrating, for example, antiproliferative activity against cancer cell lines94,99 (including leukaemia cell lines)96, strategies for the treatment of metabolic syndromes such as obesity and type 2 diabetes95, male contraception97, cellular assays measuring inhibition of TNF or IL‑6 release from whole blood stimulated with lipopolysaccharide (LPS)100–103,106, as well as cellular assays for assessing the inhibition of MYC RNA in MV4;11 cells98. As discussed below in the sections related to biology, most of these assays have also been extensively evaluated in the research literature. Some in vivo data are also presented in these patents, demonstrating, for example, inhibition of LPS-induced IL‑6 secretion in mice105. Isoxazole-azepine template (acetyl-lysine mimetic targeting BET bromodomains). An interesting combination of isoxazole and azepine ring systems resulted in a chemical template with a high potency for BET bromodomains. Compound 3 from this series bound to the first

bromodomain of BRD4 with the same acetylated lysine binding mode that has been observed for closely related triazolodiazepines such as (+)-JQ1 and I‑BET762, while retaining in vitro selectivity when tested against another 20 diverse bromodomains60. When tested in liver microsomes, the compound was found to be stable; furthermore, the pharmacokinetic profiles of this inhibitor in rats and dogs suggested that it would be suitable for in vivo applications60. A patent has described isoxazoleazepines that target BET bromodomains, which include compounds with IC50 values below 500 nM in assays that measured the inhibition of LPS-induced IL‑6 secretion in THP‑1 cells and in mice105. 3,5‑dimethyl-isoxazole template (acetyl-lysine mimetic targeting BET and CREBBP bromodomains). The 3,5‑dimethyl-isoxazole template was first used to develop ligands against BET bromodomains using a structurebased approach, whereby an initial fragment hit was gradually augmented and used to obtain a co‑crystal structure with the first bromodomain of BRD4, giving insight into potential points of expansion of the chemical scaffold. Despite the small size of these molecules, they have good ligand efficiency and selectivity; isoxazole 4d has IC50 values of 1.6 μM and 4.8 μM for in vitro binding to the first bromodomain modules of BRD2 and BRD4, respectively, as well as good ligand efficiency values (0.43 and 0.39, respectively)57. The scaffold was further optimized, yielding isoxazole 9, which had improved potency for the first bromodomain of BRD4 (IC50 value = 371 nM) and retained good ligand efficiency (0.36)58. The isoxazole scaffold was also used for the develop­ ment of the highly potent and specific BET inhibitor I-BET151, which also had improved pharmacokinetic properties compared to triazolobenzodiazepine scaffolds59. I-BET151 is a dimethylisoxazole that selectively binds to bromodomains of the BET subfamily; its binding mode has been extensively validated using chemoproteomics and crystal structure data, and is very strong (Kd = 100 nM; determined by SPR)11,107. This inhibitor blocks the clonogenic growth of MLL-fusion-driven leukaemia cells by altering transcriptional programmes that are responsible for cell-cycle progression and apoptosis11. The 5- and 6‑isoxazolylbenzimidazole template was developed to introduce selectivity between the structurally similar bromodomains of BRD4 and CREBBP61. A two-step synthesis followed by regioisomer separation, or an elegant three-step regioselective synthesis, yielded a chemotype (compound 15) with multiple positions for modification, high potency (IC50 value = 180 nM for the first bromodomain module of BRD4, determined by a competitive peptide displacement assay) and over 100‑fold greater selectivity against the bromodomain of CREBBP. This template has been further optimized: an initial fragment hit was gradually augmented to a benzoimidazole scaffold carrying a dimethylisoxazole acetylated lysine mimetic, and through several structure-guided chem­ istry iterations this template resulted in SGC‑CBP30 (also known as isoxazole 61); see the SGC‑CBP30 overview and the I-CBP112 overview on the SGC website for further information. Preliminary data suggest that SGC‑CBP30 is

344 | MAY 2014 | VOLUME 13

www.nature.com/reviews/drugdisc © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS the first to have a very high affinity for the bromodomains of CREBBP and E1A‑associated protein p300 (EP300), and high selectivity over BET bromodomains. In an effort to improve the solubility of the 7‑isoxazolo­ quinoline ligands that had previously been described to target BET bromodomains and also as potent upregulators of APOA1 expression in stably transfected human hepatocellular carcinoma (HepG2) cells93, an isoxazole template fused to 1,5‑naphthyridine derivatives was developed. The resulting series of compounds had  single-digit micromolar IC50 values in vitro against the bromodomains of BRD2, BRD3 and BRD4. Two compounds from this series (compound 36 and compound 38) had good cellular activity and were found to be suitable for chronic oral administration. They had antiinflammatory effects in a mouse model of acute inflammation62, as discussed below in the section discussing bromodomain inhibitors in inflammation. The 3,5‑dimethyl-isoxazole template (which is an acetyl-lysine mimetic), in combination with quinoline templates, has been disclosed in patents108,109 (described below). Compounds from these series were active (with IC50 values of 100‑fold higher selectivity for the bromodomains of BRD4, CREBBP and BRD9 over the bromodomain of cat eye syndrome critical region protein 2 (CECR2), with IC50 values between 150 nM and 200 nM. This compound

NATURE REVIEWS | DRUG DISCOVERY

VOLUME 13 | MAY 2014 | 345 © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS displaced a multi-bromodomain construct of CREBBP from chromatin in a human osteosarcoma cell line (U2OS cells), demonstrating that it had on‑target properties in cells. It will be interesting to test promiscuous templates such as these in models of diseases such as leukaemia, where the bromodomains of BRD4, BRD9 and CREBBP — which are targeted by the template — seem to be part of the same complex 65. Kinase inhibitors with bromodomain activity. The screening of a library of diverse kinase inhibitors, which included tool compounds and clinically approved inhibitors, recently identified several compounds that had low nanomolar activity against BRD4 (REF. 66). Compounds that significantly inhibited the BRD4–histone peptide interaction include the polo-like kinase (PLK) inhibitor BI‑2536 and the related compound volasertib (also known as BI‑6727), the ribosomal S6 kinase (RSK) inhibitor BI‑D1870, the Janus kinase (JAK) inhibitor fedratinib (also known as TG‑101348), the p38 mitogenactivated protein kinase (p38 MAPK) inhibitor SB‑203580, the dual specificity protein kinase TTK (also known as MPS1) inhibitor AZ‑3146, an isoxazoloquinoline developed for targeting PIM kinases, the focal adhesion kinase (FAK) inhibitor PF‑431396, and the phosphoinositide 3‑kinase (PI3K) and mammalian target of rapamycin (mTOR) inhibitors GSK2636771 and PP‑242, respectively. Moreover, the clinically used cyclin-dependent kinase (CDK) inhibitor dinaciclib inhibits BET bromodomains123, as does the PI3K inhibitor LY294002 (REF. 124), albeit at a much lower potency. A similar study found that several kinase inhibitors bound to the first bromodomain of BRD4, and their mode of interaction was established in co‑crystal structures. These inhibitors included the PLK1 inhibitor BI2536, TG101209 (which inhibits JAK2, RET and FMS-like tyrosine kinase 3 (FLT3)), the JAK2 inhibitor TG101348, NU7441 (which inhibits the DNA-dependent protein kinase (DNAPK) and PI3K) and the pan-kinase inhibitor GW612286X. Further inhibitors included SB610251B (which inhibits p38α MAPK, RET and SRC), SB614067R (which inhibits BRAF, p38α MAPK and lymphocyte-oriented kinase (LOK; also known as STK10)), the p38α MAPK and p38β MAPK inhibitors SB202190, SB251527 and SB284827BT, the CDK inhibitor flavo­ piridol, the glycogen synthase kinase-α (GSKα) and GSKβ inhibitor SB409514, dinaciclib (which inhibits CDK1, CDK2 CDK5 and CDK9) as well as fostamatinib (which inhibits spleen tyrosine kinase and Bruton’s tyrosine kinase)125. The high hit rate from kinase libraries in bromodomain screens is surprising and suggests that these two diverse target families share binding site characteristics. However, the acetyl-lysine mimetic groups found on these kinase inhibitors do not always coincide with the chemical moieties that are important for kinase inhibition, and may point towards the solvent space in kinases or interact with the conserved Lys–Arg pair located in the kinase back pocket. Of particular interest are the PLK1 inhibitor BI‑2536 and the dual JAK2–FLT3 kinase inhibitor fedratinib, both of which show high potency towards

the first bromodomain of BRD4 (with Kd values of 37 nM and 164 nM, respectively). These compounds acted as BET inhibitors in cellular assays at clinically relevant concentrations. Importantly, the structural comparison of bromodomain and kinase binding modes indicated features that may allow the rational design of equipotent dual inhibitors of kinases and bromodomains66.

Fragment templates Screening of fragments, either in vitro or in silico, has resulted in the identification of several novel chemotypes. For example, virtual screening of over 9 million compounds from the Dictionary of Natural Products, the ChEMBL database126 and the ZINC library yielded hits that were further characterized biophysically in order to assess binding to the first bromodomain of BRD4 (REF. 67). The confirmed hits comprised six novel inhibitors with affinities in the nanomolar to low micromolar range, and included the poisonous alkaloid colchiceine (XD1) and its methoxylated analogue colchicine (XD25) as well as the 4‑acyl pyrrole compound XD14. Highresolution crystal structures, which provided insight into the binding modes of these compounds, offered novel starting points for further structure-based design. XD14, which was the most potent compound from the validated hits and had a very good ligand efficiency value of 0.31, inhibited the growth of human cancer cell lines67. A number of  small-molecule fragments68–71 that bind to bromodomains with weak affinities may yield potent and selective bromodomain inhibitors following further optimization. To establish novel starting points for the design of inhibitors, molecular docking and crystal-structure-guided lead optimization of fragment libraries have been used68,69. For instance, when 487 commercially available drug-like fragments from the ZINC library were docked onto the available structure of the first bromodomain of BRD4 in complex with (+)-JQ1, which was followed by co‑crystallization studies with the same bromodomain, this resulted in the identification of several novel 2‑thiazolidinone scaffolds that bound to this bromodomain. Further optimization resulted in compounds that had IC50 values in the low micromolar to high nanomolar range and inhibited the proliferation of HT‑29 human adenocarcinoma cells with low GI50 values of 20–60 μM68. Although these ligands were designed to target the first bromodomain of BRD4, the 2‑thiazolidinone template is a promising starting point for developing potent and selective inhibitors of acetyl-lysine–bromodomain interactions68. Triazolopyrimidines, methylquinolines and chloropyridinones69 were identified as weak BET bromodomain binders though structure-guided virtual screening that was based on either published acetyl-lysine mimetics to generate substructures that maintained the key pharmacophore (for example, 1,2,4‑triazole to iso­xazole), or through pharmacophore shape as well as similarity searches based on the two‑dimensional structure derived from the published complex of (+)-JQ1 crystallized with the first bromodomain of BRD4. Structural characterization confirmed the predicted binding modes, which suggests that this is a good

346 | MAY 2014 | VOLUME 13

www.nature.com/reviews/drugdisc © 2014 Macmillan Publishers Limited. All rights reserved

REVIEWS method for rapidly identifying novel hits that can be expanded and optimized to yield potent inhibitors. Similar efforts have also yielded indolizine-ethanones and N‑phenylacetamides70 as well as methyltriazoles71 as scaffolds for potential expansion. The bromodomain of BAZ2B binds to several small fragment molecules71. In vitro screening of 1,300 compounds resulted in a number of weak hits (with affinities in the range of 40 μM to 1 mM). Structure-guided optimization of one scaffold led to compound 6, which exhibited a Kd of 65 μM against the bromodomain of BAZ2B71. Preliminary data suggest that GSK2801, which was developed from the indolizine-ethanone template, is highly potent and selective for the bromodomains of BAZ2B and BAZ2A (see the GSK2801 overview on the SGC website for further information). Other bromodomain inhibitors disclosed in patent applications. Other inhibitors that are not part of the categories described above have also been disclosed in patents. These include triazolopyridazine‑6‑amines127, which were active against BRD4 and modulated MYC RNA levels in a Burkitt lymphoma cell line, as well as N‑methyl­ pyrrolopyridinones and N‑methylpyrrolopyridazinones128, which were described as potential agents for the treatment of inflammatory diseases, cancer and AIDS, and had nanomolar EC50 values (effector concentration for halfmaximum response) in cellular assays and were active at low concentrations (