Preventing histone recognition by bromodomains emerges as an attractive therapeutic approach in cancer. Overexpression of ATAD2 (ATPase family AAA domain-containing 2 isoform A) in cancer cells is associated with poor prognosis making the bromodomain of ATAD2 a promising epigenetic therapeutic target. In the development of an in vitro assay and identification of small molecule ligands, we conducted structure-guided studies which revealed a conformationally flexible ATAD2 bromodomain. Structural studies on apo–, peptide–and small molecule–ATAD2 complexes (by co-crystallization) revealed that the bromodomain adopts a ‘closed’, histone-compatible conformation and a more ‘open’ ligand-compatible conformation of the binding site respectively. An unexpected conformational change of the conserved asparagine residue plays an important role in driving the peptide-binding conformation remodelling. We also identified dimethylisoxazole-containing ligands as ATAD2 binders which aided in the validation of the in vitro screen and in the analysis of these conformational studies.
The epigenetic regulation of chromatin structure and function is critical for transcriptional control, cell identity and development. Several groundbreaking studies have demonstrated the importance of post-translational modifications of histones, proteins and DNA for the epigenetic control of gene expression [1–3]. Notably, the identification of chromatin-associated proteins that catalyse, recognize and remove these chemical markers has led to the concept of the ‘histone code’ and the idea of writers, readers and erasers . These proteins work in concert with transcription factors to control fundamental cellular processes including proliferation, development, differentiation and genome integrity. Consequently, disruption of epigenetic control can lead to aberrant gene expression and human disease. Indeed, growing genomic evidence links mutations, amplifications, deletions and rearrangements of genes encoding epigenetic regulators to the development of human cancer. Consequently, therapeutic targeting of chromatin-associated proteins is a growing area of drug discovery. In fact, the first epigenetic-based therapies for cancer treatment which act globally on these post-translational processes (i.e. histone deacetylase inhibitors and inhibitors of DNA methylation) have been approved. However, preclinical evidence suggests that targeting specific writers [such as the histone methyl transferases EZH2 (enhancer of zeste homologue 2) and DOT1L (disruptor of telomere silencing 1-like)] or reader proteins [such as the BET (bromodomain and extra-terminal) family of bromodomains] may be beneficial in defined patient populations and selected tumour types. For example, DOT1L may be useful in 11q23-related leukaemias associated with mixed lineage leukemia (MLL)-rearrangement and inhibition of the reader protein BRD4 (bromodomain containing protein 4) may prove beneficial in certain tumour types as BRD4 appears critical for Myc protein (MYC) transcriptional pathways .
ATAD2 (ATPase family AAA domain-containing 2 isoform A) is an epigenetic regulator that has been highlighted as a promising target for anti-cancer therapeutic intervention . Studies revealed that high ATAD2 expression is significantly associated with poor survival and progression of cancers including lung and prostate [6,7]. Copy-number gain and co-amplification of ATAD2 with c-MYC is also observed across many tumours and it is the most significantly up-regulated bromodomain-containing gene across the TCGA (The Cancer Genome Atlas) tumour atlas. In addition, ATAD2 has been characterized as a transcription co-regulator of the hormonally-controlled developmental process interacting with the oestrogen receptor (ERα) and the androgen receptor (AR) in breast and prostate cancer respectively [8,9]. Specifically, ATAD2 expression has been hypothesized to lead to an amplification loop in these cancers driving highly proliferative transcriptional programs including EF2 (elongation factor 2) and cMYC. Finally, numerous knockdown studies of the ATAD2 protein in cancer (and normal) cells suggest a potential role in cell proliferation and transformation [6,8,9].
The domain architecture of ATAD2 includes an AAA ATPase domain and a bromodomain, which could both be targeted by small molecule inhibitors . Although these inhibitors of AAA ATPases have been reported for dynein  and p97/VCP (valosin-containing protein) [12,13], this enzyme class is challenging from an inhibitor selectivity/specificity point of view. In contrast, selective bromodomain inhibitors have emerged with selective BET-family bromodomain inhibitors showing promising preclinical  and clinical activity [14,15]. Bromodomains are structurally and evolutionarily conserved ~110 amino acid modules present in a large number of chromatin-associated proteins (including ATPase-dependent chromatin remodelling complexes and in nearly all nuclear histone acetyltransferases). For the majority of bromodomains, there is little definitive knowledge about their cognate sequences and physiological recognition motifs.
The purpose of our study was to identify chemical probes in order to develop ligands of the ATAD2 bromodomain. We hypothesized that we could obtain inhibitors by targeting the acetyllysine-binding pocket, as reported for other bromodomains [16–18]. However, prior to embarking on a medicinal chemistry effort, structural biology was used to identify acetyllysine pocket-binders, such as histone peptides or small molecule probes, in order to characterize their binding mode at the molecular level. To do this we first identified the Histone 4 Lys5 acetyl peptide (H4K5ac) as an active site binding partner using biophysical and X-ray crystallographic approaches. We focused on co-crystallization rather than soaking technique and extensive characterization of the H4K5ac complex revealed an unexpected conformational change accompanying the peptide recognition and provided an undescribed ATAD2 form in a closed state. Including structural data from a novel empty form of ATAD2 bromodomain we compared a total of six states, confirming experimentally the evidence of protein dynamics that has been described by molecular dynamics simulation  and providing a snapshot of the mechanism for peptide recognition and accompanying loop flexibility. With the recent identification of ATAD2 small-molecule binders by our laboratory and others [20,21], our study highlights that conformational flexibility of the bromodomain of ATAD2 has implications for histone–peptide recognition towards the development of ATAD2 bromodomain inhibitors.
Peptides used for α-screen assay, direct binding and crystallization studies
Biotinylated and non-biotinylated peptides used for α-screen assay, ITC (isothermal titration calorimetry) and crystallization were purchased from Anaspec: H2AK36ac (19–38)-biotin, SRAGLQFPVGRVHRLLRK(ac)GNK-biotin; H2BK85ac (81–93)-biotin, AHYNK(ac)RSTITSREK-biotin; H3K14ac (1–21)-biotin, ARTKQTARKSTGGK(ac)APRKQLAGGK-biotin; H3K56ac (44–57)-biotin, GTVALREIRRYQK(ac)SK-biotin; H4K5ac (1–25)-biotin, SGRGK(ac)GGKGLGKGGAKRHRK-VLRDNGSGSK-biotin; H4K8ac (1–25)-biotin, SGRGKGGK-(ac)GLGKGGAKRHRKVLRDNGSGSK-biotin; H4K12ac (1–25)-biotin, SGRGKGGKGLGK(ac)GGAKRHRKVLRDNGS-GSK-biotin; H4K5/8/12/16-ac4 (1–25)-biotin, SGRGK(ac)-GGK(ac)GLGK(ac)GGAK(ac)RHRKVLRDNGSGSK-biotin; H4 (1–25)-biotin, SGRGKGGKGLGKGGAKRHRKVLRD-NGSGSK-biotin; H4(1–20), SGRGKGGKGLGKGGAKR-HRK; H4K5ac(1–20), SGRGK(ac)GGKGLGKGGAKRHRK; H3K14ac (1–21), ARTKQTARKSTGGK(ac)APRKQLAGGK.
Plasmids, cloning and site-directed mutagenesis
The Escherichia coli strains BL21 (DE3) star were purchased from Invitrogen. The ATAD2 (EC 184.108.40.206) bromodomain coding sequence was amplified using a pNIC28–Bsa4 vector provided by the structural genomics consortium. WT (wild-type) and mutated genes were sequenced and the corresponding plasmids were then used for transforming the E. coli strain BL21 (DE3) star for protein expression. The transformants were grown at 22°C in Luria–Bertani (LB) medium in the presence of kanamycin at 50 mg/ml for 24 h. The expression of both recombinant proteins was then induced by the addition of 1 mM IPTG and growth was continued for 24 h at 16°C. The cells were then pelleted by centrifugation and served as the source for protein purification. Soluble protein was purified using Ni–NTA (nitrilotriacetic acid; Qiagen) gravity flow affinity chromatography followed by tobacco etch virus (TEV) cleavage and size exclusion chromatography. The protein was concentrated to 12–14 mg/ml in 25 mM HEPES, pH 7.5, 150 mM NaCl and 10 mM DTT column buffer.
ATAD2 AlphaScreen assay
Recombinant human histidine-tagged ATAD2 (produced in-house), biotinylated H4K5ac (1–25) peptide and test compound were added to a 384-well OptiPlate (Perkin Elmer) and incubated at room temperature for 1 h. The assay buffer consisted of 50 mM HEPES, pH 7.5, 100 mM NaCl and 0.12 mM Triton X-100. Final concentrations for this reaction were as follows: 100 nM histidine–ATAD2, 100 nM peptide, variable concentrations of compound (3-fold serial dilutions) and 1% (v/v) DMSO. Streptavidin donor beads and nickel chelate acceptor beads, both from a PerkinElmer AlphaScreen Histidine Detection Kit, were then added to the optiplate to a final concentration of 10 μg/ml each. Following 2 h incubation at room temperature, the microplate was read on an Envision Plate Reader (PerkinElmer). Percentage of control (POC) values were calculated from the following formula: POC = sample signal − average background signal/average maximum signal–average background signal × 100. The average maximum signal was obtained from wells containing all assay components except test compound. The average background signal pertained to wells with all assay components except ATAD2 and test compound.
Measurement of small molecule Kd by DiscoverX and ligand efficiency calculation
Crystallization conditions for the double-mutant alanine Y1063A–N1064A were similar to the conditions reported by the structural genomic consortium for WT-ATAD2 . Protein crystallized at 4°C by vapour diffusion in a hanging drop with a crystallization buffer containing 2 M ammonium sulfate, 0.1 M Bis–Tris, pH 5.5. Initial co-crystallization conditions of WT-ATAD2 in complex with H3K14ac and H4K5ac peptides were found by using a sitting drop-based sparse-matrix screening strategy at 4°C. Peptide solution stocks were diluted in the protein column buffer at 5 mM. Protein at 8–10 mg/ml was incubated with H3K14ac or H4K5ac peptide to reach a 1:3 ratio. For both complexes, the best crystals were obtained after 3–5 days of equilibration using a sitting drop method by mixing 1 μl of the peptide-protein solution with an equal volume of the reservoir equilibrated over 0.3 ml of the crystallization buffer at 4°C. Co-crystals with H4K5ac were obtained at 4°C using a crystallization buffer containing 0.1 M ammonium acetate, 100 mM Bis–Tris, pH 5.5, 15%–20% w/v PEG-10000. Co-crystals with H3K14ac were obtained at 4°C using a crystallization buffer containing 2.0 M ammonium sulfate, 100 mM Tris, pH 8.5. ATAD2 crystals showing the double conformation of Asn1064 were obtained at 4°C using a crystallization buffer containing: 0.2 M ammonium sulfate/0.1 M Bis–Tris, pH 5.5/25% w/v PEG-3350. For cryoprotection, all crystals were transferred in their relative crystallization buffer supplemented by 20% glycerol. Co-crystallizations with compound 1 and compound 2 were performed at room temperature with a hanging drop method by mixing an equal volume of the protein incubated with the compound of interest (1 mM final concentration 1% DMSO) and the solution reservoir containing 2.1–2.4 M ammonium sulfate, 0.1 M buffer (Bis–Tris, pH 5.2–5.7), 10% glycerol. In order to get 100% occupancy in the binding site, co-crystals with compounds were next soaked for 24 h in a pre-equilibrated hanging drop containing 2 M ammonium sulfate, 0.1 M buffer Bis–Tris, pH 5.5, 10% glycerol and the compound of interest at 10 mM final concentration.
All X-ray diffraction data sets were collected from frozen single crystals at the Advanced Light Source (beamline 8.3.1) and processed with the programs ELVES  and/or MOSFLM, SCALA and TRUNCATE from the CCP4 program suites . Molecular replacement solutions were obtained using the BALBES molecular replacement pipeline  and the crystal structure PDB ID: 3DAI . Iterative model rebuilding and refinement were performed by using the programs COOT and REFMAC5 [28,29]. Figures of the different structures of ATAD2 were generated using PyMOL (DeLano Scientific).
Isothermal titration calorimetry
ITC measurements were carried out at 10°C using the nano ITC (TA instruments) with a stirring of 250 rpm. ITC titrations were performed with one 0.2-μl injection of peptide, followed by 20 consecutive injections of 2.0 μl of the peptide with injection durations of 8 s and 240 s intervals between injections. The sample chamber was filled with 100 μM of WT or mutant ATAD2 protein. Solution stocks of peptides were prepared at 5 mM using the column buffer used during the protein purification. Solutions of peptide titrants were next freshly prepared by diluting them to reach a 1.5 mM final concentration. Titrant solutions and protein were centrifuged for 10 min at 14000 g, and supernatants were loaded into the syringe and cell respectively. The final molar ratio of ligand:protein exceeded 2.5. The heat released was measured following each injection. Data from the experiment were analysed with Nanoanalyze software (TA instruments) supplied by the manufacturer.
Synthetic route to compound 2
To a solution of 3,5-dimethylisoxazol-4-yl)boronic acid (123 mg, 0.869 mmol) and methyl 3-amino-5-bromobenzoate (200 mg, 0.869 mmol) in dimethoxyethane was added sodium carbonate powder (193 mg, 1.83 mmol) in water (1 ml) and palladium tetrakis(triphenylphosphine) catalyst (30 mg, 3 mol%). The reaction mixture was degassed by sparging with a stream of nitrogen and then refluxed for 16 h. The cooled reaction mixture was partitioned between water and ethyl acetate, the organic layer was dried over sodium sulfate, filtered, concentrated and then purified by flash chromatography (1:1 ethyl acetate/hexanes) to give methyl 3-amino-5-(3,5-dimethylisoxazol-4-yl)benzoate as a pale-yellow solid (190 mg, 89%). MS-ESI: m/z calculated for (C13H14N2O3 + H)+ 247.1, found 246.8. 1H NMR (600 MHz, d6-DMSO) δ 7.20 (s, 1H), 7.01 (s, 1H), 6.78 (s, 1H), 5.54 (s, 2H), 3.82 (s, 3H), 2.38 (s, 3H), 2.20 (s, 3H).
Methyl 3-(3,5-dimethylisoxazol-4-yl)-5-(phenylsulfonamido)benzoate (compound 1)
A solution of methyl 3-amino-5-(3,5-dimethylisoxazol-4-yl)benzoate (30 mg, 0.122 mmol) and pyridine (0.029 ml, 0.365 mmol) in dichloromethane (1 ml) was treated with benzenesulfonyl chloride (22 mg, 0.122 mmol) and the reaction mixture stirred at ambient temperature for 1 day. The reaction mixture was concentrated then purified by mass-directed prep-HPLC (mobile phase: A=0.1% TFA (trifluoroacetic acid)/water, B=0.1% TFA/acetonitrile; gradient: B=30%–70% in 12 min; column: C18) to give methyl 3-(3,5-dimethylisoxazol-4-yl)-5-(phenylsulfonamido)benzoate as a white solid (14 mg, 30%). MS-ESI: m/z calculated for (C19H18N2O5S + H)+ 387.1, found 386.8. 1H NMR (600 MHz, CDCl3) δ 7.83 (d, J=7.5 Hz, 2H), δ 7.67(d, J=1.9 Hz, 2H), 7.58 (t, J=7.8 Hz, 1H), 7.48 (t, J=8.0 Hz, 2H), 7.29 (t, J=1.9 Hz, 1H), 7.20 (s, 1H), 3.92 (s, 3H), 2.34 (s, 3H), 2.19 (s, 3H).
3-(3,5-Dimethylisoxazol-4-yl)-5-(phenylsulfonamido)benzoic acid (compound 2)
A solution of methyl 3-(3,5-dimethylisoxazol-4-yl)-5-(phenylsulfonamido)benzoate (14 mg, 0.036 mmol) in water (0.3 ml), methanol (0.3 ml) and THF (tetrahydrofuran; 0.3 ml) was treated with lithium hydroxide hydrate (3.0 mg, 0.072 mmol). The reaction mixture was heated to 50°C for 4 h. The cooled reaction mixture was then acidified with aqueous HCl (0.012 ml, 0.12 mmol) and purified by prep-HPLC (mobile phase: A=0.1% TFA/water, B=0.1% TFA/acetonitrile; gradient: B= 20%–50% in 12 min; column: C18) to give compound 2 as a white solid (5 mg, 37%). MS-ESI: m/z calculated for (C18H16N2O5S + H)+ 373.1, found 373.4. 1H NMR (600 MHz, d6-DMSO) δ 13.20 (br-s, 1H), 10.66 (s, 1H), 7.80 (d, J=7.8 Hz, 2H), 7.68 (s, 1H), 7.64 (t, J=7.6 Hz, 1H), 7.58 (t, J=7.6 Hz, 2H), 7.54 (s, 1H), 7.26 (s, 1H), 2.31 (s, 3H), 2.11 (s, 3H).
Quantification of the relative movement in the ZA loop
The relative position of Val1018 has been used to quantify the ZA loop movement compare with the apo structure of reference PDB ID: 3DAI. The relative distance between the Cγ2 of Val1018 and the atom Cγ from the acetyllysine side chain [Δdist. = d Val1018 (Cγ2) − K5ac (Cγ)structure − d Val1018 (Cγ2) − K5(ac) (Cγ)reference] has been measured for each state. Positive differences indicate wider conformations and negative differences indicate narrow conformations.
ATAD2 bromodomain recognizes H4K5ac peptide
In the search for chemical probes able to disrupt the ATAD2-acetylated histone complex, we sought to develop a high-throughput bromodomain histone peptide displacement assay. The ATAD2 bromodomain is an acetyllysine reader and histone interactions have been reported with H4K5ac or H3K14ac peptides based on co-immunoprecipitation studies in mouse and human cells respectively [7,30]. However, to our surprise, such interactions have not been confirmed from panning of histone peptide panels by peptide array . We used a high-sensitivity AlphaScreen assay widely used for bromodomains [31,32] to screen for acetylated histone peptides. Interactions are detected by using a chemiluminescent read-out triggered by the close proximity between a donor-bead-coated peptide and an acceptor-bead-coated protein. We screened a panel of histone peptides and characterized H4K5ac as a specific acetyllysine binder with a Kd of 22 μM (Supplementary Figures S1a and S1b). Despite the previous report  of H3K14ac recognized by ATAD2, we were surprised to find it did not respond in our AlphaScreen assay. We evaluated the structural basis for H4K5ac recognition by ATAD2 and investigated the H3K14ac binding mode using X-ray crystallography. The assay development and the acetyllysine binding validation studies are documented in Supplementary Notes S1 and S2, Supplementary Table S1 and ‘Methods’ sections.
Structural basis of H4K5ac read-out by ATAD2
The ATAD2 bromodomain contains a canonical left-handed four-helix bundle (αZ, αA, αB and αC), topped with two loops; ZA and BC loops (Figure 1a). The acetyllysine-binding pocket is described structurally by a cavity created with two helices (αB and αC) and the ZA and BC loops. Specific features of ATAD2 reported in Figure 1(a) highlight the presence of a specific arginine-valine-phenylalanine (RVF) shelf and proline residues located in the loops that may confer a substantial degree of flexibility to the structure.
Structural basis for H4K5ac readout by ATAD2
Following the submission of our manuscript, a group from the structural genomic consortium identified H4K5ac peptide as an ATAD2 binder. Despite packing constraints, an X-ray structure obtained by peptide soaking confirmed the interaction . Because recent molecular dynamics studies suggested some flexibility of the ZA and BC loop on bromodomains , in the present study we prioritized a co-crystallization strategy. We identified alternative crystallogenesis conditions and obtained the complex by co-crystallization. This complex allowed us to improve our knowledge of the H4K5ac readout and demonstrated that the peptide binds to the acetyllysine-binding site, as illustrated by the Fo–Fc omit map (Supplementary Figure S2a). As seen for the recent complex obtained by soaking , the polypeptide follows the groove created by the ZA and BC loops. In our co-crystal structure residues 1-SGRGKacG-6 were defined, allowing us to determine the molecular basis of the recognition. A summary of data collection and refinement statistics has been reported in Supplementary Table S2.
Both, hydrophobic contacts and hydrogen bonds contribute to the peptide readout. Five hydrophobic contacts, involving residues from the ZA loop (Val1008, Val1013, Val1018, Tyr1063) and helix αC (Ile1074), were found in the acetyllysine-binding pocket (Figure 1b). Notably, at the bottom of the cavity, all the hydrophobic interactions did not induce any side-chain rearrangements despite a narrow environment. These interactions engaged the sp2-hybridized carbon of the acetyllysine which interacts with both the ZA loop and the helix αC (through residues Val1013 and Ile1074 respectively) and the sp3-hybridized carbon which makes contact with the RVF shelf (Val1008). Two additional hydrophobic contacts stabilize the Cγ of the acetyllysine side chain in the upper part of the cavity (Val1018 and Tyr1063; Figure 1b). Taken together, these observations show that the hydrophobic contacts surrounding the acetyllysine cavity play a key role in peptide recognition.
In addition to hydrophobic contacts, two residues from the peptide at positions 3 and 5 (Arg3 and Lys5 acetyl) form hydrogen bonds and contribute to the specific nature of binding (Figure 1c). We observe the common signature of interactions of acetyllysine peptides to bromodomains  in which ATAD2 acetyllysine interacts directly with the conserved asparagine (Asn1064) and engages the ZA loop and αB helix via a series of hydrogen bonds through water molecules (Figure 1c). The Arg3 side chain interacts with two carbonyls of the protein's backbone at the top of αB helix (Glu1062 and Tyr1063; Figure 1c). Taken together, these observations highlight the requirement of Arg3 for the specificity of the acetyllysine peptide recognition.
ATAD2 loop motion facilitates histone recognition leading to a closed state
Superimposition of the apo  and our peptide-bound structures reveals evidence of both ZA and BC loop flexibility which accompanies the recognition of H4K5ac (Figure 2a). Computational studies have suggested apparent bromodomain flexibility [19,33,34], however no loop rearrangements have been described experimentally in the bromodomain family upon histone binding, describing an additional mode of molecular recognition for bromodomains. Thus, we decided to compare the movement of each loop in detail.
ATAD2 loop motion accompanies histone recognition
In the peptide bound complex, the ZA loop movement shrinks the acetyllysine-binding site allowing Val1013 and Val1018 to shift inward toward the cavity (by 1.1 at the bottom and 1.7 Å 1 Å=0.1 nm respectively; Figure 2b). The ZA loop movement is mediated by two prolines (Pro1012 and Pro1028) which act as hinges (Figure 2b) which agrees with Pizzitutti et al.  who showed that Pro371 of the yeast transcriptional adaptor Gcn5p bromodomain (corresponding to Pro1012 in the present study) plays a key role in the molecular recognition of the acetylated histone H4 tail. Many proteins require loop motion for the specific recognition of their substrate, cofactor or protein partners in a closed form , therefore we can speculate that in ATAD2, a closed form that is able to accommodate the acetyl histone peptide, exists.
BC loop flexibility also accompanies the peptide recognition. In the H4K5ac complex, binding of the histone is associated with a concomitant flip of the Pro1065 peptide bond and the rearrangements of two residues (Asp1066 and Arg1075) which stabilize the BC loop in a peptide-compatible conformation (Figure 2c).
Evidence of conformational flexibility of ATAD2
Because we captured ATAD2–H4K5ac complex in a closed form, we next investigated whether ATAD2 could also adopt multiple conformations in the absence of peptide. Such conformations could offer attractive opportunities for structure-guided drug design intervention and paradoxically the flexibility may explain the weak affinity of ATAD2 ligands and thus the recent experimental observation that ATAD2 is challenging to drug [20,21]. Thus, we screened for new ATAD2 crystallogenesis conditions and obtained crystals in a number of different conditions. Among them, one crystal provided us with four novel peptide-free structures of apo-ATAD2. Comparison with the reported apo structure  with our four additional apo forms and the H4K5ac-bound complex provides an opportunity to understand the implications of dynamic flexibility in the ATAD2 bromodomain.
Superposition of the four additional apo forms revealed differences in ZA loop positions (Figure 3a). A visual graphical scheme represented in Figure 3(b) highlights the relative movements of the ZA loop and describes six different states caught by X-ray crystallography. To compare the states, we used as a reference the position of the ZA loop observed in the ligand-free structure PDB ID: 3DAI, which is defined as state ①. Of the four additional apo forms, three of them are wider (states ②, ③ and ④) than the reference and one form adopts an intermediate closed conformation (state ⑤), which was found in the H4K5ac-bond form (state ⑥). Overlay of the X-ray crystal structures of ATAD2 bound to fragments recently published [20,21] showed that ligands binds in state ①. Surprisingly in one of the open states (state ④), our density map showed an unexpected dual conformation for the conserved asparagine (Supplementary Figure S2b). In the predominant orientation, Asn1064 adopts a position commonly found in the bromodomain family by pointing within the acetyllysine-binding cavity (occupancy of 60%). Strikingly, in its alternate conformation, Asn1064 faces the BC loop and engages in a hydrogen bond with the peptide-flipped carbonyl of Pro1065, assisting the transition into the histone-compatible conformation (Figure 3c). The reversal of Asn1064 is supported by two additional hydrogen bonds involving residues from helix αC (Asp1071 and Arg1075). Thus, by assisting the peptide-bond flip of Pro1065, Asn1064 triggers the BC loop conformational change from the apo form to a histone-compatible state. Therefore, we can hypothesize that we caught a pre-activated state which sheds light on to an undescribed structural and functional role of the evolutionarily conserved asparagine in ATAD2.
Conformational flexibility of ATAD2
H3K14ac binds to the C-helix of ATAD2, a potential interface for protein–protein interaction
In the absence of a response from H3K14ac in our AlphaScreen assay, we attempted to co-crystalize ATAD2 in the presence of this histone peptide to check for specificity. Novel crystallogenesis conditions were found. Surprisingly, the H3K14ac-bound complex solved at 2.7 Å reveals that H3K14ac interacts along the end of αC helix, far from the acetyllysine-binding site (Figure 4a). The binding pose showed that the acetyllysine hydrogen bonds with two residues (Arg1005, His1076) and highlights the presence of a groove which accommodates the peptide backbone between the αZ and αC helices (Figure 4b). However, since bromodomains are protein–protein partner modules, we questioned whether this interaction mapped a previously unrecognized protein–protein interface. We performed computational simulations using Sitemap  and identified a potential protein–protein interface at the surface of ATAD2 which partially overlaps the experimentally observed H3K14ac-binding pose (Figure 4c). Confirmation of the binding of H3K14ac at a site distal from the expected binding site raises the possibility of an additional binding surface which might have important implications in bromodomain biology.
H3K14ac binds non-specifically to ATAD2
Isoxazole-based small molecules used as ATAD2 chemical probes targeting the acetyllysine pocket.
There were no reported ligands for ATAD2 when we initiated our studies, so we sought a suitable ligand with which to validate an ATAD2 AlphaScreen assay as well as to initiate a medicinal chemistry effort. We performed an X-ray crystallography screen with fragments as well as known acetyllysine mimetic templates. We identified two dimethylisoxazole-containing, compounds 1 and 2 (Figure 5a and Supplementary Figure S3), which is a template that are known inhibitors of the well-studied bromodomains of the BET family [37–40]. We liked the modular features that this template provided and envisioned that an appropriately bi-substituted version would allow one to access both the RVF-shelf and the ZA-channel. We also tested the known broad-spectrum bromodomain inhibitor bromosporine (Figure 5a) (http://www.thesgc.org/chemical-probes/bromosporine) in our AlphaScreen assay. We confirmed that the three molecules bound to the ATAD2 acetyllysine pocket (Supplementary Note S3, Supplementary Figure S3c), with high micromolar affinities observed in both the AlphaScreen and an orthogonal DiscoveRx competitive displacement assay (see Methods; Figure 5a). This is in contrast with the low micromolar affinities reported for similar dimethylisoxazoles against the BET bromodomain family and corresponding LEs of up to 0.39 [37–40]. Despite the weak affinities observed, compound 1 displayed a higher LE of 0.28 than compound 2 and bromosporine (0.20 and 0.21 respectively) and provides a reasonable starting point from which to design more potent inhibitors. The modest LEs observed are similar to other recently reported fragments [20,21] and probably reflect the lower druggability of ATAD2 compared with other bromodomains such as BRD4. However, since structure-guided analysis revealed that the H4K5ac complex adopts a closed form, we investigated whether ATAD2 could adopt this state in the presence of compound 1 and compound 2.
Structure-guided analysis of compound 1 and compound 2 in complex with ATAD2
The co-crystallization of compound 1 and compound 2 provided complexes in an ATAD2 open form (resolution of 1.90 Å and 1.74 Å respectively) similar to the previously reported structure [20,21,24]. All structures superimposed well, showing no significant rearrangements of the ZA and BC loops and the two complexes revealed an acetyllysine mimetic binding mode for both ligands (Supplementary Figure S3c). These results validate the strategy we employed to develop a robust AlphaScreen assay and provided further insight as to the structural requirements for ligand compared with peptide binding.
The binding pose of the dimethylisoxazole in compound 1 and compound 2 complexes revealed a common signature of key interactions, also reported in the BET family [37–40] (Figures 5b and 5c). In the ATAD2 bromodomain, the isoxazole moiety forms hydrogen bonds with Asn1064 and a conserved water molecule bridges the heterocycle to the conserved tyrosine Tyr1021 (Figures 5b and 5c). Therefore, compound 1 and compound 2 are validated acetyllysine mimetics. However, to enter at the bottom of the narrow cavity, both complexes show that the orientation of the isoxazole is guided by hydrophobic contact from the ZA loop (Val1013) and the 5-methyl substituent of the isoxazole (Figures 5b and 5c). Due to these constraints, side chains of Val1008 rearrange to prevent steric clashes, as exemplified by the binding mode of compound 1 (Figure 5d). It has been demonstrated that a conformational variability of the protein target can impair ligand affinity , therefore our structure-guided analysis provides an explanation of the weak affinity measured for these ligands. In addition, it is worth emphasizing that in the H4K5ac peptide-bound form, compounds 1 and 2 cannot enter the narrow acetyllysine-binding pocket which explains why the ligand-bound complexes were obtained in an open form. Despite a common signature, each molecule displays specific interactions. Whereas compound 1 interacts with the RVF shelf (salt bridge with Arg1007; Figure 5b), compound 2 loses this salt bridge. Instead, compound 2 expands to the ZA channel and interacts with the flexible part of the ZA loop through hydrogen bonds (Asp1014) and hydrophobic contacts (Arg1007, Lys1011, Pro1012, Glu1017 and Val1018; Figure 5c). Phenyl ring orientation of compound 1 makes adjusted hydrophobic contact with the RVF shelf. On the other hand, the equivalent phenyl ring in compound 2 is rotated away in order to accommodate where the sulfonamide derivation of the small molecule anchors (Supplementary Figure S3c). Therefore, the change of binding mode observed by X-ray crystallography correlates with the weaker binding measured in vitro for compound 2 compared with compound 1 (Figure 5a).
In the search for small molecule inhibitors of the human ATAD2-acetylated histone complex, we characterized the H4K5ac peptide as a specific, acetyllysine-binding site partner and highlighted evidence of protein dynamics in the ATAD2 bromodomain. From a structure-guided point of view, our findings are important because they clearly demonstrate that protein flexibility can impair protein–ligand interactions . To our knowledge, such flexibility has not been described by X-ray crystallography for any bromodomain. The proline residues which define the loop motion are conserved in the family (Supplementary Figure S4) suggesting that this loop flexibility is not unique to ATAD2. Sorting bromodomains based on their flexibility may complement the well-established structural classification of bromodomains  and could improve the specific classification based on the acetyllysine-binding sites that emerged from computational druggability analyses .
Molecular dynamics simulations conducted by Steiner et al.  predicted that the conserved asparagine in ATAD2 can adopt an alternate conformation. Nevertheless, the movement of Asn1064 we observed experimentally is completely novel. In the simulation, the Asn1064 side chain points only within the active site, whereas we demonstrated that the ATAD2 flexibility is much larger than predicted and Asn1064 can reverse its position to point between the αB and αC helix. Based on our X-ray crystallographic snapshot, we propose that Asn1064 assists the BC loop transition into a histone compatible conformation. However, the role of the conserved asparagine may not be generalized to the entire bromodomain family because some histone peptides do not interact with the BC loop. Taken together our data support the results found in the molecular dynamics simulation, and alternate conformations of Asn1064 have to be taken into account for future virtual screening campaigns.
From a rational design point of view, the predicted second binding site is not ideal for drug binding (Dscore 0.68) . However, our study demonstrates that ATAD2 is flexible, thus the second binding site might become more accommodating if the hydrophobic residues rearrange to create a more enclosed site. This has been shown for some therapeutic targets, as exemplified by the Bcl-2 (B-cell lymphoma 2) family with abbott laboratories (ABT)-737 inhibitor class . Although speculative, the H3K14ac complex combined with the computational predictions suggests a potential site for a protein or small molecule. Searching for chemical tools that could dock in an induced-fit second binding site might be envisioned to explore whether bromodomain rearrangement may provide a closed form allowing H4K5ac recognition.
Finally, the binding conformation of compounds 1 and 2 as well as other recently reported ligand [20,21], rearrangements of Val1018 may have an affect on ATAD2 ligand recognition. The flexibility of the ATAD2 bromodomain should be taken into account for further design of more potent ATAD2 inhibitors.
ATPase family AAA domain-containing 2 isoform A
bromodomain and extra-terminal
bromodomain containing protein 4
disruptor of telomere silencing 1-like
Histone 3 Lys14 acetylated peptide
Histone 4 Lys5 acetylated peptide
isothermal titration calorimetry
percentage of control
Yanai Zhan, Jennifer P. Bardenhagen, Elisabetta Leo, Xi Shi, Mary K. Geck Do and Jannik N. Andersen developed the AlphaScreen assay. Guillaume Poncet-Montange and Gilbert R. Lee purified protein. Guillaume Poncet-Montange and Paul Leonard prepared co-crystals. Guillaume Poncet-Montange solved and interpreted co-crystal structures. Alessia Petrocchi and Wylie S. Palmer synthesized compounds 1 and 2. Mario G. Cardozo performed computational studies. Guillaume Poncet-Montange, Wylie S. Palmer, Jannik N. Andersen, Philip Jones and John E. Ladbury directed the studies and interpreted the data. Guillaume Poncet-Montange, Wylie S. Palmer and John E. Ladbury wrote the paper with assistance from co-authors.
We thank Dr Todd Link and Dr Cindy Benod for technical help and critical review of the paper. We would also like to thank James Holton and George Meigs for technical help at the Advance Light Source synchrotron facility.
This work was supported by the M. D. Anderson Cancer Research Institute [institutional support]; beamline 8.3.1 was built by the University of California Campus-Laboratory Collaboration Grant with support from the National Science Foundation; the University of California, Berkeley; the University of California, San Francisco, the W. M. Keck Foundation and Henry Wheeler. Operation is supported by the National Institutes of Health [grant number GM073210, M082250, GM094625], the Department of Energy Integrated Diffraction Analysis Technologies, Plexxikon Inc and the M.D. Anderson Cancer Research Institute. The Advanced Light Source and all its beamlines are supported by the Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences Division, of the US Department of Energy [grant number DEAC02- 05CH11231] at Lawrence Berkeley National Laboratory.
Present address: School of Molecular and Cellular Biology, University of Leeds, LC Miall Building, Leeds, LS2 9JT, U.K.
PDB accession codes: atomic coordinates and structure factors have been deposited in the protein data bank under the accession code PDB ID: 4TT2, PDB ID: 4TT4, PDB ID: 4TT6, PDB ID: 4TTE, PDB ID: 4TU4, PDB ID: 4TU6.