The BRD3 ET domain recognizes a short peptide motif through a mechanism that is conserved across chromatin remodelers and transcriptional regulators

Members of the bromodomain and extra-terminal domain (BET) family of proteins (bromodomain-containing (BRD) 2, 3, 4, and T) are widely expressed and highly conserved regulators of gene expression in eukaryotes. These proteins have been intimately linked to human disease, and more than a dozen clinical trials are currently underway to test BET-protein inhibitors as modulators of cancer. However, although it is clear that these proteins use their bromodomains to bind both histones and transcription factors bearing acetylated lysine residues, the molecular mechanisms by which BET family proteins regulate gene expression are not well defined. In particular, the functions of the other domains such as the ET domain have been less extensively studied. Here, we examine the properties of the ET domain of BRD3 as a protein/protein interaction module. Using a combination of pulldown and biophysical assays, we demonstrate that BRD3 binds to a range of chromatin-remodeling complexes, including the NuRD, BAF, and INO80 complexes, via a short linear “KIKL” motif in one of the complex subunits. NMR-based structural analysis revealed that, surprisingly, this mode of interaction is shared by the AF9 and ENL transcriptional coregulators that contain an acetyl-lysine-binding YEATS domain and regulate transcriptional elongation. This observation establishes a functional commonality between these two families of cancer-related transcriptional regulators. In summary, our data provide insight into the mechanisms by which BET family proteins might link chromatin acetylation to transcriptional outcomes and uncover an unexpected functional similarity between BET and YEATS family proteins.

BET proteins play important roles in the regulation of gene transcription (9 -11). In mammals, BRD2/4 are essential for normal development; homozygous deletion of either gene in mice is embryonic lethal (12)(13)(14). BRD2 and BRD3 (but not BRD4) are highly enriched at erythroid gene promoters where their direct interaction with the transcription factor GATA1 regulates the expression of many erythroid genes (15,16). Furthermore, BRD2 and to a lesser degree BRD3 are found at CTCF-occupied sites where they contribute to genome organization and chromatin boundary function (17). As well as being targeted by several viruses during infection (18 -20), BET proteins have also emerged as key players in oncogenesis. Fusions of BRD3/4 with the protein Nuclear in Testis give rise to NUT midline carcinoma, an aggressive malignancy, and BRD4 has This work was supported by Australian National Health and Medical Research Council Grants APP1063301 and APP1058916 (to J. P. M.) and National Institutes of Health Grant R01DK054937 (to G. A. B.). The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This article contains Figs. S1 and S2, Tables S1-S5, and PDB validation report. The atomic coordinates and structure factors ( been shown to occupy super-enhancer regions that drive oncogenic transcription programs in a range of cancers (21)(22)(23)(24)(25). As a consequence, BRD4 in particular has become a major target for the design of cancer therapeutics that inhibit the acetyllysine-binding properties of its bromodomains (26 -28).
Although many studies have focused on the acetyl-lysinebinding specificity of BET bromodomains and on the design of small-molecule inhibitors of these domains, the roles of other parts of these proteins, and the biochemical mechanisms by which BET proteins actually regulate gene expression, are less well understood. For example, although the functional importance of BRD2/3 in erythropoiesis is clear and it is known that the direct interaction between GATA1 and BRD2/3 is important for GATA1 chromatin occupancy (6,15), the biochemical mechanism underlying these effects is not well understood.
The ET domain of BRD4 has been shown to be the cellular target of the viral proteins Kaposi's sarcoma-associated herpesvirus latency-associated nuclear antigen (KSHV-LANA), and the murine leukemia virus integrase (MLV-IN) (29 -31).
Recently, NMR structures of the BRD4 ET domain in complex with peptides derived from these viral proteins have been reported, as has the structure of a complex formed between BRD4 ET and a peptide from the histone methyltransferase NSD3 (32,33). Overall, however, the role played by the ET domain in normal biology remains relatively under-explored.
The remodeling of chromatin, an essential aspect of genomic function, is achieved by large, multisubunit protein complexes that contain ATP-dependent DNA translocase domains (34). Several classes of these chromatin-remodeling complexes have been defined in mammals, including (i) the Nucleosome Remodeling and Deacetylation (NuRD) complex, in which CHD3, CHD4, or CHD5 is the ATP-driven DNA translocase (35,36); (ii) the BAF and PBAF complexes, for which remodeling is driven by BRG1 or BRM (37,38); and (iii) INO80 family complexes, which are involved in DNA damage repair and the exchange of variant histones such as H2AX into nucleosomes (39,40).
The mechanisms by which these machines are directed to specific genomic loci are areas of active research. For example, the NuRD complex has been shown to colocalize with GATA-1 during erythroid differentiation and to interact directly with FOG-1 (41)(42)(43), which is recruited to these sites through its zinc finger-mediated interaction with GATA-1 (44,45). Intriguingly, the NuRD subunit CHD4 was among a number of cellular interaction partners identified in pulldowns that used a BRD4 fragment containing the ET domain as bait (46). This result raises the possibility that the BRD3 ET domain might also interact with CHD4 and potentially contribute to recruitment or retention of the NuRD complex to GATA-1 sites during hematopoiesis.
In an effort to better define the biochemical mechanisms by which BET family proteins regulate gene expression, we searched for binding partners of BRD3. Among partners that we identified were multiple members of the NuRD, BAF, and INO80 chromatin-remodeling complexes. Biochemical and sequence analysis led us to demonstrate that the ET domain of BRD3 was sufficient to mediate direct interactions with the CHD4, BRG1, and INO80B components of the respective complexes. In each case BRD3-ET recognized a short linear motif comprising a KIKL-like sequence. We determined three-dimensional structures of BRD3-ET bound separately to a CHD4 and a BRG1 KIKL peptide, demonstrating that the ET domain recognizes a wide range of chromatin-remodeling complexes via a conserved mechanism. Analysis of the structures also reveals that, strikingly, the architecture of the interaction is conserved in another class of transcriptional regulators: the AF9 and ENL YEATS-domain proteins that form part of the superelongation complex (SEC) and are prominent MLL fusion partners in a subset of acute leukemias (47). Our analysis provides general insight into the molecular mechanisms by which BET family proteins might regulate gene expression and points to a remarkable functional similarity between BET and YEATS family proteins.

Immobilized BRD3 can capture a number of gene-regulatory complexes
To delineate the mechanisms by which BRD3 regulates gene expression, we immobilized full-length murine FLAG-tagged Brd3 on anti-FLAG-Sepharose beads and treated the beads with nuclear extract from HEK293 cells (Fig. 1A). Among the 752 proteins retained by Brd3 that could be identified by MS, multiple members of several multiprotein complexes associated with gene regulation were observed. These complexes include NuRD, SWI/SNF, INO80, TFIID, TFIIH, Mediator, and the PAF complex, as well as a range of elongation, splicing, and DNA replication factors (Table 1 and Tables S1-S5). Of these, two of the proteins detected with the largest number of peptides were the chromatin-remodeling enzymes CHD4 and BRG1. An interaction between CHD4 and a C-terminal fragment of both BRD3 and BRD4 (that includes the ET domain) had previously been detected in a similar pulldown experiment (46).
To corroborate these interactions, we transfected HEK293 cells with a FLAG-Brd3 construct, captured the FLAG-Brd3 protein on anti-FLAG-Sepharose beads, ran it on SDS-PAGE, and probed the gel with antibodies against components of the NuRD complex (CHD4 and MTA2, Fig. 1B) or the BAF complex (BRG1 and BAF170, Fig. 1C). In both cases, the two complex subunits that were queried were clearly detected. To assess which part of Brd3 was responsible for interacting with the BAF complex, the experiment was repeated using Brd3 deletion constructs. Fig. 1D indicates that the ET domain is essential for the Brd3/BAF complex interaction.

BRD3-binding motif of CHD4 maps to a disordered region
It was shown previously that BRD4(608 -671), which encompasses the ET domain but not the bromodomains or C-terminal domain (Fig. 1E), can interact with full-length CHD4 in pulldown assays (46). The same study showed that a larger region of BRD3 (residues 570 -726) can likewise bind CHD4. Following these findings, we first mapped the BRD3interacting domain of CHD4 using pulldown assays. When fulllength BRD3 was coexpressed in HEK293 cells with the N-terminal, middle, or C-terminal sections of CHD4 ( Fig. 2A), only

The BRD3 ET domain binds chromatin remodeling complexes
the N-terminal third (residues 1-355) was able to robustly bind BRD3 (Fig. 2B). This portion of the protein contains an HMGbox-like domain (48) and is otherwise predicted to be disordered (by programs such as DISOPRED3 (49)). We next made a series of bacterially expressed GST-tagged CHD4 fragments focused on the N-terminal third of the protein ( Fig. 2A), and we used these as baits to pull down HA-tagged BRD3-L expressed in mammalian cells (Fig. 2C). These pulldowns indicated that CHD4-Ne (residues 265-310), a region predicted to be largely disordered, was both necessary and sufficient for interaction with BRD3.
Sequence similarity with the BRD4 ET domain, together with the published interaction data (46), suggested that BRD3-ET alone might also interact with CHD4. To ascertain whether BRD3-ET was sufficient for interaction with CHD4, we turned to NMR spectroscopy. 15 N HSQC spectra of BRD3-L and BRD3-ET are shown in Fig. 3A. The majority of peaks in the spectrum of BRD3-L (Fig. 3A, black) appear as intense, overlapping signals with amide proton shifts in the 8.0 -8.6 ppm region of the spectrum, a pattern characteristic of the disordered polypeptide. A number of weaker and more dispersed signals are also observed. In contrast, peaks in the 15 N HSQC of BRD3-ET (Fig. 3A, red) are well-dispersed, as expected of a small, folded protein domain. An overlay of the two 15 N HSQC spectra reveals that almost all signals in the BRD3-ET spectrum are essentially unchanged in the BRD3-L spectrum, indicating that the folded ET domain takes up the same conformation in the context of the longer polypeptide.

The BRD3 ET domain binds chromatin remodeling complexes
Titration of 15 N BRD3-L with CHD4-Ne does not appear to induce a significant disorder-to-order transition (Fig. 3B). Instead, only the signals attributable to BRD3-ET are affected. Furthermore, the pattern of intensity changes suggests the formation of a complex under an intermediate-slow exchange regime. To confirm that BRD3-ET is sufficient to mediate the interaction with CHD4, a titration was conducted with 15 Nlabeled BRD3-ET and unlabeled CHD4-Ne (Fig. 3, C and D). The CHD4-induced changes in peak positions were almost exactly reproduced in BRD3-ET (Fig. 3D), indicating that the ET domain is likely to be the minimal interaction domain for CHD4.

BRD3-ET interacts with a conserved linear motif found associated with many classes of chromatin-remodeling complexes
The 15 N HSQC spectrum of free CHD4-Ne was partially assigned using a combination of homonuclear 2D spectra ( 1 H, 1 H NOESY and TOCSY) and a 15 N-separated NOESY. Because of extensive overlap of lysine resonances, complete assignment of CHD4-Ne amide resonances was not possible; however, a number of peaks affected by titration with BRD3-ET were able to be identified (Figs. 4A and 5). Comparison of CHD4-Ne with a 23-residue MLV-IN peptide known to bind the BRD4 ET domain (50) revealed a highly related "PLKI(K/ R)L" sequence ( Fig. 4A) in the vicinity of the BRD3-affected peaks, suggesting that this might constitute the core ET-binding motif. We therefore made a synthetic peptide corresponding to residues 290 -301 of CHD4 (CHD4-Ng, Fig. 2A, Table 2), which contains the hypothesized ET-binding site. To verify that CHD4-Ng binds BRD3-ET in the same manner as the longer, recombinant CHD4-Ne peptide, we carried out a 15 N HSQC titration of BRD3-ET with CHD4-Ng (Fig. 4B). The chemical shift perturbations induced by CHD4-Ne and CHD4-Ng in BRD3-ET are very similar (Fig. S1), indicating that CHD4-Ng is sufficient to recapitulate the interaction of BRD3-ET with CHD4-Ne. A fit of the titration data to an intermediate chemical exchange model, using an approach previously described (51), yielded a K D estimate of ϳ2 M. We also used surface plasmon resonance to measure this affinity, which yielded a K D of 95 Ϯ 10 M (Fig. 5A, Table 2). The SPR and NMR were carried out at 4 and 25°C, respectively, which could be the source of this difference.
Because the PLKIKL sequence in both CHD4-Ng and MLV-IN was sufficient to bind BRD3-ET, we hypothesized that it might represent a common motif by which the ET domain interacts with partner proteins. Indeed, this motif is present in both NSD3 and GLTSCR1 (Fig. 4A), which along with CHD4 have been previously identified as potential ET domain interaction partners in a pulldown-MS screen (46). Interrogation of the human proteome using SCANSITE3 (scansite3.mit.edu (52)) revealed instances of this motif in a number of other putative BET interaction partners, including INO80B, a subunit of the INO80 complex involved in chromatin remodeling and DNA damage repair (53) and BRG1 (SMARCA4), a member of the SWI/SNF chromatin remodeling complex that has been previously found in complexes containing BRD2 (Fig. 4A) (54). For INO80B, BRG1, and NSD3, the motif was located in a region of the protein that is likely to be disordered based on disorder predictions.
To assess whether the mode of interaction is conserved between different ET-domain-binding partners, synthetic peptides derived from NSD3, BRG1, and INO80B were tested in 15 N HSQC titrations (Fig. 4B and Table 2); a titration was also carried out with an MLV-IN peptide for comparison. Fig. 4C shows that the chemical perturbation shift patterns induced in BRD3-ET by each peptide are very similar, indicating that the same binding pocket on BRD3-ET mediates all of these interactions. The interaction of BRD3-ET with NSD3 appeared to be the weakest, based on the larger excess of NSD3 peptide required to reach saturation, the magnitude of the chemical shift changes induced by NSD3, and the fact that it occurs in a faster chemical exchange regime. Surface plasmon resonance measurements (Fig. 6) yielded a K D of 7 Ϯ 1 M for the ET/BRG1 interaction, 80 Ϯ 10 M for ET/INO80B, and 950 Ϯ 100 M for ET/NSD3.

Three-dimensional structures of BRD3-CHD4 and BRD3-BRG1 complexes reveal a conserved interaction mode
We next determined the three-dimensional structure of the BRD3-CHD4 complex by NMR methods. Samples of the 1:1 complex were generated by titrating 15 N/ 13 C-labeled BRD3-ET to a slight excess of CHD4-Ng (ϳ1:1.1), and a range of standard

Components of NuRD, BAF, and INO80 complexes identified in the Brd3 pulldown experiment
The numbers of peptides identified for each protein in the Brd3-loaded beads and the beads alone are indicated. IP, immunoprecipitation.

No. of peptides identified
The BRD3 ET domain binds chromatin remodeling complexes 2D and 3D NMR spectra was recorded. A total of 1324 NOEs, which included 49 intermolecular NOEs, were obtained, and the structure is shown in Fig. 7, A-C. The geometry of the structure is good, with Ͼ99% of residues in the ordered region lying in allowed regions of the Ramachandran plot. The fold of BRD3-ET consists of three ␣-helices, designated ␣1 (residues 574 -586), ␣2 (residues 589 -602), and ␣3 (residues 623-637), consistent with the structure of the free BRD4 ET domain (1). The CHD4 peptide binds in a groove that is formed by residues from the ␣1 helix and the ␣2-␣3 loop of BRD3-ET (Fig. 7, B and C), resulting in a buried surface area of ϳ1150 Å 2 . A short antiparallel ␤-sheet is formed between CHD4 Ile-296 -Leu-298 and Ile-614 -Ile-616 in the ␣2-␣3 loop of BRD3, as inferred from CO-N distances. BRD3 residues with large CHD4-induced chemical shift perturbations (Fig. 4C) are shown mapped onto the structure of the complex in Fig. 7B. Hydrophobic residues (Leu-294, Ile-296, and Leu-298) alternate with charged residues (Lys-295 and Lys-297) in the CHD4 peptide, and this arrangement is mirrored by residues on the BRD3 ␣2-␣3 loop. Leu-294 interacts with the aromatic ring of Phe-618; Ile-296 contacts Ile-616; and Leu-298 packs between Ile-614 and Leu-592 (in the ␣2 helix). A network of electrostatic interactions is also apparent: the side-chain amine of Lys-297 lies adjacent to the carboxyl groups of Glu-613/Glu-615 and that of Lys-295 contacts Glu-615/Asp-617 ( Fig. 7C). Other residues that do not directly contact the peptide also display large chemical shift changes, which is likely due to the large movement in the ␣2-␣3 loop and helix ␣3 induced by peptide binding. These residues include Val-595 and Val-596 on the ␣2 helix, which pack against the binding-pocket residues Leu-592 and Ile-614, respectively, and residues Arg-607-Asn-610, which form a small 3 10 helix at the N terminus of the ␣2-␣3 loop that may contribute to stabilization of the loop conformation in the peptide-bound state.
To ascertain whether this interaction is important for mediating an interaction between full-length CHD4 and BRD3, we expressed human CHD4 (either WT or a mutant in which the Lys-297-Leu-298 sequence was mutated to AA) in HEK293 cells and applied cell lysate from these cells to GST-BRD3-ET bound to GSH-Sepharose beads. As shown in Fig. 8, binding of the mutant to BRD3-ET was significantly attenuated.
Finally, we used the same NMR-based approach to determine the three-dimensional structure of the BRD3-BRG1 complex. A total of 873 NOEs, including 71 intermolecular NOEs, were used to derive the structure shown in Fig. 7, D-F. Line broadening resulted in fewer NOEs than observed for the BRD3-CHD4 complex, and consequently a higher root mean square deviation for the overlay of the 20 conformers used to represent the structure (Table 3). Overall, however, the structure and binding mode are essentially the same as for the

The BRD3 ET domain binds chromatin remodeling complexes
BRD3-CHD4 complex, with the three hydrophobic residues Val-1595, Ile-1597, and Leu-1599 packing against BRD3-ET and the three alternating residues Lys-1594, Lys-1596, and Lys-1598 making contacts with the acidic surface of the binding pocket.

ET domains are highly conserved
During our work, structures of the ET domain of BRD4 bound to peptides from the MLV-IN (33), the transcriptional coregulator NSD3 (residues 152-163 and 593-605) (32), and the KSHV-LANA protein (32) were reported. The sequences of human BRD2-, BRD3-, and BRD4-ET are conserved (Ͼ85% identity and only one conservative substitution on the peptidebinding surface; Fig. S2A), and a comparison of the structures of BRD3-ET-CHD4, BRD3-ET-BRG1, BRD4-ET-NSD3, and BRD4-ET-LANA shows that the interaction mode is also conserved (Fig. 9). All of the interactions occur in the same peptide-binding pocket of the ET domain and involve the formation of an intermolecular two-stranded ␤-sheet that positions alternating hydrophobic and charged residues in the peptide to contact the ET domain. In contrast, the MLV-IN peptide is longer and folds back onto itself to form a third antiparallel ␤-strand in the complex with BRD4-ET. Interestingly, our data show that an MLV-IN peptide containing only the core motif can bind BRD3-ET, albeit with significantly lower affinity than that reported for the BRD4/MLV-IN interaction (160 nM). Furthermore, the similarity of chemical shift perturbation data for CHD4-Ne and the shorter CHD4-Ng confirm that, for CHD4, the shorter peptide contains the full ET-binding motif.
In addition to the high conservation of ET domains among human BET proteins, high similarity is observed throughout evolution. For example, only one conservative substitution distinguishes human BRD3-ET from the corresponding domain in Xenopus laevis. Furthermore, biochemical data (55) and sequence analysis indicate that the yeast BET protein Bdf1

The BRD3 ET domain binds chromatin remodeling complexes
interacts with the TAF7 component of the TFIID complex via a similar binding mechanism that involves Bdf1-ET and an LKLKN motif in TAF7 (Fig. 4A).
This high sequence similarity throughout evolution argues for a critical role for ET/partner interactions in BET protein function. In contrast, the high degree of identity between family members within a species suggests that a degree of functional overlap is likely to exist between BRD2, -3, and -4, given that they have an overlapping pattern of expression across mammalian tissues.
However, genetic evidence suggests that BET proteins perform both redundant as well as nonredundant roles (10,16,56); whether the ET domain contributes to discrimination between interaction partners has yet to be determined. This study, together with published data (33,57), indicates that different partners display a range of affinities for the BRD3 and BRD4 ET domains. In addition, there are residues in the ␣2-␣3 loop of the ET domain (namely Arg-607-Asn-610 in BRD3) that exhibit large chemical shift changes upon peptide binding but are not conserved between BET family members (Fig. S2A), and these might conceivably influence binding affinity and/or specificity. Outside of the bromodomains and ET domain, BET proteins display lower sequence conservation, and these regions might play a greater role in delineating paralogue-specific in vivo interactions of each family member.

Conservation of ET-binding motifs in transcriptional regulators
The ET-interacting peptide motif is relatively short and therefore is likely to occur by chance in many proteins, especially large proteins such as CHD4. Conservation of putative binding sites, both between paralogs and across species, might be of use in distinguishing those that are functionally relevant. For example, at least one additional potential binding site for the ET domain has been identified in both NSD3 (KKKIKK, residues 596 -601) and CHD4 (LKKVKL, residues 689 -694) (32). However, only the first ET-binding site (EIKLKI, residues 152-157) is fully conserved between NSD2 and NSD3, with partial conservation observed in NSD1 (Fig. S2E). Consistent with this observation, mutation of the first NSD3 site has been shown to abolish the NSD3/ET interaction in pulldown experiments, whereas mutation of the second site has little effect (32). Similarly, the first site in CHD4 (PLKIKL, residues 293-298) is conserved in human and mouse CHD3/4 (Fig. S2B), both of which can participate in formation of the NuRD complex. In contrast, the second site is not conserved between any of the nine members of the CHD family of proteins in humans. Therefore, it might be predicted that the ET domain interacts with mammalian CHD3/4 primarily through the first site. The Drosophila homolog of CHD4, dMi-2␤, also displays the first ETbinding site (Fig. S2B), which we predict will interact with the ET domain of the sole BET protein fs(1)h in Drosophila melanogaster. The ability of any KIKL sequence to interact with an ET domain will also depend on sequence context due to issues of folding and steric accessibility.
A counterpart for the BRG1 (SMARCA4) ET-binding site also exists in the BRG1 paralog BRM (SMARCA2) (Fig. S2C); both can serve as the ATPase subunit in the mammalian SWI/ SNF remodeling complexes, consistent with the hypothesis that the ET domain plays a central role in the recruitment of ATP-dependent chromatin-remodeling complexes to generegulatory elements. The INO80B site is conserved from human to frog (Fig. S2D).
Curiously, BRD4 itself also contains a potential ET-binding site within the disordered C-terminal domain (DLKIKN, residues 1193-1198). Intramolecular interaction between the second BRD4 bromodomain and an adjacent phosphorylated control region has previously been proposed to regulate its interaction with acetylated chromatin (58), and it is possible that additional autoregulation might arise through this ET/CTD interaction.

AF9/ENL leukemia-associated proteins display unexpected structural and functional commonality with BET proteins
A BLAST search of the human proteome using BRD3-ET as a query does not return any hits other than BRD4-ET. However,

The BRD3 ET domain binds chromatin remodeling complexes
a structure-based search using DALI (59) reveals that the ET domain fold closely resembles the structure of the ANC1 homology domain (AHD) of AF9 (Fig. 10A) (60); the two domains share 25% sequence identity (Fig. 10B, top), suggesting the idea that they are descended from a common ancestor. AF9 and its close paralog ENL (which also contains an AHD) are common translocation partners of MLL in human leukemia (61). Unexpectedly, the AF9-AHD is intrinsically disordered in isolation    (62) but takes up an ET-like fold upon interacting with peptides from AF4 (62), CBX8 (PDB code 2N4Q), 4 or DOT1L (60). AF4 and DOT1L (an H3K79 methyltransferase) are components of complexes associated with transcriptional elongation (the super-elongation (63) and Dot1 (47) complexes, respectively). In a remarkable recapitulation of the ET/peptide interaction mechanism, the AF9/partner interactions are mediated by the formation of a short antiparallel ␤-sheet between the ␣2-␣3 loop of AF9 and an amphipathic peptide from AF4/CBX8/ DOT1L (Fig. 10A). There is a clear similarity between the peptide-binding sites of AHD and ET domains (Fig. 10, A and C,  bottom), although the alternation between charged and hydrophobic residues (Fig. 10, B and C) is less distinct for the AHDbinding motifs than for the ET-binding sequences.

The BRD3 ET domain binds chromatin remodeling complexes
Similar to the BET proteins, AF9/ENL harbor an N-terminal domain that recognizes acetylated lysine side chains, in this case a YEATS domain (64,65), rather than the twin bromodomains of the BET family. This architectural similarity gives rise to clear functional commonalities between the two protein families. Both BRD4 and ENL can read the histone (or transcription factor) acetylation status of target genes and promote transcriptional elongation. Both are also essential for the proliferation of a subset of aggressive leukemias (24,66,67); inhibition of their activity impairs proliferation through selective repression of the transcriptional program initiated by the oncoprotein c-Myc (24,66,67).
At the biochemical level, the signal encoded in the acetylated lysine residue(s) is interpreted by the respective epigenetic reader domains (YEATS and bromodomain) and then translated into a downstream action by the ET/AHD domain recognizing a short, linear motif in specified partner proteins. It is at this point that the activities of the two classes of proteins appear to diverge, despite utilizing the same structural mechanism for signal transmission. Thus, targets of the AF9/ ENL AHD include the lysine methyltransferase DOT1L, which is the only enzyme that methylates histone H3K79. In contrast, targets of the ET domain are dominated by chromatin remodelers, including CHD4, CHD8 (indirectly through the interaction of ET with a short form of NSD3 (68)), BRG1, and the INO80 complex subunit INO80B. Thus the ET/peptide interactions represent a conserved regulatory mechanism across (at least) these two families of generegulatory proteins.
From an evolutionary perspective, a search of the protein sequence database using PHMMER (69) reveals that ET-related sequences are found very widely throughout Eukaryota, including in deep-branching species such as the plankton Emiliania huxleyi, the ciliate Tetrahymena thermophila, and the oomycete (water mold) Phytophthora infestans (where they are found alongside bromodomains). In contrast, AHD-related sequences appear to be restricted to Metazoa (animals, from Homo sapiens to the sponge Amphimedon queenslandica). The most parsimonious explanation for this observation is that the intrinsically disordered AHD domain has evolved from an ancestral ET domain. The selective advantage afforded by the intrinsically disordered nature of the AHD (if any) is currently not clear, but it might involve an ability to rapidly exchange binding partners (62), even highaffinity ones.  Bacterially-expressed GST-BRD3-ET immobilized on GSH beads was used to pull down mammalian-expressed full-length FLAG-CHD4 (WT or K297A/ L298A mutant). Western blots using anti-GST and anti-FLAG antibodies are shown. Root mean square deviation from lowest energy structure a All backbone atoms (N, C␣, CЈ) (Å) 0.

The BRD3 ET domain binds chromatin remodeling complexes Functional implications of ET interactions with transcriptional coregulators
The first BRD3 bromodomain has been shown to bind to acetylated GATA-1, and this interaction has been proposed to stabilize chromatin occupancy of GATA-1, possibly through the simultaneous binding of the second bromodomain to acetylated histones (15). The NuRD complex, of which CHD4 is a component, is also suggested to be recruited to promoters of GATA-1 target genes via Friend of GATA-1 (FOG-1), which interacts with the RBBP4/7 and MTA1/2 subunits of NuRD (42,70). The interaction between the BRD3 ET domain and CHD4 suggests that BRD3 may play an additional role in the recruitment or retention of the NuRD complex at GATA-1 sites. It is possible that both the BRD3/CHD4 and FOG-1/ RBBP4 interactions are required for the functional recruitment of NuRD, because RBBP4/7 can also participate in other regulatory complexes (71), and yet only the NuRD complex is recruited to GATA-1-dependent promoters. An analogous arrangement might also exist for BRD4, which is also able to bind the acetylated transcription factor RelA (7). The ubiquitous expression pattern of BRD2/3/4 suggests that recognition of acetylated transcription factors by BETprotein bromodomains, followed by ET-mediated interactions with KIKL motifs, might be a general mechanism through which chromatin remodeling complexes are recruited to promoters/enhancers.
Given that recruitment of transcriptional coregulators by the ET and AHD domains appears to play an important role in oncogenesis, targeting the ET domain might represent an alternative approach in the design of therapeutics for BET-dependent malignancies. The high degree of conservation in the ET domain might, like the bromodomains, present a challenge in terms of designing small molecules that selectively inhibit individual BET family members. Nevertheless, the unusual dependence of malignant cells on BET-occupied enhancers suggests that disrupting the ET-mediated interactions of BET proteins with transcriptional regulators that drive oncogenic expression programs might be a viable therapeutic strategy (72). In this regard, our findings provide fundamental insight into the mechanisms by which BET proteins link chromatin acetylation to transcriptional outcomes in normal physiology and in human disease. showing the ET/AHD as a surface and the partner peptide as cartoon/stick representation. B, top, amino acid sequences of the ET and AHD domains of human BRD3 and AF9, respectively. Identical residues are shown in orange and similar residues in yellow. The sequence that forms a ␤-sheet with the partner is underlined. Bottom, alignment of the ET-binding regions of CHD4 and BRG1 with the corresponding regions from the AHD-binding proteins CBX8, AF4, and BCoR. Sequences from the human proteins are shown in all cases. C, comparison of the interface residues of ET/AHD domains (left) and of the partner peptides (right); strong conservation is apparent in the former case, whereas the alternating pattern of hydrophobic and basic residues observed in ET-binding sequences is not as strongly conserved in AHD-binding sequences.

Constructs
All constructs described in this work are derived from human gene sequences unless otherwise specified. The full-length murine Brd3 coding sequence was cloned into the pEF1␣ vector with a single FLAG tag (DYKDDDDK) at the N terminus. Brd3 deletion mutants were amplified by PCR with a single FLAG tag at the N terminus and inserted into pEF1␣. BRD3-L (residues 420 -726) was cloned into pcDNA3.1 with a C-terminal hemagglutinin (HA) tag (YPYDVPDYA) and both BRD3-L and BRD3-ET (577-644) were cloned into pGEX-6P (for bacterial expression as GST-fusion proteins) using BamHI/EcoRI sites. CHD4 constructs were cloned either into pcDNA3.1 (with an N-terminal FLAG tag or HA tag) or into pGEX-6P. CHD4 KL-AA (K297A/L298A) was generated by Gibson assembly of synthetic gene fragments.

Protein overexpression in E. coli
Escherichia coli Rosetta (DE3) pLysS cells transformed with a CHD4 or BRD3 encoding plasmid were cultured at 37°C with shaking. Log-phase cultures (OD 0.4 -0.8) were induced by addition of isopropyl ␤-D-thiogalactopyranoside (IPTG, 0.2-0.4 mM), and cultured for a further 16 -24 h at 25°C. For preparation of isotopically-labeled protein, log-phase cells were washed in M9 minimal media salts and equilibrated for 1 h at 25°C in minimal media containing 15 NH 4 Cl and (in the case of 15 N/ 13 C-labeled samples) D-[ 13 C]glucose. The cultures were then induced with IPTG and protein expressed as described above. For production of 15 N/ 13 C-labeled BRD3-ET, cultures were grown completely in minimal media containing 15 NH 4 Cl and D-[ 13 C]glucose over 12-14 h at 37°C prior to induction.

Protein purification
Cell pellets containing overexpressed GST-BRD3-ET and GST-BRD3-L were lysed in lysis buffer (50 mM Tris, pH 7.3, 500 mM NaCl, 100 M MgCl 2 , 1 mM DTT, 0.5 M PMSF, 25 g ml Ϫ1 DNase I) and purified using GSH-affinity chromatography. The N-terminal GST tag was removed by incubation of GSH elution fractions with HRV-3C protease (produced inhouse) at 4°C. Recombinant protein was separated from GST and other contaminants using column chromatography. BRD3-L was purified by ion exchange on a UNO TM S1 column (Bio-Rad) in 50 mM Tris, pH 7.5, 50 -1000 mM NaCl, 1 mM DTT. BRD3-ET was purified using size-exclusion chromatography on a Superdex 75 preparative grade HiLoad 16/60 column (GE Healthcare) in 50 mM Tris, pH 7.3, 150 mM NaCl, 1 mM DTT. Separation of proteins was monitored using UV absorbance at 215 and/or 280 nm using a BioLogic QuadTec UVvisible detector (Bio-Rad) and SDS-PAGE.

Preparation of nuclear extracts
Cells were scraped from the plate in cold PBS. After centrifugation, the cell pellet was lysed in Buffer A (10 mM HEPES-KOH, pH 7.9, 1.5 mM MgCl 2 , 10 mM KCl, protease inhibitor mixture (Roche Applied Science), 1 mM PMSF, 1 mM DTT) on ice for 30 min. After vigorous vortexing, nuclei were pelleted by centrifugation and lysed in Buffer C (20 mM HEPES-KOH, pH 7.9, 25% glycerol, 420 mM NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, protease inhibitor mixture, 1 mM PMSF, 1 mM DTT) on ice for 30 min. Material was then spun at maximum speed in a tabletop centrifuge for 10 min at 4°C. The resulting supernatant was diluted in Buffer C lacking NaCl to reduce the salt concentration to 150 mM before being used for immunoprecipitations.

Immunoprecipitations
Immunoprecipitations were performed with anti-FLAG beads (Sigma). After extensive washing (150 mM NaCl, 50 mM Tris-HCl, pH 7.5, 0.5% Igepal), elution was carried out with 3ϫ FLAG peptide (Sigma), and bound proteins were run on SDS-PAGE and analyzed by Western blotting or MS. For MS, SDSpolyacrylamide gels were stained with Colloidal Blue and cut into four segments: Ͼ100, 95 to 50, 50 to 25, and Ͻ25 kDa. Mass spectrometry was performed at the Taplin Facility at Harvard University.

GST pulldown assays
Clarified cell lysates from 0.5 to 1 ml of E. coli Rosetta2 cells expressing GST-CHD4 proteins were incubated with GSH-Sepharose 4B beads for 30 min at 4°C. The beads were subsequently washed with 5 column volumes of lysis buffer, followed by equivalent washing into pulldown buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 0.5% Triton X-100, 1 mM DTT, 1ϫ cOmplete EDTA-free protease inhibitor mixture). 1 ml of HEK293FT cells expressing HA-BRD3-L was lysed by incubation in pulldown buffer for 30 min at 4°C. Clarified HEK293FT cell lysate was incubated with 20 l of GST-CHD4 beads for 1 h at 4°C. The beads were subsequently washed with 5 column volumes of pulldown buffer. Protein retained on the beads and 10% input of HEK293FT cell lysate were analyzed by SDS-PAGE followed by Western blotting.

Handling of synthetic peptides
Peptides derived from murine leukemia virus integrase (MLV-IN; residues 1191-1202) and from human CHD4 (resi-

Surface plasmon resonance
SPR was performed on a Biacore T200 with a buffer comprising 20 mM HEPES and 150 mM NaCl, pH 7.5, as both the immobilization and running buffer. The active and reference flow cells of a Biacore T200 CM5 series S sensor chip (GE Healthcare) were activated with 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride and N-hydroxysuccinimide at 25°C according to the manufacturer's instructions. BRD3-ET (100 g ml Ϫ1 ) in 10 mM sodium acetate, pH 4, was then injected for 15 min at 10 l min Ϫ1 , and the flow cells then blocked with an injection of 1 M ethanolamine. Following protein immobilization, the system was cooled to 4°C. Duplicate samples of BRG1, INO80B, NSD3, or CHD4 peptides (0.02-56 M in running buffer) were injected at a flow rate of 20 l min Ϫ1 over immobilized BRD3-ET. Equilibrium dissociation constants (K d ) were calculated by nonlinear least-squares fitting to a simple 1:1 Langmuir binding isotherm, as implemented in the Biacore T200 Evaluation software. Each affinity was measured at least twice on different days, and the quoted uncertainty is an estimate based on the spread of measured affinities as well as estimates of uncertainty in protein concentrations. 15 N/ 13 C-Labeled BRD3-ET samples (500 M) were prepared in either 20 mM sodium phosphate, pH 6.5, 50 mM NaCl, 1 mM DTT (CHD4), or 20 mM MES (pH 6.5), 50 mM NaCl, 1 mM tris(2-carboxyethyl)phosphine (TCEP) and then titrated with unlabeled CHD4 (1.1 molar equivalents) or BRG1 (2 molar equivalents) peptide. For acquisition of HCCH-TOCSY and 13 C NOESY-HSQC spectra, the protein-peptide complex samples were lyophilized and reconstituted in an equal volume of D 2 O. For all other NMR experiments, 5-10% (v/v) D 2 O was added directly to the samples. 2,2-Dimethyl-2-silapentane-5sulfonic acid was added as a chemical shift reference to all samples to a final concentration of 150 -300 M. Spectra were acquired at 298 K on Bruker Avance III 600 or 800 MHz NMR spectrometers, each fitted with a cryogenic TCI probe head and using standard pulse sequences from the Bruker library. The spectra were processed using TOPSPIN3 (Bruker, Karlsruhe, Germany), and NMRFAM-SPARKY (74,75) was used for spectral analysis.

Structure determination by NMR spectroscopy
Backbone assignments of peptide-bound BRD3-ET were made using HNCACB, CBCA(CO)NH, HNCA, HNCO, and HN(CA)CO spectra. Side-chain resonances were assigned using HBHA(CO)NH, H(C)(CO)NH-TOCSY, CC(CO)NH-TOCSY, and HCCH-TOCSY spectra, and aromatic spin systems were assigned using 2D 1 H, 1 H-NOESY spectra (150 ms mixing time). Resonances of the BRD3-bound peptide were assigned using 2D 15 N/ 13 C (F2,F1)-filtered 1 H, 1 H TOCSY and 15 N/ 13 C (F2,F1)-filtered 1 H, 1 H NOESY spectra (150 ms mixing time). Intramolecular protein NOEs were identified in 15 N-and 13 C-edited NOESY-HSQC spectra (150 ms mixing time), and intermolecular NOEs were identified in 15 N/ 13 C (F1)-filtered NOESY-HSQC spectra (100 ms mixing time). / torsion angle restraints were predicted from chemical shifts using TALOSϩ (76). CYANA 3.0 (77-79) was used for automated NOE assignment and initial structure calculations. 200 -1000 starting conformers were generated and subjected to simulated annealing with 20,000 torsion angle dynamics steps. Further molecular dynamics under the Crystallographic and NMR System (CNS) force field was performed on the lowest energy conformer from the CYANA calculation, using the final set of NOEs from the CYANA calculations, and the 100 (CHD4 complex) or 500 (BRG1 complex) lowest-energy structures from this process were refined with a 7-Å shell of water molecules according to the RECOORD protocol (80). Structural statistics of the ensemble of 20 lowest-energy structures at the end of water refinement were analyzed using the Protein Validation Software Suite web server (81). The structures of the BRD3-ET-CHD4 (PDB code 6BGG) and BRD3-ET-BRG1 (PDB code 6BGH) complexes have been deposited in the PDB. 15 N HSQC spectra of 200 M 15 N-labeled BRD3-ET, BRD3-L, and CHD4-Ne were acquired, and unlabeled peptide or protein was titrated into the sample in increments of 0.25 M eq (up to 2 M eq); 15 N HSQC spectra were acquired after each addition. The 15 N HSQC spectrum of BRD3-ET was assigned at the completion of each titration from HNCACB and CBCA-(CO)NH spectra. The 15 N HSQC spectrum of CHD4-Ne (residues 265-310) was partially assigned using 2D 1 H, 1 H NOESY/ TOCSY spectra and a 15  where ⌬␦HN is the chemical shift change in the proton dimension, and ⌬␦N is the chemical shift change in the nitrogen dimension.