Abstract Recognition of methylarginine marks by effector proteins (“readers”) is a critical link between arginine methylation and various cellular processes. Recently, we identified methylation of AKT1 at arginine-391 (R391), but the reader for this methylation has yet to be characterized. Here, we show that bromodomain-containing protein 9 (BRD9), a reader of acetylated lysine, unexpectedly recognizes methylated R391 of AKT1 through an aromatic cage in its bromodomain. Disrupting the methylarginine reader function of BRD9 suppresses AKT activation and tumorigenesis. RNA sequencing data show that BRD9 and AKT coregulate a hallmark transcriptional program in part through enhancer of zeste homolog 2 (EZH2)–mediated methylation of histone-3 lysine-27. We also find that inhibitors of BRD9 and EZH2 display synergistic effects on suppression of cell proliferation and tumor growth. Collectively, our study reveals a previously unknown function of BRD9 and a potential therapeutic strategy for cancer treatment by combining BRD9 and EZH2 inhibitors. __________________________________________________________________ BRD9 activates the AKT-EZH2 pathway to facilitate tumorigenesis through its noncanonical role as a methylarginine reader. INTRODUCTION Posttranslational modifications (PTMs) refer to addition of chemical groups or polypeptides to amino acid residues of proteins ([34]1). PTMs are dynamically regulated by a plethora of players that can be classified as writers, erasers, and readers. Writers and erasers are enzymes that add and remove various PTMs, respectively. Readers typically contain a specialized domain that recognizes respective modifications and interprets the PTM language to cellular processes ([35]2). Arginine methylation has been documented as a ubiquitous PTM and plays important roles in many fundamental cellular processes, including transcription, RNA splicing, DNA repair, and signaling transduction ([36]3–[37]6). It involves the addition of a methyl group to the guanidino nitrogen atoms of arginine residue by the writers, protein arginine methyltransferases (PRMTs), resulting in three types: monomethylarginine (MMA), asymmetric dimethylarginine (ADMA), and symmetric dimethylarginine (SDMA). Nine members of the PRMT family in mammal can be classified into three categories: Type I enzymes (PRMT1, 2, 3, 4, 6, and 8) catalyze the formation of MMA and ADMA; type II enzymes (PRMT5 and PRMT9) catalyze the formation of MMA and SDMA; and type III enzyme (PRMT7) only catalyzes the formation of MMA ([38]7). PRMTs have been reported to overexpress in several types of cancers and play important role in cancer progression, thereby they are considered as potential therapeutic targets ([39]8, [40]9). Notably, recent clinical trial studies of methylthioadenosine-cooperative PRMT5 inhibitors showed promising antitumor effect in patients with methylthioadenosine phosphorylase–deleted cancer ([41]10, [42]11). However, PRMT5 is essential for normal cell survival and development ([43]12) and genetic alterations in PRMT5 are rare in most cancer types ([44]http://cbioportal.org/). Further research is necessary to clarify the regulation and exact role of PRMT5 in various cancers. Emerged evidence suggests that recognition of methylarginine (Rme) marks by effector proteins (known as readers) is a key mechanism for transducing arginine methylation signals to various cellular processes ([45]13). Like many other PTMs, arginine methylation can generate docking sites for its binding modules. Compared to multiple reading domains of lysine methylation (Kme), the Tudor domain is the only well-known module that binds to Rme marks ([46]14). Tudor domains typically contain about 60 amino acids, in which the conserved aromatic residues structurally form an “aromatic cage” to accommodate Rme. Among more than 30 Tudor domain-containing proteins in humans, only 4 of them have been well characterized as the Rme readers. Survival motor neuron (SMN) and survival motor neuron domain containing 1 (SMNDC1) recognize both SDMA and ADMA marks with lower affinity for ADMA ([47]15, [48]16), whereas staphylococcal nuclease domain-containing protein 1 (SND1) and Tudor domain containing protein 3 (TDRD3) prefer to bind SDMA and ADMA motifs, respectively ([49]17–[50]19). Existing evidence also suggests that other five Tudor domain containing proteins, including TDRD1 ([51]20), TDRD2 ([52]17), TDRD6 ([53]21), TDRD7 ([54]22), and TDRD8 ([55]23), are putative arginine methylation readers. In addition to these Tudor domain containing readers, proteins or complexes without Tudor domain are also capable of recognizing Rme marks, including breast cancer type 1 susceptibility protein (BRCA1) ([56]24), forkhead box O3A (FOXO3a) ([57]25), tripartite motif-containing protein 29 (TRIM29) ([58]26), SWI/SNF related BAF chromatin remodeling complex subunit ATPase 4 (SMARCA4) ([59]27), and the polymerase-associated factor 1 (PAF1) complexes ([60]28). Notably, a recent study identified squamous cell carcinoma antigen recognized by T cells 3 (SART3) as a reader of SDMA through its half-atetratricopeptide (HAT) repeat domain ([61]29). These studies suggest that recognition of Rme marks may depend on the methylation sites and its surrounding sequence. Thus, numerous Rme readers likely exist to interpret the prevalent Rme events, which remain largely unknown. The serine/threonine kinase AKT is a critical node that links growth factors, cytokines, and other cellular stimuli to cell proliferation and survival. Hyperactivation of the AKT pathway has been observed in various cancers including breast cancer ([62]30, [63]31). We and another group recently demonstrated that PRMT5-mediaed symmetric dimethylation of AKT is critical for its activation and oncogenic function ([64]32, [65]33). Moreover, we identified R391 in AKT1 as the main methylarginine by using an antibody that specifically recognizes symmetric dimethylation of R391 (termed AKT1-R391me2s hereafter) ([66]33). However, the readers and downstream signaling pathway of this methylation event remain to be identified. Here, through an approach of unbiased peptide-based pull-down coupled with mass spectrometry analyses, in vitro and cell-based assays, and orthotopic breast cancer xenograft mouse studies, we define that BRD9 functions as a reader of AKT1-R391me2s to promote AKT activation and tumor growth, which is in part by directly linking AKT signaling to EZH2-mediated trimethylation of histone-3 lysine-27 (H3K27me3) for transcriptional regulation. Moreover, we demonstrate that BRD9 and EZH2 inhibitors synergistically suppress breast cancer cell survival and tumor growth. RESULTS BRD9 is capable of binding with AKT1-R391me2s Recently, we reported that PRMT5-mediated AKT1-R391me2s leads to a disruption of the intramolecular interaction between pleckstrin homology domain (PH) and kinase domain (KD), as well as R391 interaction with surrounding residues, including E319, A329, and K386 ([67]33). Thus, we hypothesized that these conformational changes might provide an open space to accommodate an effector protein. To test this hypothesis, we sought to identify putative readers for AKT1-R391me2s using peptide-based interaction proteomics ([68]34). Biotinylated AKT1 peptides with or without symmetric dimethylation of R391 (termed R391me2s and R391me0, respectively) were incubated with cell lysates to pull down interaction proteins, followed by liquid chromatography–tandem mass spectrometry (LC-MS/MS)–based label-free proteomics analysis ([69]Fig. 1A). After filtering to remove reversed database hits, common cell culture contaminants, proteins with less than two peptides, and common protein contaminants associated with streptavidin enrichment (histones, elongation factors, and carboxylases) ([70]35), 21 proteins were only detected in the R391me2s enrichment including BRD9 ([71]Fig. 1B and data S1). We selected BRD9 for further study because it has been reported to positively regulate AKT activation, but the mechanism remains unknown ([72]36). Fig. 1. BRD9 recognizes and binds to methylated AKT1-R391. [73]Fig. 1. [74]Open in a new tab (A) Schematic of peptide pull-down and mass spectrometry experiment using biotin-labeled AKT1-R391 peptides with or without symmetrical dimethylarginine (R391me2s and R391me0). (B) A scatterplot comparing the log[2] iBAQ protein intensities between the R391me2s and R391me0. BRD9 was marked as red. n = 1 biological replicate. (C) Immunoblot (IB) analysis of pull-down products by biotin-labeled AKT1 peptides and myc-BRD9. (D) Schematic showing the domains of BRD9. (E and F) IB analysis of WCL and immunoprecipitates (IPs) derived from HEK293T cells transfected with the indicated constructs. EV, empty vector. (G) IB analysis of pull-down products derived from biotin-labeled peptides and BRD9-BD or BRD4-BD1 purified from bacteria. (H) Zoomed-out (left) and zoomed-in (right) images depicting the proposed interaction of R391-me2s (blue) from AKT (orange) with BRD9-BD (gray) based on MD simulations. Hydrogen bonds (yellow) and key residues on BRD9 (purple) were shown. (I) IB analysis of WCL and hemagglutinin (HA)–IP derived from HEK293T cells transfected with Myc-BRD9-WT or mutants and HA-AKT1 constructs. To validate BRD9 as a reader of AKT1-R391me2s, we first confirmed that only the AKT1-R391me2s peptide can pull down the full-length BRD9 ([75]Fig. 1C). The interaction between endogenous BRD9 and AKT1 was further validated by coimmunoprecipitation experiments (fig. S1, A and B). AKT1 consists of an N-terminal PH domain, a KD, and a C-terminal regulatory region (Tail) ([76]37). As expected, BRD9 mainly bound to the KD that harbors R391 (fig. S1, C and D). Furthermore, inhibition of R391 methylation by either a PRMT5 inhibitor or mutating R391 to lysine (R391K), largely blocked AKT1 interaction with BRD9 (fig. S1, E and F). These results demonstrate that BRD9 binds to AKT1 in an R391me2s-dependent manner. We next investigated the molecular mechanism by which BRD9 recognizes AKT1-R391me2s. BRD9 contains a bromodomain (BD) at the N terminus and a DUF3512 domain at the C terminus. The BD functions as a reader of acetylated lysine residues (Kac) ([77]38), whereas the DUF3512 domain serves as a scaffold for assembly of the noncanonical BAF complex (ncBAF) ([78]39–[79]41). To define the role of these two domains, we generated truncated mutants expressing either the N-terminal BD or C-terminal DUF3512 ([80]Fig. 1D). The BD is sufficient and necessary for BRD9 interaction with the KD of AKT1 in cells ([81]Fig. 1, E and F). Moreover, AKT1 did not interact with SMARCA2/4 (two subunits of the ncBAF complex) and depletion of SMARCA2/4 by proteolysis-targeting chimera (PROTAC) degrader AU-15330 has minimal effects on BRD9 interaction with AKT1 (fig. S1G). Furthermore, we purified recombinant glutathione S-transferase (GST)–tagged BRD9-BD protein from bacteria and performed in vitro peptide pull-down assays. The results showed that only the R391me2s peptide can pull down BRD9-BD ([82]Fig. 1G). Of note, BRD4-BD1 was not pulled down by the AKT1 peptides ([83]Fig. 1G), although it is structurally similar to BRD9-BD. We also found that compared to BRD9-BD, the Tudor domain of SMN and SND1 displayed much weaker or no binding to R391me2s (fig. S1H). In addition, the BRD9-BD exhibited a greater affinity for R391me2s than H4R3me2s (fig. S1I). These results suggest that BRD9 functions as a major reader of AKT1-R391me2s through its BD domain and independent of the ncBAF complex. To pinpoint the residues that are involved in recognizing AKT1-R391me2s, we performed rigid peptide docking on the crystal structure of BRD9-BD [Protein Data Bank (PDB): 6V1B]. A pocket consisting of three residues—Y173, I176, and Y215—was identified as a possible cage for accommodation of AKT1-R391me2s (fig. S1J), which is different from the Kac binding pocket consisting of G159, F160, F161, N216, and Y222 ([84]42). When the simulations were run with the AKT1-R391me2s peptide as a ligand, R391me2s is recognized by Y215 and held into place by hydrogen bonding to both Y173 and Y215 ([85]Fig. 1H). Although I176 is positioned away from the predicted binding pocket with AKT, one thing that was noted from both the crystal structure and our molecular dynamics (MD) simulations experiments was that this binding region is extremely dynamic and does not adopt a set secondary structure. I176 [root mean square fluctuation (RMSF = 6.68 Å)] and Y173 (RMSF = 5.49 Å) are among the most dynamic side chains in BRD9, whereas Y215 also has an RMSF = 3.18 Å (fig. S1K). It is likely that this unstructured and extremely dynamic region confers BRD9 with the ability to scan its local environment and capture Rme marks, such as AKT1-R391me2s. Given that no data-driven structural models are available at this time to assess the AlphaFold model with biochemical data, it warrants a further structural study to investigate how BRD9-BD can accommodate Rme. Consistent with the structural simulations data, mutating either of these residues, to a less extent for I176, reduced the interaction between BRD9 and AKT1, whereas mutating all these three residues (3Mut) completely blocked their interactions ([86]Fig. 1I). We further performed peptide pull-down assays and found that BRD9-3Mut is incapable of binding to the AKT1-R391me2s peptide (fig. S1L). Of note, BRD9-BD-WT and BRD9-BD-3Mut displayed very similar thermal stability and secondary structures with consistent overlap (fig. S1, M and N), demonstrating that the 3Mut does not disrupt the folding and structure of the BRD9-BD domain. These results suggest that the three residues form an “aromatic cage”–like pocket for binding of R391me2s. Notably, mutating the gatekeeper residue Y222 that is required for Kac binding had minimal impact on BRD9 interaction with AKT1 (fig. S1O). However, I-BRD9, a selective BRD9 inhibitor that blocks the Kac binding pocket ([87]43), markedly impaired BRD9 and AKT1 interaction in cells (fig. S1P). Consistently, I-BRD9 blocked the AKT1-R391me2s peptide binding of BRD9-BD in vitro (fig. S1Q). Mechanically, the structure of BRD9-BD with I-BRD9 showed that I-BRD9 not only forms a hydrogen bond to a water molecule bound by Y173 but also causes the conformation change of surrounding residues ([88]43). We also found that acetylated histone H4 peptide displayed greater capacity than AKT1-R391me2s peptide to compete for binding of BRD9-BD (fig. S1, R to T), suggesting that BRD9-BD has a stronger affinity to Kac than Rme and cannot simultaneously bind both PTMs. This is likely because binding to one PTM may cause conformation change or spatial hindrance for BRD9-BD access to another PTM, which warrants further structural study. Together, these results demonstrate that, in addition to its well-known function as a Kac reader, BRD9 also serves as an Rme reader through a distinct pocket in its BD to bind AKT1-R391me2s and likely other methylated proteins. BRD9 promotes AKT activation in an R391me2s-dependent manner Having demonstrated BRD9 as a reader of AKT1-R391me2s, we next investigated whether BRD9 plays a role in R391me2s-mediated AKT1 activation. Notably, depletion of BRD9 by CRISPR-Cas9 attenuated phosphorylation of AKT and its downstream substrate, glycogen synthase kinase-3β (GSK-3β), in various breast cancer cell lines, including MDA-MB-231 ([89]Fig. 2A), BT-549 (fig. S2A), MCF-7 (fig. S2B), and T-47D (fig. S2C). Consistently, both short- and long-term treatments of these cells with dBRD9, a selective BRD9 degrader ([90]44), efficiently reduced BRD9 expression and AKT activation ([91]Fig. 2B and fig. S2, D to G). We also performed in vitro kinase assays and found that AKT1 immunopurified from BRD9-depleted cells largely lost its ability to phosphorylate GSK-3β (fig. S2H). Moreover, abrogation of BRD9 expression also suppressed AKT signaling in response to insulin and EGF stimulation ([92]Fig. 2C and fig. S2, I and J). To determine BRD9-mediated activation of AKT1 through recognition of R391me2s, we ectopically expressed BRD9 wild-type (WT) or BRD9-3Mut in cells with depletion of endogenous BRD9. Notably, the AKT signaling was restored by BRD9-WT, but not the BRD9-3Mut that is unable to bind R391me2s ([93]Fig. 2D). Moreover, treatment of cells with I-BRD9, which blocks BRD9 access to R391me2s (fig. S1, P and Q), also diminished AKT activation in a time-dependent manner ([94]Fig. 2E). In contrast, overexpression of BRD9 further enhanced the phosphorylation of AKT1-WT but not the methylation-deficient mutant, AKT1-R391K ([95]Fig. 2F). Fig. 2. The methylarginine reader function of BRD9 is required for AKT activation. [96]Fig. 2. [97]Open in a new tab (A) IB analysis of WCL derived from MDA-MB-231 cells depleted of BRD9 by two independent sgRNAs. sgGFP is a negative control. (B) IB analysis of WCL derived for MDA-MB-231 cells treated with 100 nM dBRD9 for the indicated time. (C) IB analysis of WCL derived from control (sgGFP) or BRD9-depleted MDA-MB-231 cells. Cells were serum starved for 16 hours before being treated with 100 nM insulin for 1 hour. (D) IB analysis of WCL derived from BRD9-depleted MDA-MB-231 cells stably expressing HA-BRD9-WT or HA-BRD9-3Mut. (E) IB analysis of WCL derived for MDA-MB-231 cells treated with 5 μM I-BRD9 for the indicated time. (F) IB analysis of WCL and HA-IPs derived from HEK293T cells transfected with indicated constructs. (G) IB analysis of WCL and GST pull-downs derived from HEK293T cells transfected with indicated constructs. (H) Schematic model depicting that BRD9 recognizes AKT1-R391me2s to impair the interaction between PH and KD domains, leading to AKT membrane translocation and activation. PRMT5-mediated AKT1-R391me2s can disrupt the intramolecular interaction between the PH and KD domains and thereby release AKT1 autoinhibition ([98]33). To investigate whether binding of BRD9 to R391me2s is involved in this process, we treated cells with dBRD9 or I-BRD9 and found that both treatments enhance the interaction between the PH and KD domains (fig. S3, A and B). Consequently, depletion of BRD9 reduced AKT1 membrane recruitment (fig. S3, C and D). In contrast, overexpression of BRD9-WT, but not the BRD9-3Mut, weakened PH and KD interaction ([99]Fig. 2G). We also found that, although BRD9 is predominantly localized in the nucleus, a certain amount of BRD9 is also localized to the cytoplasm (fig. S3E), suggesting that BRD9 may bind AKT1-R391me2s and acetylated histones in the cytoplasm and nucleus, respectively. Together, these results demonstrate that, upon methylation at R391 by PRMT5, AKT1 conformation shifts from the closed state to open state. Subsequently, BRD9 recognizes R391me2s and facilitates or maintains AKT1 in the open state, thereby promoting AKT1 membrane translocation for activation ([100]Fig. 2H). BRD9 facilitates breast tumor growth through its methylarginine reader function Recent studies have shown that BRD9 plays an important role in various cancers, including acute myeloid leukemia ([101]45), malignant rhabdoid tumors ([102]40), prostate cancer ([103]46), and synovial sarcoma ([104]47). Our previous study demonstrated that PRMT5-mediated AKT1-R391me2s is critical for AKT activation and breast cancer growth ([105]33). Having addressed BRD9 as an R391me2s effector to promote AKT1 activation, we next investigated its biological function in breast cancer. Upon analyzing the expression data of breast cancer in The Cancer Genome Atlas (TCGA), we found that, compared to the normal breast tissues, the mRNA levels of BRD9 are significantly increased in breast tumor tissues, with the highest expression levels in triple-negative breast cancer (TNBC) ([106]Fig. 3A). Moreover, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis also revealed that the phosphatidylinositol 3-kinase (PI3K)–AKT pathway is enriched in breast cancer patients with high BRD9 expression ([107]Fig. 3, B and C), which is consistent with our finding that overexpression of BRD9 promotes AKT activation in cells ([108]Fig. 2F). Fig. 3. BRD9 deficiency suppresses breast tumor growth in part through AKT. [109]Fig. 3. [110]Open in a new tab (A) Analysis of BRD9 mRNA levels in breast tumors and normal breast tissues in the TCGA dataset. **P < 0.01, ***P < 0.001, one-way ANOVA. (B) KEGG pathway enrichment analysis of DEGs between BRD9-high and BRD9-low breast cancer samples in the TCGA dataset. (C) Gene Set Enrichment Analysis (GSEA) of the PI3K-AKT signaling pathway in BRD9-high and BRD9-low breast cancer samples in the TCGA dataset. NES, normalized enrichment score. (D) Control (sgGFP) or BRD9-depleted MDA-MB-231 cells were subjected to xenograft mouse assays. Tumor size was measured every other day. Data are shown as means ± SEM of n = 8 tumors for each group. ***P < 0.001, two-way ANOVA. (E) Dissected tumors were weighed. Data are shown as the means ± SEM of n = 8 tumors for each group. *P < 0.05, Student’s t test. (F) IB analysis of lysates derived from tumor tissues in (E). (G) MDA-MB-231 cells expressing BRD9-WT or BRD9-3Mut were subjected to xenograft mouse assays. Tumor size was measured every 3 days. Data are shown as means ± SEM of n = 8 tumors for each group. ***P < 0.001, two-way ANOVA. (H) Dissected tumors were weighed. Data are shown as the means ± SEM of n = 8 tumors for each group. **P < 0.01, Student’s t test. (I) IB analysis of lysates derived from tumor tissues in (H). To address the role of BRD9 in breast cancer, we depleted BRD9 in two TNBC cell lines, MDA-MB-231 and BT-549. Proliferating and colony-forming capability of these BRD9-depleted cells was significantly decreased compared to the control cells (fig. S4, A to D). We next investigated the role of BRD9 in tumor growth using a xenograft mouse model by injecting BRD9-depleted MDA-MB-231 cells (sgGFP cells as a control) into the mammary fat pad. Notably, BRD9 depletion significantly suppressed tumor growth and tumor mass ([111]Fig. 3, D and E, and fig. S4E). Immunoblot analysis showed that BRD9-depleted tumors exhibited much lower AKT signaling than the control tumors ([112]Fig. 3F), suggesting that BRD9 promotes tumor growth in part by activating the AKT pathway. To assess the contribution of BRD9’s Rme reader function to breast cancer, we reconstituted BRD9-WT or BRD9-3Mut in MDA-MB-231 cells with depletion of endogenous BRD9. Notably, cells expressing BRD9-3Mut displayed much slower cell proliferation and less colony formation than cells expressing BRD9-WT (fig. S4, F to H). Xenograft mouse experiments showed that the BRD9-3Mut markedly suppresses tumor growth and AKT activation ([113]Fig. 3, G to I, and fig. S4I), which phenocopies BRD9 depletion. Together, these results demonstrate that the Rme reader function of BRD9 contributes to its tumor-promoting role in breast cancer in part through recognition of R391me2s to activate the AKT pathway. BRD9 and AKT coregulate a hallmark transcriptional program We next sought to identify the downstream signaling of the BRD9/AKT axis. BRD9 is known as an important transcriptional modulator through several mechanisms, including chromatin remodeling ([114]39–[115]41), recognition or influence of histone acetylation ([116]42, [117]48), and interaction with transcription factors and regulators ([118]46, [119]49). Likewise, AKT plays a critical role in modulating gene expression by directly or indirectly regulating transcription factors and epigenetic modulators ([120]31, [121]50). To identify the transcriptional program controlled by the BD of BRD9 and AKT, we performed RNA sequencing (RNA-seq) experiments in MDA-MB-231 cells treated with either I-BRD9 or ipatasertib, a selective and adenosine triphosphate–competitive pan-AKT inhibitor ([122]51). Analysis of differentially expressed genes (DEGs) showed that I-BRD9 and ipatasertib treatment groups share 116 down-regulated genes and 70 up-regulated genes, which account for 59 and 32% of AKT-down-regulated and up-regulated genes, respectively ([123]Fig. 4, A and B). These 186 shared genes were then subjected to MSigDB hallmark pathway enrichment analysis ([124]52). Nine hallmark gene sets were significantly enriched, including genes involved in epithelial-mesenchymal transition (EMT), hypoxia, cholesterol homeostasis, apoptosis, and angiogenesis, as well as genes regulated by transcription factors, including MYC, nuclear factor κB (NF-κB), and signal transducer and activator of transcription 5 (STAT5) ([125]Fig. 4C). We also analyzed the association between these shared genes and epigenetic modifications (epi-marks) using epidecodeR, which can perform statistical testing on log[2]FC (fold change) of DEGs to group genes as “0” (no epi-marks in these genes) and “1+” (one or more epi-marks in these genes) ([126]53). Seventy-five of the 116 common down-regulated genes were significantly marked by H3K27me3 ([127]Fig. 4D). Moreover, ~40% of these H3K27me3-associated genes were also regulated by PRMT5 (fig. S5, A and B). These results suggest that BRD9 and AKT coregulate a subset of genes likely through both transcription factors and H3K27me3 modification. Fig. 4. BRD9 and AKT share common transcriptional targets in breast cancer cells. [128]Fig. 4. [129]Open in a new tab (A) Venn diagrams depicting common significant down-regulated and up-regulated genes (Q value < 0.05) between I-BRD9 and ipatasertib (AKTi). MDA-MB-231cells were treated with 5 μM I-BRD9 or 1 μM ipatasertib for 16 hours prior to determination of gene expression by RNA-seq. DMSO treatment as a negative control. (B) Heatmap showing mRNA expression changes of commonly regulated genes described in (A). (C) Identification of hallmark signaling pathways shared by I-BRD9 and ipatasertib. (D) Boxplots of common down-regulated genes based on the number of H3K27me3 mark analyzed using epidecodeR software. “0” represents no H3K27me3 mark, and “1+” represents ≥1 H3K27me3 mark. (E) RT-qPCR analysis of mRNA levels of select genes commonly regulated by I-BRD9 and ipatasertib. MDA-MB-231 cells were treated with 5 μM I-BRD9 or 1 μM ipatasertib for 16 hours. Data are shown as means ± SD of n = 3 biological replicates. **P < 0.01 and ***P < 0.001, one-way ANOVA and Tukey post hoc test. (F) RT-qPCR analysis of mRNA levels of select common genes in MDA-MB-231 cells expressing BRD9-WT or BRD9-3Mut. Data are shown as means ± SD of n = 3 biological replicates. *P < 0.05 and **P < 0.01, Student’s t test. We next performed reverse transcription–quantitative polymerase chain reaction (RT-qPCR) and confirmed that the shared genes in these hallmark pathways were transcriptionally repressed in cells treated with I-BRD9 or ipatasertib ([130]Fig. 4E), which was further validated in cells genetically depleted of BRD9 or AKT1 (fig. S5, C and D). The mRNA levels of these genes were also significantly decreased in cells expressing BRD9-3Mut, compared to cells expressing BRD9-WT ([131]Fig. 4F). These results suggest that the Rme reader function of BRD9 plays a critical role in regulating AKT-dependent transcription program, which may contribute to the tumor promoting function of BRD9 and AKT. BRD9 and AKT regulate transcription in part through EZH2-mediated H3K27me3 Previous studies demonstrated that AKT-mediated phosphorylation of EZH2 suppresses its binding to histone H3, resulting in a decrease in H3K27me3 and subsequent activation of gene transcription ([132]54, [133]55). Because BRD9 regulates AKT activation and shares common genes with AKT, we speculated that BRD9 may regulate AKT/EZH2-mediated H3K27me3 to control gene expression. To address this hypothesis, we genetically depleted BRD9 by CRISPR-Cas9 or chemically degraded BRD9 by dBRD9 in MDA-MB-231 and MDA-MB-436 cells. Notably, loss of BRD9 led to a notable elevation of H3K27me3 in both cell lines ([134]Fig. 5, A and B, and fig. S6, A and B), which could be completely blocked by GSK126, a highly selective and SAM-competitive inhibitor of EZH2 ([135]Fig. 5C) ([136]56). Moreover, treatment of cells with I-BRD9 also markedly enhanced H3K27me3 signal in cells ([137]Fig. 5D and fig. S6C), which was suppressed by GSK126 treatment (fig. S6D). In contrast, overexpression of BRD9-WT, but not the BRD9-3Mut, decreased H3K27me3 levels (fig. S6E). These results demonstrate that BRD9 negatively regulates H3K27me3 in an EZH2-dependent manner. Fig. 5. BRD9 regulates H3K27me3 levels through the AKT1-EZH2 axis. [138]Fig. 5. [139]Open in a new tab (A) IB analysis of WCL derived from MDA-MB-231 cells infected with lentivirus of sgGFP or sgBRD9. (B) IB analysis of WCL derived from MDA-MB-231 cells treated with indicated doses of dBRD9 for 16 hours. (C) IB analysis of WCL derived from MDA-MB-231 cells infected with lentivirus of sgGFP or sgBRD9. Cells were treated with 5 μM GSK126 for 16 hours before harvesting. (D) IB analysis of WCL derived from MDA-MB-231 cells treated with indicated doses of I-BRD9 for 16 hours. (E) IB analysis of WCL derived from MDA-MB-231 cells stably expressing EV or Myr-AKT1 and depleted of BRD9 infected with lentivirus sgGFP (−) or sgBRD9 (+). (F) IB analysis of WCL derived from control (sgGFP) or AKT1-depleted MDA-MB-231 cells. Cells were treated with 5 μM I-BRD9 for 16 hours before harvesting. (G) RT-qPCR analysis of mRNA levels of select genes in MDA-MB-231 cells treated with DMSO or 5 μM GSK126 for 48 hours. Data are shown as means ± SD of n = 3 biological replicates. *P < 0.05, **P < 0.01, and ***P < 0.001, Student’s t test. To investigate whether BRD9 regulates H3K27me3 through the AKT-EZH2 axis, we first confirmed that H3K27me3 is elevated in cells treated with PI3K inhibitors, LY294002 and BKM120, or the AKT inhibitor ipatasertib (fig. S6, F to H), which is consistent with a previous report ([140]55). Similar results were obtained upon genetic depletion of AKT1 (fig. S6I). Overexpression of the constitutively active form of AKT1 (myr-AKT1) largely suppressed H3K27me3 induction caused by BRD9 depletion or I-BRD9 treatment ([141]Fig. 5E and fig. S6J). To further support this finding, we showed that PTEN depletion, which results in hyperactivation of the PI3K-AKT pathway ([142]57), also markedly inhibited H3K27me3 in BRD9-deficient cells (fig. S6, K and L). However, combination of I-BRD9 treatment and AKT1 depletion did not further induce H3K27me3 ([143]Fig. 5F), suggesting that AKT and BRD9 are in the same pathway to regulate H3K27me3. Consistent with the data that the ncBAF complex is not required for BDR9 interaction with AKT1 (fig. S1G), depletion of SMACARC2/4 by AU-15330 has minor effects on AKT phosphorylation and H3K27me3 (fig. S6M). We also observed that I-BRD9 largely blocked AKT-mediated phosphorylation of EZH2, which was detected by a phospho-AKT substrate antibody (fig. S6N). Last, inhibition of EZH2 by GSK126 increased the transcription of common genes of BRD9 and AKT ([144]Fig. 5G). These results reveal a mechanism that BRD9, specific to its BD here, can regulate transcription through the AKT-EZH2-H3K27me3 pathway, in addition to its functions as a Kac reader and a subunit of the ncBAF complex. BRD9 inhibitor synergizes with EZH2 inhibitor in breast cancer Following the milestone discoveries of selective inhibitors for BD and extraterminal BDs ([145]58, [146]59), targeting epigenetic readers for cancer therapy has gained extensive attention and achieved tremendous progress ([147]60). As emerging evidence supports a critical role of BRD9 in various cancers, it has become an attractive therapeutic target and multiple selective BRD9 inhibitors or degraders have been developed ([148]61). However, the antitumor effect of BRD9 inhibitors in breast cancer, particularly TNBC, has not been evaluated. Likewise, EZH2 has emerged as a promising target with numerous inhibitors undergoing clinical trials as monotherapy or combination with other drugs ([149]62). Given the interplay between BRD9 and EZH2, we next evaluated the effect of their inhibitors as a monoreagent or in combination in breast cancer. A synergistic antiviability effect of I-BRD9 and GSK126 was observed in several breast cancer cell lines, including MDA-MB-231 ([150]Fig. 6A), MCF-7 (fig. S7A), and MDA-MB-436 (fig. S7B). Moreover, combined treatment resulted in more potent antiproliferative effects and much less colony formation compared to I-BRD9 or GSK126 alone ([151]Fig. 6, B to D, and fig. S7, C and D). We further investigated the antitumor effects of I-BRD9 and GSK126 using an orthotopic xenograft mouse model ([152]Fig. 6E). Compared to the control (vehicle), all treatments significantly suppressed tumor growth, but the combination group displayed the most notable antitumor effects ([153]Fig. 6, F to H). In line with these observations, combined treatment markedly induced cell apoptosis compared to the monoreagent treatment ([154]Fig. 6I). Thus, cotargeting BRD9 and EZH2 can be an efficient strategy for better treatment of breast cancer. Fig. 6. Combined treatment of BRD9 and EZH2 inhibitors leads to synergistic growth inhibition of breast cancer. [155]Fig. 6. [156]Open in a new tab (A) Inhibition of cell viability and dose-response matrixes analyzed by SynergyFinder. MDA-MB-231 cells were treated with the indicated doses of I-BRD9 and GSK126 for 96 hours prior to analysis of cell viability. (B) MDA-MB-231 cells treated with I-BRD9 or GSK126 were subjected to cell proliferation assays. Data are shown as means ± SD of n = 3 biological replicates. ***P < 0.001, two-way ANOVA and Tukey post hoc test. (C) MDA-MB-231 cells treated with I-BRD9 or GSK126 were subjected to colony formation assays. Representative images are shown. (D) Quantification of colonies in (C). Data are shown as means ± SD of n = 3 biological replicates. *P < 0.05 and ***P < 0.001, one-way ANOVA and Tukey post hoc test. (E) Schematic of a mouse xenograft assay to evaluate the antitumor effects of I-BRD9 and GSK126. (F) Tumor growth curve upon treatment of I-BRD9 and GSK126. Data are shown as means ± SEM of n = 6 mice for each group. *P < 0.05, two-way ANOVA and Tukey post hoc test. (G and H) Dissected tumors were weighed. Data are shown as the means ± SEM of n = 6 tumors for each group. *P < 0.05, one-way ANOVA and Tukey post hoc test. (I) Representative images of TUNEL assays in xenograft tumors in (G). (J) Schematic depicting the function of the BRD9-AKT-EZH2 axis in regulating transcription and tumor growth. DISCUSSION BDs are well known by their functions as epigenetic “readers” of Kac on histones. According to the distinctive architecture in structure, BDs found in 46 different proteins are classified into eight families ([157]38). BRD9 belongs to the BD family IV and can bind histone peptides with Kac and butyryllysine (Kbu) modifications ([158]38, [159]42). Here, we showed that BRD9 recognizes Rme marks, at least the AKT1-R391me2s and H4R3me2s, through its BD. We further defined that a noncanonical “aromatic cage” in the BD consisting of Y173, I176, and Y215 is responsible for Rme recognition, which is structurally adjacent to but different from its Kac binding pocket. Therefore, this finding reveals an unexpected Rme reader function of BRD9. Recent studies have also identified BRD9 as a crucial subunit of distinct SWI/SNF chromatin remodeling complexes called GBAF or ncBAF through its DUF3512 domain, which cooperates with its BD to recruit chromatin modifiers and transcription factors to control unique transcriptional programs involved in cell proliferation, metastasis, apoptosis, and differentiation ([160]39–[161]41, [162]47). Depletion or inhibition of BRD9 suppressed cell proliferation and xenograft tumor growth of prostate cancer ([163]46), SMARCB1-deficient pediatric malignant rhabdoid tumors ([164]40), SS18-SSX fusion-dependent synovial sarcoma ([165]47), and acute myeloid leukemia ([166]45). These studies suggest that the role of BRD9 in tumor progression is likely cell type and context dependent. A recent study showed that BRD9 plays an important role in PI3KCA/KRAS-driven oncogenic transformation in breast epithelial cells ([167]63). However, the role and mechanism of BRD9 in breast cancer progression remain unknown. Our results demonstrate that depletion of BRD9 suppresses breast cancer cell proliferation and xenograft tumor growth. Mechanically, BRD9 promotes tumor growth through the AKT-EZH2-H3K27me3 axis that is one of the most critical pathways in breast cancer ([168]64, [169]65). The TCGA data showed that the expression of BRD9 is elevated in tumors compared to corresponding normal tissues. Moreover, high mRNA levels of BRD9 are associated with shorter survival time in patients of liver cancer, clear cell renal cell cancer, and sarcoma ([170]66). Consistently, our analysis showed that BRD9 expression is elevated in breast cancer, particularly TNBC. Therefore, BRD9 has been considered as a tumor promoter and a potential therapeutic target. Two strategies have been developed to disable the function of BRD9, including the BD inhibitors ([171]43, [172]67) and PROTAC degraders ([173]44, [174]68). Both treatments have been shown to significantly suppress cell proliferation and survival in multiple cancer cell culture models, with acute myeloid leukemia cell lines displaying the most sensitive phenotypes ([175]45, [176]46, [177]69). Emerged studies have also suggested that combination of BRD9 inhibitors with other antitumor agents is a more effective strategy ([178]70, [179]71). Our study revealed a synergistic antitumor effect of BRD9 and EZH2 inhibitors in breast cancer in part through AKT. In an agreement with our finding, a recent study showed that AKT and EZH2 inhibitors synergically suppress TNBC tumor growth ([180]72). EZH2 is aberrantly overexpressed in more than 50% of invasive breast cancer and its expression levels inversely correlate with poor prognosis ([181]65). Extensive evidence has demonstrated that EZH2 is a key driver of breast cancer initiation and metastasis ([182]73). Although its canonical function is to catalyze H3K27me3 for silencing tumor suppressor gene ([183]74), EZH2 also has noncanonical functions that activate transcription through directly interacting with transcription factors or methylating nonhistone substrates ([184]75). Therefore, EZH2 can serve as both transcriptional repressor and activator to promote cancer progression. AKT-mediated phosphorylation of EZH2 at Ser^21 (EZH2-pS21) serves as a switch controlling its canonical and noncanonical functions. The phosphorylated form of EZH2 impairs its binding to histone H3 and consequently decreases H3K27me3 to re-express tumor suppressor genes in breast cancer cells ([185]55). In contrast, EZH2-pS21 is required for EZH2-mediated methylation and activation of oncogenic transcription factors, including STAT3 in glioblastoma ([186]76) and androgen receptor in prostate cancer ([187]77). Our RNA-seq data indicate that BRD9 may regulate EZH2-pS21 to control the activity of transcription factors, including MYC, NF-κB, and STAT5, which warrants further investigations. Collectively, our study demonstrates that BRD9 acts as a methylarginine reader to recognize AKT1-R391me2s and activate AKT, which, in turn, phosphorylates EZH2 to dictate H3K27me3-dependent and H3K27me3-independent transcriptional programs, thereby promoting breast cancer. Thus, a combination of BRD9 and EZH2 inhibitors will inactivate both H3K27me3 and oncogenic transcription factors to achieve better antitumor activity ([188]Fig. 6J). MATERIALS AND METHODS Cell culture, transfection, and lentiviral infection Human embryonic kidney (HEK) 293T, MCF-7, MDA-MB-231, and their derived cell lines were cultured in Dulbecco’s modified Eagle’s medium (DMEM). BT-549, T-47D, MDA-MB-436, and their derived cell lines were maintained in RPMI 1640 medium. Both the DMEM and RPMI 1640 medium were supplemented with 10% fetal bovine serum, penicillin (100 U/ml), and streptomycin (100 μg/ml). Cells were maintained at 37°C with 5% CO[2]. Cell transfection was performed using Lipofectamine and Plus reagents following the manufacturer’s instructions. Lentivirus package and infection were described previously ([189]33). Briefly, targeted construct was cotransfected with pMD2.G and psPAX2 plasmids into HEK293T cells using polyethylenimine as a transfection reagent. Forty-eight hours posttranscription, supernatants containing virus were collected and filtered with 0.45 μM PES filters. Cells infected with the targeted virus were cultured in media supplemented with hygromycin (200 μg/ml), puromycin (2 μg/ml), or blasticidin (10 μg/ml) for 3 to 5 days to eliminate the noninfected cells. Reagents Insulin (41400045), EGF ([190]PHG03111), and subcellular protein fractionation kit (78840) were obtained from Thermo Fisher Scientific. I-BRD9 (HY-18975), GSK126 (HY-13470), ipatasertib (HY-15186), LY294002 (HY-10108), BKM120 (HY-70063), and AU-15330 (HY-145388) were purchased from MedChemExpress. dBRD9-A (6943) was purchased from TOCRIS. Plasmids Myc-BRD9, Myc-BRD9-N, Myc-BRD9-C, and Myc-BRD9-ΔBD were generated by cloning the corresponding cDNA into the pRK5-Myc vector. Myc-BRD9-Y173A, Myc-BRD9-I176R, Myc-BRD9-Y215A, Myc-BRD9-3Mut, and Myc-BRD9-Y222I were generated using the QuikChange XL site-directed mutagenesis kit from Agilent (#20518). pLJM1-hygro-HA-BRD9 and pLJM-hygro-HA-BRD9-3Mut were generated by cloning the corresponding cDNA into the pLJM1-HA vector. pGEX-GSK-3β, pGEX-BRD9-BD, and pGEX-BRD4-BD1 were generated by inserting the cDNA into pGEX-6P-1 vector. HA-AKT1, HA-AKT1-R391K, HA-AKT1-KD, CMV-GST-AKT1-PH, CMV-GST-AKT1-KD, and CMV-GST-AKT1-Tail were generated previously ([191]33). Various single guide RNAs (sgRNAs) were designed at [192]https://synthego.com and were cloned into the lentiCRISPR v2 vector (Addgene, 52961). Myr-AKT1 (64606) and HA-EZH2 (173717) were purchased from Addgene. GST-SMN-Tudor and GST-SND1-Tudor constructs were obtained from the laboratory of M. Bedford at the MD Anderson Cancer Center. Antibodies All primary antibodies were diluted in TBST buffer with 5% nonfat milk at 1:1000 to 1:10,000 for Western blotting. Anti-AKT-pT308 antibody (13038), anti-AKT-pS473 antibody (4060), anti-AKT pan antibody (4685), anti-pGSK-3β antibody (5558), anti-GSK-3β antibody (12456), anti-AKT1 antibody (2938), anti-PRMT5 antibody (79998), anti-phospho-AKT substrate antibody (9614), anti-E-cadherin antibody (3195), anti-Myc-tag rabbit antibody (2278), anti-biotin antibody (7075), anti-SMARCA2 antibody (11966), anti-GST-Tag rabbit antibody (2625), and anti-HA-Tag rabbit antibody (3724) were purchased from Cell Signaling Technology. Anti-BRD9 rabbit antibody (A303-781A) was purchased from Bethyl Laboratories. Anti-histone H3 mouse antibody (sc-517576), anti-SMARCA4 antibody (sc-17796), and anti-PTEN mouse antibody (sc-7974) were purchased from Santa Cruz Biotechnology. Anti-TriMethyl-Histone H3-K27 rabbit antibody (A2363) was purchased from ABclonal. Anti-Tubulin antibody (66240-1-lg) was purchased from Proteintech. Monoclonal anti-HA Tag antibody (901503) and anti-c-Myc mouse antibody (626802) were purchased from BioLegend. Anti-rabbit immunoglobulin G (IgG)–peroxidase secondary antibody (A4914) and anti-mouse IgG-peroxidase secondary antibody were purchased from Sigma-Aldrich. Immunoblot and immunoprecipitation analysis Cells were washed with ice-cold phosphate-buffered saline (PBS) and lysed with Triton lysis buffer [40 mM Hepes (pH 7.4), 5.5 mM MgCl[2], 150 mM NaCl, 1 mM EDTA, and 1% Triton X-100 supplemented with protein and phosphorylation inhibitors] or EBC buffer [50 mM Tris (pH 7.5), 120 mM NaCl, and 0.5% NP-40 supplemented with protein and phosphorylation inhibitors]. For histone extraction, cells were lysed with Triton lysis buffer and then sonicated 4x for 5 s at 50% power. After incubating for 15 min at 4°C, the samples were centrifuged at 13,200 rpm for 10 min at 4°C. Protein concentrations were measured by a Bio-Rad protein assay reagent. Equal amounts of whole-cell lysate (WCL) were resolved by SDS–polyacrylamide gel electrophoresis (PAGE), transferred onto a polyvinylidene difluoride membrane, and immunoblotted with antibodies. For immunoprecipitation, 1000 to 3000 μg of WCL was incubated with primary antibodies conjugated with agarose for 3 to 4 hours at 4°C. The beads were then washed five times with NETN buffer [150 mM NaCl, 20 mM Tris (pH 8.0), 1 mM EDTA, and 0.5% NP-40] before being subjected to SDS-PAGE for further analysis. Purification of GST-tagged protein from E. coli Recombinant GST-GSK-3β, GST-BRD9-BD, GST-BRD9-BD-3Mut, GST-SMN-Tudor, and GST-SND1-Tudor proteins were purified from BL21(DE3) Escherichia coli (E. coli) cells transformed with respective plasmids. Specifically, a single colony was grown in LB medium until reaching an optical density of 0.4 to 0.6 and then induced with 0.1 mM IPTG (isopropyl-β-d-thiogalactopyranoside) at 25°C for 16 hours. The E. coli cells were collected and resuspended in GST buffer [25 mM Tris (pH 8.0), 5 mM dithiothreitol (DTT), and 150 mM NaCl] before being sonicated (12 cycles of 5 s each at 50% power). After centrifugation, the supernatant was incubated with Glutathione Sepharose beads (Cytiva, 17075605) for 3 hours at 4°C. The protein-bound beads were washed three times with GST buffer. GST-tagged proteins were eluted with elution buffer [10 mM l-glutathione and 50 mM tris-HCl (pH 8.0)]. Biotinylated peptide pull-down assay AKT1 peptides [Biotin-Ahx-GLLKKDPKQRLGGGSEDA (R391me0), Biotin-Ahx-GLLKKDPKQ(Rme2s)LGGGSEDA (R391me2s), and unconjugated KDPKQ(R-me2s)LGGG (me2s)] were synthesized at ABclonal. Biotin-H4R3me2s (AS-65424), Biotin-H4ac (AS-65248), and unconjugated H4ac (AS-65241) peptides were purchased from ANASPEC Inc. (AS-65424). Peptides (12.5 μg) and 10 μl of Dynabeads Myone Streptavidin C1 (Invitrogen, 65-001) were incubated in 300 μl of binding buffer [50 mM tris-HCl (pH 7.5), 150 mM NaCl, 1 mM EDTA, 2 mM DTT, and 0.5% NP-40] overnight at 4°C. Five micrograms of recombinant GST-BRD9-BD or GST-BRD9-BD-3Mut proteins was added to the reaction for 3 hours at 4°C. The beads were washed five times with binding buffer before being resolved by SDS-PAGE. Mass spectrometry identification of AKT1-R391me2s interacting proteins Twenty-five micrograms of biotin-labeled R391me0 or R391me2s peptides and 20 μl of Streptavidin C1 beads were incubated in 300 μl of binding buffer (see Biotinylated peptide pull-down assay) overnight at 4°C. The beads were then added to WCLs derived from MDA-MB-231 cells and incubated overnight at 4°C. The samples were washed five times with binding buffer and then washed twice with 50 mM ammonium bicarbonate. Proteins were reduced with 1 mM DTT for 30 min followed by alkylation with 5.5 mM iodoacetamide for 15 min in the dark. Proteins were digested directly off the beads with a trypsin-Lys C solution overnight at 37°C with shaking at 300 rpm. Resulting peptides were desalted using 0.6 μl of C18 ZipTips and dried under vacuum. Peptides were analyzed by LC-MS/MS with an EASY nLC 1200 in-line with an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific). Peptides were pressure loaded onto C18 reversed-phase column [Acclaim PepMap RSLC, 75 μm by 50 cm (C18, 2 μm, 100 Å); Thermo Fisher Scientific, 164536] and separated using a gradient of 5 to 35% B in 180 min (Solvent A: 5% acetonitrile/0.1% formic acid; Solvent B: 80% acetonitrile/0.1% formic acid) at a flow rate of 300 nl/min. Mass spectra were acquired in data-dependent mode with high-resolution (60,000) FTMS survey scans, mass range of mass/charge ratio (m/z) 375 to 1575, followed by MS/MS of the most intense precursors with a cycle time of 3 s. The AGC target value was 4.0 × 10^5 for the survey scan. Peptides were fragmented by HCD with a precursor isolation window of 1.6 m/z, maximum injection time of 50 ms, and collision energy of 35%. Monoisotopic-precursor selection was set to “peptide.” Precursors within 10 ppm (parts per million) mass tolerance were dynamically excluded from resequencing for 25 s. Advanced peak determination was not enabled. Precursor ions with charge states 2 to 7 were included. Data were searched against a human protein database (UniProt, 02 February 2024, 20,433 entries) as well as a reversed, decoy database and common cell culture contaminants using MaxQuant v1.6.14 (Max Planck Institute). Search parameters permitted two missed cleavages, a minimum peptide length of seven amino acids, fixed cysteine carbamidomethylation, and variable protein N-terminal acetylation and methionine oxidation. For identification, a false discovery rate of <0.01 was required at the PSM, peptide, and protein levels. Data were processed using Perseus v1.6.15.0 (Max Planck Institute). Proteins were filtered to remove matches to the reversed database, common cell culture contaminants, proteins identified with less than two peptides, and contaminants associated with streptavidin beads (histones, elongation factors, and carboxylases) ([193]35). iBAQ protein intensities were log[2] transformed, and missing values were imputed with a constant value of 13 to visualize the data. A scatterplot comparing the log[2] iBAQ protein intensities between the R391me0 and R391me2s was generated. Initial rigid docking procedures Preliminary docking efforts of the BRD9-BD-AKT interaction were performed using AlphaFold2 ([194]78). The docking peptide of AKT was joined to the C terminus of the BD of BRD9 (PDB: 6V1B) by a 30-glycine-residue linker. The polypeptide was used as an input for AlphaFold2 as recently published ([195]79). After removing the polyglycine linker, the initial rigid-docking model was formed. Differential scanning fluorimetry experiments Differential scanning fluorimetry experiments were performed according to the protocol described previously ([196]80). Briefly, GST-BRD9-BD-WT and GST-BRD9-BD-3Mut recombinant proteins (final concentration of 4 μM) were mixed with SYPRO Orange dye at 0.5X (Thermo Fisher Scientific, S6650; the stock concentration of 5000X). Twenty microliters of the protein/dye mixture was distributed into 384-well plates. SYPRO Orange dye without protein was used as a negative control. The assays were performed using the Bio-Rad CFX384 Touch Real-Time PCR Detection System with a temperature range of 25° to 95°C and record of fluorescence every 1°C. The data were plotted as relative fluorescence to the max peak of each sample. Melting temperatures were calculated using DSFworld ([197]https://gestwickilab.shinyapps.io/dsfworld/) and presented as the means ± SD of three independent experiments. MD simulations Unbiased MD simulations were performed using Desmond with the Schrodinger Computational Suite. For studies with the WT protein, BRD9 (PDB: 6V1B) was imported into the Maestro GUI. I-BRD9 was deleted from the structure, and the apo protein was prepared via the Protein Preparation Wizard using default conditions. Using Desmond’s System Builder, the protein was explicitly solvated in the TIP3P solvation model containing 150 mM NaCl using the OPLS_2005 force field, and the charges were neutralized by adding Cl^− when necessary. An 8 Å–by–8 Å–by–8 Å box was used to generate the system, and the resulting system was minimized to a protein RMSD of <0.3 Å. This minimized system was used as the starting pose for MD simulations. Four independent runs of 1 μs were completed. Simulations were performed in the NPγT ensemble using the Langevin barostat (1.01325 bar) and thermostat (300 K). The system was relaxed before simulation and gradually brought to temperature with decreasing constraints as per the default series of Desmond simulations. All simulations began from a random starting seed and velocities were randomized. Frames were recorded at an interval of 50 ps. The MD simulations with the triple mutant BRD9 were prepared and run in an identical manner, beginning with the BRD9 triple mutant protein that was generated by making the desired mutant of WT BRD4 in the Schrodinger Computational Suite. Simulations with the AKT peptide began from the pose generated using AlphaFold2 as described above. Each of the runs for both BRD9 variants were independently clustered using the “Trajectory Clustering Tool” within the Maestro GUI. The poses of the most populated cluster from the runs were then superimposed to obtain the consensus structure from the MD simulations. RMSF values of residue side chains were calculated using the “Simulations Interactions Diagram” wizard within Maestro. RNA-seq and RT-qPCR Total RNAs were prepared from MDA-MB-231 cells treated with dimethyl sulfoxide (DMSO), I-BRD9, or ipatasertib using the RNeasy Plus Mini Kit (Qiagen, 74134). Each was in triplicate. RNA-seq was performed at the Beijing Genomics Institute (BGI). RNA quality was assessed using an Agilent 4150 analyzer. mRNA was purified using poly-T oligo-attached magnetic beads. The purified mRNA was fragmented and then used for synthesis of the first-strand cDNA using random hexamer primers, followed by second-strand cDNA synthesis using dTTP (3′-deoxythymidine 5′-triphosphate) for library construction. Qualified libraries were then pooled and sequenced on the DNBSEQ platform. Raw reads were filtered by software SOAPnuke to obtain clean reads, followed by mapping to the human genome with HISAT2 and Bowtie2. Gene expression level is estimated using the FPKM value. DEGs were detected with DEseq2. The Venn diagram and pathway analysis were performed in the Dr. Tom system provided by the BGI. For RT-qPCR, the iScript Reverse Transcription Supermix (Bio-Rad, 1708841) was used to synthesize the first-stand cDNA. qPCR assays were carried out using EvaGreen qPCR Master Mix (Biotium, 31041) with normalization to glyceraldehyde-3-phosphate dehydrogenase (GAPDH). PCR primers are listed in table S1. TUNEL assay Terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick end labeling (TUNEL) assay was conducted using the CF488A-TUNEL Apoptosis Assay Kit (Biotium, 30063). Paraffin-embedded tumor sections were deparaffinized and rehydrated through a series of graded ethanol solutions after xylene treatment. Subsequently, the sections were washed with PBS and permeabilized with proteinase K (20 mg/ml) for 30 min at 37°C. Next, the sections were incubated with 100 μl of TUNEL Equilibration Buffer for 5 min, followed by incubation with 50 μl of TUNEL reaction mix for 2 hours at 37°C. Last, the sections were mounted using Vibrance Antifade mounting medium (Vector Laboratories, H-1700). Colony formation assay A total of 200 to 300 cells per well were seeded in 6-well plates and cultured for 8 to 12 days until visible colonies formed. The colonies were then fixed with fixative buffer containing 10% acetic acid and 10% methanol for 20 min, followed by staining with 0.4% crystal violet dissolved in 20% ethanol for an additional 20 min. After staining, the colonies were washed with distilled water and air-dried. Last, the colonies were scanned and manually counted. Xenograft mouse assays MDA-MB-231 cells or MDA-MB-231 cells expressing sgGFP, sgBRD9, BRD9-WT, or BRD9-3Mut were injected into the mammary fat pad of 8-week-old female nude mice. Five days postinjection, the tumors were measured every other day using an electronic caliper. For inhibitor treatment experiments, when the average tumor volume reached ≥100 mm^3, mice were randomly assigned into four groups and treated daily with vehicle, I-BRD9 (30 mg/kg), GSK126 (50 mg/kg), or combined I-BRD9 and GSK126. Again, the tumors were measured every other day using an electronic caliper. Tumor volume was calculated by the formula: L x W^2 x 0.5 (where L is the longest diameter and W is the shortest diameter). Mice were euthanized at the endpoint, and solid tumors were dissected and weighed. All mice are housed at 22°C with 50 to 60% humidity under a 12-hour light/12-hour dark cycle. All animal experiments were conducted following the protocol IACUC-2018-00604-1 approved by the MUSC Institutional Animal Care and Use Committee. Drug combination assay and synergy analysis MDA-MB-231, MCF-7, and MDA-MB-436 cells were treated with serially diluted I-BRD9 and GSK126. Cell viability was measured after treatment for 4 days using the CellTiter-Glo Cell Viability kit according to the manufacturer’s instructions (Promega, G7572). Synergistic scores were calculated using SynergyFinder ([198]https://synergyfinder.fimm.fi/). A score greater than 10 was considered as synergy. Statistical analysis Most cell-based experiments were repeated at least three times. As indicated in figure legends, all quantitative data are presented as the means ± SD or means ± SEM. Statistical significance was determined using a Student’s t test or analysis of variance (ANOVA). A P value of <0.05 was considered significant. Acknowledgments