Integration of the Transcriptome and Genome-Wide Landscape of BRD2 and BRD4 Binding Motifs Identifies Key Superenhancer Genes and Reveals the Mechanism of Bet Inhibitor Action in Rheumatoid Arthritis Synovial Fibroblasts

This article is featured in Top Reads, p.243

Rheumatoid arthritis (RA) is a systemic autoimmune disorder characterized by activation of adaptive as well as innate immune systems, infiltration of immune cells in the joints, and hyperproliferation of fibroblasts-like synoviocytes (FLS), leading to joint inflammation and destruction of bone and cartilage. In RA, the healthy two to three layered synovium is transformed into a multilayered pannus-like structure with a dramatically increased number of FLS and macrophages that secrete a multitude of proinflammatory cytokines and matrix metalloproteases (MMPs), leading to RA pathogenesis (1, 2). Because activated T/B cells, plasmablasts/plasma cells, and myeloid cells infiltrate RA synovium, therapeutics targeting these cell types (CTLA4-Ig and anti-CD20 Abs) and their effector molecules (anti-TNF and anti–IL-6R Abs) have been approved for the treatment of RA (3). Despite the adoption of these biologics in the armamentarium of RA therapeutics, deeper response and treatment-free remission in RA has remained elusive. Although most patients on biologic disease-modifying antirheumatic drugs show some improvement, only a small number achieve low disease activity (4, 5). Thus, there is an unmet medical need for better treatments that either alone or in combination with existing therapeutic regimens result in a deeper response and long-lasting disease remission in a greater number of patients. Because RA-FLS display a unique aggressive behavior and epigenetic imprinting may be the underlying basis of their activated phenotype (1), targeting epigenetic pathways may provide a novel approach for the treatment of RA.

The bromodomain and extraterminal (BET) family consists of three somatically expressed bromodomain (BRD) proteins, namely BRD2, BRD3, and BRD4, that bind to chromatin and regulate gene transcription. BET proteins, which are chromatin readers with binding specificity to the lysine residues present on histone tails, activate gene expression by facilitating the recruitment of transcriptional factors and regulatory complexes to promoter and enhancer regions of responsive genes (6–9). To understand the role of BET proteins in biological systems, small molecule inhibitors (JQ1 and I-BET) that competitively bind to BRDs of these proteins and block their interaction with the acetylated histone tails have been developed (10, 11). BET proteins are detected in RA synovial tissue, and their selective small molecule inhibitors have been shown to decrease the expression of proinflammatory cytokines and MMPs in vitro in RA synovial fibroblasts (12–14). In addition, JQ1 showed efficacy in a collagen-induced arthritis (CIA) murine model in vivo (13, 14). Because epigenetic pathways, in particular histone acetylation, are intricately involved in inflammatory and proliferative gene regulations (15) and BET inhibitors show promising activity in vitro and in vivo in RA models, we used the active enantiomer of JQ1 for understanding the biology of BRD proteins in RA-FLS. JQ1-treated RA-FLS were subjected to transcriptomic and genomic profiling to identify JQ1-regulated genes as well as BRD2 and BRD4 cistromes that were impacted by the BET protein inhibitor. In addition, the effect of JQ1 on open and closed regions of chromatin was examined by assay for transposable accessible chromatin by high throughput sequencing (ATAC-Seq). We demonstrate that JQ1 specifically inhibited IL-1β–induced activation and proliferation of RA-FLS in vitro. In addition, integration of transcriptional profiling and BRD2/BRD4/Pol2 cistrome datasets obtained from JQ1-treated IL-1β–activated RA-FLS identified key superenhancer genes and JQ1-responsive pathways that were enriched in synovial tissue samples from established RA or various stages of RA disease progression. Moreover, JQ1 treatment of RA-FLS resulted in an alteration of chromatin conformation, leading to reduced binding of proinflammatory transcription factors. Our data demonstrate that the BET inhibitor functions by downregulating multiple superenhancer genes and inflammatory pathways that are either targets of the current generation of therapeutics or that have been validated in preclinical models.

Human FLS isolated from established RA patients (Articular Engineering, Northbrook, IL) were cultured in Connaught Medical Research Laboratories media supplemented with 20% FBS, 1× l-glutamine, and 1% penicillin/streptomycin (Life Technologies, Carlsbad, CA). For assay set up, cells were washed, trypsinized, centrifuged, washed to remove trypsin, resuspended in Connaught Medical Research Laboratories media, and seeded in 96-well plates at 5000 cells/well. After overnight incubation, BET inhibitor JQ1⁺ or its inactive control enantiomer JQ1⁻ (Tocris Bioscience, Bristol, U.K.) were added to wells at various concentrations from 0.001 to 1 μM. After 2-h incubation, cells were stimulated with IL-1β (1 ng/ml; R&D Systems, Minneapolis, MN) and continued incubation for another 24 h. Supernatants were harvested for Luminex analysis (MilliporeSigma, Burlington, MA) to evaluate cytokine/chemokine and MMP secretion as per manufacturer’s instructions. Cell viability was measured by MTT assay (MilliporeSigma) as per manufacturer’s protocol. No more than passage 5 cells were used for these experiments.

RA-FLS were seeded at 8000 cells/well in a 96-well plate. After 24 h, JQ1⁺ or control JQ1⁻ were added at concentrations ranging from 0.001 to 10 μM generated with a 3-fold dilution of the compounds. At the same time, cells were also stimulated with 1 ng/ml IL-1β (R&D Systems). After overnight incubation, BrdU was added for 4 h, and proliferation was quantitated by BrdU ELISA as per manufacturer’s protocol (Roche Diagnostics, Basel, Switzerland).

RA-FLS (n = 2 donors) were seeded at 6 × 10⁶/T225 flasks and treated with JQ1 (1 μM) or DMSO and stimulated with IL-1β at 1 ng/ml. After 24-h incubation, cells were fixed with paraformaldehyde for 15 min with agitation at room temperature, and the cross-linking reaction was terminated by incubation of cells with glycine for 5 min. Cells were rinsed twice with ice-cold PBS, scraped from flasks with a cell scraper, and centrifuged for 5 min, 4°C, at 1000 × g. Cell pellets were resuspended in ice-cold wash buffer PBS-Igepal and centrifuged again to pellet cells. Wash step was repeated one more time with 1 mM PMSF in the wash buffer. Cells were centrifuged and pellets snap frozen in dry ice and stored at −80°C. Cell pellets were sent to Active Motif (Carlsbad, CA) for chromatin immunoprecipitation sequencing (ChIP-Seq) analysis. Chromatin immunoprecipitation and Ab pulldown with Abs against BRD2, BRD4, and Pol2, followed by sequencing, was performed at Active Motif. BRD2 and BRD4 Abs were obtained from Bethyl Laboratories (Montgomery, TX), and Pol2 Abs were procured from Active Motif. Quality control and alignment of the resulting 75-bp sequencing reads were also performed at Active Motif. Alignment to the human genome (hg19) was performed using the Burrows-Wheeler Aligner (16) with default settings. Reads that had >2 mismatches and multimapping reads were removed followed by removal of PCR duplicates through deduplication with samtools (H. Li, manuscript posted on arXiv, DOI: arXiv:1303.3997v2). The resulting binary alignment map files were used for further peak calling and analysis. MACS2 (17) was used to call peaks with the input as control. Both narrow and broad peaks were called using MACS2; however, for the purposes of differential analysis, the narrowPeak calls were used. Comparisons between JQ1-treated and DMSO controls to identify changes in occupancy were performed using the DiffBind (18) R package.

RA-FLS (three RA donors) were seeded at 3 × 10⁵cells/well in a six-well plate with 2 ml media and incubated overnight for cells to attach. Cells were treated with JQ1 (1 μM) or DMSO and stimulated with 1 ng/ml IL-1β. After 24-h incubation, total RNA was extracted, and the quality of all RNA samples was evaluated using an Agilent Bioanalyzer. RNA sequencing (RNA-Seq) was performed by BioProcessing Solutions (Piscataway, NJ). Sequencing libraries were prepared using TruSeq Stranded Total RNA RiboZero protocol from Illumina. Libraries were pooled and sequenced with an Illumina HiSeq 2000 with paired-end 100 bp flow cells. Raw read quality was evaluated using FastQC. Differential expression analysis was performed using the voom-limma framework implemented by the limma Bioconductor package (19). Genes that had median count values >1 were retained for further analysis. The voom command was used to log transform the counts and generate weights for further downstream differential analysis using a linear modeling framework with limma. The Benjamini–Hochberg multiple testing correction was used to generate adjusted p values, which were then used to filter for significance of the differentially expressed genes.

RA-FLS (three RA donors) were seeded at 300,000 cells/well in six-well plates and incubated overnight. Cells were treated with JQ1 (1 μM) or along with 1 ng/ml IL-1β and continued incubation for another 24 h. FLS were trypsinized, washed, counted for viability, and aliquoted at 7500 cells/vial in 1 ml freezing media and frozen at −80°C and then in liquid N2 according to vendor’s protocol (Epinomics, San Francisco, CA). ATAC-Seq (20) was performed at Epinomics, and the resulting fastq files were analyzed in house with a custom pipeline. First, quality control of the fastq files was performed using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc), followed by adaptor trimming and subsequent alignment of the trimmed reads to the human genome (hg19) using bowtie2 (http://broadinstitute.github.io/picard) with the -X 2000 option to allow for large insert sizes. PCR duplicates and alignments with a mapping score quality <10 were removed from the aligned reads using picard (21) and samtools (17), and binary alignment map files were generated for further analysis. To identify regions of chromatin accessibility, the aligned binary alignment map files from technical replicates were combined together and peak calling with MACS2 performed using -nomodel –shift–extsize. Downstream differential analysis upon peak calling was performed using DiffBind as with the ChIP-Seq data.

Annotation of ChIP-Seq and ATAC-Seq was performed using the R packages:ChIP-Enrich (22) and ChIPpeakAnno (23) in combination with human genome annotation packages. HOMER (24) was used for the identification of transcription factor binding motifs.

JQ1, a BET protein inhibitor, suppresses TNF-α–induced expression of IL-6, IL-8, and IL-1β in RA-FLS (13). We examined the effect of the active isomer of JQ1 (JQ1⁺) and its inactive isomer (JQ1⁻) on cytokine production in IL-1β– or TNF-α–activated RA-FLS. We found that JQ1⁺ but not its inactive isomer JQ1⁻ inhibited the IL-1β–induced production of proinflammatory cytokines G-CSF, GM-CSF, GRO, IL-6, IL-8, MCP-1, and VEGF in a dose-dependent manner (Fig. 1A). In addition, JQ1⁺ also specifically inhibited the TNF-α–induced levels of G-CSF, GM-CSF, IL-6, IL-8, MCP-1, and RANTES in RA-FLS (data not shown). Because MMPs play an important role in the pathogenesis of RA, we next examined the effect of the BET inhibitor JQ1⁺ on IL-1β–induced production of MMPs in RA-FLS. JQ1⁺ but not its inactive isomer JQ1⁻ decreased the production of IL-1β–induced MMP1 and MMP3 in a dose-dependent manner (Fig. 1A). JQ1⁺ did not inhibit the levels of MMP2 and MMP10 with or without IL-1β stimulation (data not shown). JQ1⁺ did not significantly affect the viability of RA-FLS cells after 24 h of treatment (MTT assay; data not shown) when compared with DMSO-treated cells. Furthermore, JQ1 inhibited the IL-1β–induced proliferation of RA-FLS in a dose-responsive manner (Fig. 1B).

To understand the global transcriptional effects of JQ1 treatment on RA-FLS, we studied changes in the genome-wide occupancy of BRD2, BRD4, and Pol2 using ChIP-Seq on pairs of JQ1 or DMSO-treated RA-FLS. A total of 18,795 BRD2 peaks (42% overlapping the transcription start sites [TSS], 18.1% inside the gene regions, 24.4% upstream of the TSS, and 14.2% in the downstream regions) and 14,613 BRD4 peaks (21.2% overlapping the TSS, 24.3% inside the gene, 33.2% upstream of the gene TSS, 19.5% downstream or overlapping the end of the gene, and 0.9% overlapping the entire gene) were detected in RA-FLS. Upon JQ1 treatment, BRD2 occupancy showed the most changes, in which its occupancy was lost in 7917 genomic locations (false discovery rate [FDR] ≤ 0.05), whereas BRD4 occupancy was lost in 1829 genomic regions (Fig. 2A). Similarly, we found that Pol2 occupancy was lost in 577 genomic regions upon JQ1 treatment (FDR ≤ 0.05). A majority of the BRD2 regions that lost occupancy after JQ1 treatment were not occupied by BRD4 (Fig. 2A), indicating that BRD2 may play a distinct nonredundant role in gene transcription in RA-FLS. Moreover, some of the genomic regions that showed a change in BRD4 occupancy overlapped with regions that had altered BRD2 occupancy (Fig. 2A), suggesting that BRD4 may function jointly with BRD2 to mediate JQ1-depdendent gene transcription. The occupancy of BRD2, BRD4, and Pol2 together was lost in 218 genomic regions (Fig. 2A). Most changes in BRD2 occupancy occurred at regions that overlap gene TSS, whereas BRD4 occupancy changes occurred in regions upstream, downstream, or inside of genes, thus demonstrating that these BET proteins display distinct genomic occupancy in mediating JQ1 action on RA-FLS (Fig. 2B). In contrast, Pol2 occupancy was lost more or less uniformly in regions that were inside a gene or were upstream or downstream of a gene (Fig. 2B, pie graph). We also found that after JQ1 treatment, BRD4 chromatin peaks were strongly correlated with changes in BRD2 occupancy, suggesting that BRD2 and BRD4 might be components of a protein complex bound to the genome (Fig. 2C). In contrast, there was not a strong correlation between BET proteins and Pol2 occupancy changes upon JQ1 treatment (Fig. 2C). As expected, the loss of Pol2 occupancy in or overlapping a gene TSS (Fig. 2D, lane 1) resulted in a corresponding loss of JQ1-mediated gene expression in IL-1β–treated RA-FLS (Fig. 2D, lane 2). The gene expression changes were highly significant as shown by positive FDR values (Fig. 2D, lane 3). Therefore, JQ1 resulted in the loss of BRD2, BRD4, and Pol2 in the genomic regions as well as a decrease in the expression of 218 genes in IL-1β–activated RA-FLS.

Because BRD4-loaded superenhancers have been described in human diffuse large B cell lymphoma, which accounted for almost a third of the total BRD4 enhancer region binding in the cell (25), we next examined RA-FLS superenhancer genes marked by binding of BRD2, BRD4, or those cobound with both BRD2 and BRD4 proteins. In addition, we also examined the effect of JQ1 treatment on BRD2/BRD4 occupancy of the superenhancer genes. The differential ChIP-Seq analysis revealed JQ1-dependent BRD2 peak changes in regions of variable lengths from few 100 bp to several kb in length, with a median length of ∼1863 bp, whereas the corresponding median length for BRD4 peak changes were ∼3100 bp, demonstrating that overall, BRD4 imparts a larger footprint on superenhancers than BRD2. To identify JQ1-mediated changes, BRD2/BRD4 co-occupied superenhancer genes (regions of >5 kb length that showed a change in both BRD2 and BRD4 occupancy) were examined. In addition, the regions that were close (within 5 kb, upstream) to a gene TSS and whose associated genes showed at least a moderate change in Pol2 occupancy (FDR < 0.15) and a significant change in gene expression upon JQ1 treatment (Benjamini–Hochberg–corrected p value <0.05) were cataloged. Most of the BRD2/BRD4 peaks impacted by JQ1 in IL-1β–treated RA-FLS covered <5 bp of DNA (Fig. 3A). A list of 10 superenhancer genes is shown in Table I, and the genomic tracks showing occupancy for BRD2 and BRD4 on IL-6 and IL-8 upstream regions with or without JQ1 treatment in IL-1β–stimulated RA-FLS are shown in Fig. 3B. Prominent among these are regions that are either upstream of or overlap the CXCL8 (IL-8), IL-6, CCL2, PRDM1, lysophosphatidic acid receptor 1 (LPAR1), WNT5A, and FLRT2 genes, which were loaded with both BRD2 and BRD4 in their genomic region and showed a loss of both BRD2 and BRD4 occupancy over larger stretches of DNA after JQ1 treatment. RA-FLS BET protein superenhancer genes IL-6 and IL-8 exhibited changes in JQ1-dependent BRD2 and BRD4 occupancy spanning lengths of ∼10 kb (BRD2) and 14 kb (BRD4) for IL-6 and 20 kb (BRD2) and 32 kb (BRD4) for IL-8 (Fig. 3B). Similarly, we found that the loss of occupancy occurs in a region of ∼9.9-kb (BRD2) and 14-kb (BRD4) length for WNT5A (data not shown). These findings suggest that JQ1 treatment resulted in a loss of superenhancer marks corresponding to the IL-6, IL-8, and WNT5A genes (Fig. 3A and data not shown). Further, we found that losses in BRD4 and BRD2 occupancy spans regions of length >12 kb for the LPAR1, PTGIS, gremlin 1 (GREM1), CCL2, FLRT2, and KCNJ15 genes, suggesting that JQ1 treatment inactivates superenhancer regions in each of these genes. The fold changes in BRD2 and BRD4 occupancy of BET protein co-occupied superenhancer genes in JQ1-treated IL-1β–stimulated RA-FLS are shown in Table I (columns 4 and 5). We next examined the effect of JQ1 on the expression of above-mentioned superenhancer genes in RNA-Seq expression profiling data obtained from compound or DMSO-treated IL-1β–activated RA-FLS. Consistent with the observed loss of BRD2, BRD4, and Pol2 occupancy (Table I, columns 4–6), highly significant downregulation of the expression of these genes upon JQ1 treatment of IL-1β–activated RA-FLS was observed (Fig. 3C, Table I, columns 2 and 3). These findings suggest that JQ1 downregulates the expression of prominent proinflammatory modulators by sharply inhibiting the formation of BRD complexes and thus adversely affecting transcription.

Given that the BET protein inhibitor JQ1 has shown efficacy in animal models of arthritis (13, 14), we next examined if the BRD2/4 pathway is activated in established RA. Therefore, we interrogated the expression of BRD2 and BRD4 co-occupied superenhancer genes in established RA (>12 mo of disease) synovial tissue disease samples (26, 27). RNA-Seq data from established RA disease joint biopsy samples (26, 27) were compared with those obtained from healthy normal synovial tissues samples. Eight out of ten BRD2/BRD4 superenhancer genes (CXCL8, IL-6, KCNJ15, GREM1, CCL2, WNT5A, PRDM1, and FLRT2) showed increased expression in established RA synovium when compared with their normal counterparts (Table II). These results suggest the involvement of BRD2/BRD4 pathway in the pathogenesis of RA.

To further understand regulatory and epigenetic changes in the RA-FLS upon JQ1 treatment, we profiled three sets of IL-1β–activated RA-FLS samples from different donors with and without treatment with JQ1 using ATAC-Seq to evaluate changes in chromatin accessibility and nucleosome occupancy. The experiment was performed on RA-FLS that were grown both in the presence and absence of IL-1β in the culture medium. Surprisingly, there were few significant changes in chromatin occupancy in the absence of IL-1β, but in the presence of IL-1β, there were substantial changes in chromatin accessibility upon JQ1 treatment, implying that inflammatory conditions enhance the sensitivity of RA-FLS cells to the BRD inhibitor at an epigenetic and chromatin level (Fig. 4A). As with changes in BRD2/4 occupancy, the majority of the regions that were altered by JQ1 treatment showed a loss of chromatin occupancy. ATAC-Seq data can be mined to find enrichment of consensus binding motifs to provide a global view of potential transcription factors that might bind to the accessible chromatin sites. An enrichment (FDR-corrected p values <10⁻³⁰) of consensus binding motifs belonging to the AP-1, NF-κB, and ETS1 families was observed, implying that these transcription factors have a reduced ability to access their target genes upon BRD2/4 inhibition (Fig. 4B). In addition, interrogation of changes in RNA expression of genes from these families (FOSL1, NFKB1, and NFKB2) revealed that the expression of these genes was significantly downregulated upon JQ1 treatment of RA-FLS (Fig. 4C), suggesting that in addition to a loss of accessibility to their target genes, the expression of AP-1 and NF-KB transcription factors themselves is directly repressed by JQ1 treatment. In contrast, JQ1 treatment increased the expression of ATF3 in RA-FLS (Fig. 4C).

Differential ChIP-Seq occupancy of BRD2, BRD4, and Pol2 and the changes in chromatin accessibility induced upon JQ1 treatment indicated that BRD2 and BRD4 inhibition appears to alter the expression of major inflammatory signaling proteins and regulatory transcription factors. Thus, an implication of our findings is that BRD2/4 inhibition could modulate major inflammatory pathways in RA-FLS. To further examine this hypothesis, pathway enrichment analysis was performed using the ingenuity pathway analysis tool on genes that were significantly differentially expressed in RA-FLS upon JQ1 treatment (FDR-adjusted p value ≤0.05) and exhibited significant changes in BRD2 and BRD4 occupancy or Pol2 occupancy within 5 kb upstream of their TSS (237 genes). Interestingly, downregulation of several prominent inflammatory pathways (p value < 10⁻³) post–JQ1 treatment was observed including those involved in IL-17F, p38 MAPK, HMGB1, IL-6, IL-1β, IL-8, TLR, and inflammasome signaling (Fig. 5A). We next interrogated the message expression of key anchor genes of these pathways in synovial tissue samples obtained from established RA as well as from various stages of RA disease progression (i.e., arthralgia, undifferentiated arthritis, and early RA) (27). We found that the expression of IL-1β, IL-6, IL-8, TLR4, HMGB1, and IL-18 was significantly increased in the synovium at various stages of RA disease progression (Fig. 5B, and data not shown). To examine if the whole pathway is enriched in RA synovium, we interrogated the enrichment of Kyoto Encyclopedia of Genes and Genomes TLR signaling and REACTOME TLR4_NFKB_MAPK modules in pre-RA and RA synovial tissue samples. These modules showed significant enrichment in early and established RA synovial tissue samples (Fig. 5C). Thus, JQ1 may not only downregulate the anchor genes but may also dampen the whole pathway in the disease tissue. To prove this point, we examined the expression of JQ1-affected pathway genes in activated RA-FLS. JQ1 indeed regulated the expression of downstream pathway genes belonging to IL-1β, IL-6, and TLR4 signaling networks in IL-1β–activated RA-FLS. Note that most of the genes covering these pathways were downregulated in IL-1β–activated RA-FLS (Fig. 5D).

FLS, the major cell types in RA synovium, perpetuate joint inflammation by getting activated through innate and adaptive immune signaling pathways. They secrete a multitude of cytokines/chemokines and MMPs and are the major effectors of tissue degradation. The pannus-like structure formed by activated proliferating FLS invades cartilage/bone and results in joint destruction (28). Observations that biological immune-targeted therapies show good therapeutic efficacy in approximately half of the patients coupled with the demonstration that cadherin-11 (a FLS-selective adhesion molecule) knockout mice display hypoplastic synovium and resistance to inflammatory arthritis and synovial hypertrophy (29) indicated that FLS play a crucial role in RA pathogenesis and development. In this study, we have explored the promise of BRD proteins as potential targets for modulating RA-FLS biology. Using an integrative systems immunology approach, we demonstrate that 1) pharmacological inhibition of BRD protein action downregulates RA-FLS activation and proliferation, 2) BET protein inhibitor JQ1 exerts its effects by disrupting the occupancy of superenhancer regions of crucial human disease relevant inflammatory genes by BRD proteins, 3) the BET inhibitor downregulates major proinflammatory pathways, and 4) JQ1 impairs the chromatin accessibility of critical proinflammatory transcription factors and also downregulates the expression of some of these key proteins.

All four BRD proteins are expressed in RA and osteoarthritis FLS; however, the levels of BRD2 and BRD4 are significantly increased in RA. Accordingly, the expression of TNF-α–induced secretion of IL-6, IL-8, MMP-1, and MMP-3 was decreased by a BET inhibitor or by BRD short hairpin RNAs (13). In line with these observations, we demonstrate that specifically, the active isomer of JQ1 but not its inactive counterpart inhibited the secretion of various cytokines/chemokines and MMPs (MMP-1 and MMP-3) in IL-1β–activated RA-FLS in a dose-responsive manner (Fig. 1). Cistrome analysis in RA-FLS revealed that even though many sites were occupied by BRD2 and BRD4, JQ1 resulted in three to four times more loss of BRD2 (42%) than BRD4 (12%) peaks (Fig. 2A). This differential loss of occupancy cannot be accounted for by difference in the binding affinity of JQ1 because it binds with equal affinities to BRD1 and BRD2 of BRD2 and BRD4 (30). Xiao et al. (13) have shown that short hairpin RNA knock-down of either BRD2 or BRD4 was as efficacious as JQ1 in downregulating the production of cytokines from TNF-α–activated RA-FLS. Therefore, taken together with our observation that the loss of BRD4 occupancy correlates with that of BRD2 (Fig. 2C), it appears that the BET protein inhibitor mediates its action in FLS predominantly by disrupting BRD2 and BRD4 co-occupied complexes. Interestingly, BRD2 predominantly occupied transcriptional start sites in RA-FLS chromatin compared with BRD4 (Fig. 2B), suggesting that these two BET proteins may also perform distinct transcriptional regulatory and chromatin architectural roles in RA-FLS. These observations are in concordance with the published data, wherein BRD2 and BRD4 have been shown to occupy distinct chromatin regions and perform nonoverlapping functions during murine Th17 cell differentiation (7). BRD2 selectively associated with the chromatin insulator protein CTCF, whereas BRD4 occupancy was observed predominantly with the STAT3 binding sites (7, 31), thus indicating that the BET proteins also perform nonoverlapping functions in the cell.

BRD4 superenhancer genes have been described in diffuse large B cell lymphoma cell lines, in which a large cellular load of BRD4 was present over a long stretch of DNA (>10 bp) in upstream regions of a few genes that were critically involved in promoting cell growth and inhibiting B cell differentiation into plasma cells (25). These superenhancers were strongly depleted of BRD4 after JQ1 treatment of diffuse large B cell lymphoma cell lines and included key master regulatory cell-specific transcription factors OCA-B, BCL-6, PAX5, IRF8, and MYC (25). Similarly, we have discovered 10 BRD2/BRD4 co-occupied superenhancer genes in JQ1-treated RA-FLS (Fig. 3B, Table I) that appear to be intricately involved in inflammatory processes in the synovium. To the best of our knowledge, this is the first report identifying BRD2/BRD4 co-occupied enhancers or superenhancers. These superenhancer genes also display increased expression in established RA synovium, indicating their involvement in disease pathogenesis (Table II). IL-6 is established as a key gene involved in RA development because two Abs targeting its receptor, namely tocilizumab and sarilumab, are used in clinic for the treatment of the disease, and an IL-6–targeting Ab (sirukumab) has also shown therapeutic efficacy (32). IL-8 has been proposed as a potential target for RA because anticitrullinated protein Abs induce osteoclast activation via secretion and autocrine action of IL-8, and neutralization of IL-8 has been proposed as a strategy for arresting RA progression at a very early stage of the disease (33). Furthermore, the loss of expression of IL-6 and IL-8 genes also implies a possible downregulation of the corresponding immune pathways initiated by these cytokines. Interestingly, GREM1, an antagonist of bone morphogenetic proteins, has been proposed as a key regulator of RA-FLS migration and hyperproliferation, and small interfering RNA–mediated knock-down or treatment with a specific GREM1 mAb resulted in inhibition of migration, proliferation, and survival of RA-FLS (34). Decreased bone morphogenetic protein signaling results in an increase in FGF and sonic hedgehog signaling pathways that are involved in cell cycle progression and antiapoptosis (34, 35).

CCL2 (MCP-1), a chemokine produced by many cell types including activated synoviocytes, is implicated in the pathology of RA and multiple sclerosis because of its role in attracting monocytes and lymphocytes to the sites of inflammation (36, 37). WNT5A is overexpressed in the synovium of RA patients, and its inhibition blocks the activation of cultured RA-FLS proliferation. Receptor tyrosine orphan receptor 2, a decoy receptor for WNT5A, reduces disease severity in CIA murine model (38). Furthermore, Wnt5a knockout mice are resistant to arthritis development in the K/BxN serum transfer-induced arthritis model (38). Similarly, LPAR1, which is highly expressed on RA-FLS and the synovium of RA patients, mediates the effect of its bioactive lipid ligand lysophosphatidic acid on the motility and IL-6/IL-8 production from synoviocytes (39). Furthermore, LPAR1 null mice failed to develop CIA, and in addition, pharmacological inhibition of LPAR1 using a small molecule inhibitor alleviated disease in the murine CIA and K/BxN serum transfer-induced arthritis models (40, 41). FLRT2 is an autoantigen, and its autoantibodies were detected in systemic lupus erythematosus, a disease in which joint involvement has also been observed in patients (42). Taken together, the above-mentioned preclinical and clinical biology of the superenhancer genes, along with our observation that the expression of these genes is significantly increased in RA synovium, highlights their importance in RA pathogenesis. JQ1 downregulates the expression of these BRD2/BRD4 co-occupied superenhancer genes by collapsing their core transcriptional machinery in RA-FLS cells.

Ingenuity pathway analysis of all genes that lost BRD2 and BRD4 or Pol2 occupancy within 5 kb of the TSS and were also regulated by JQ1 at the message level in RA-FLS revealed that they are involved in various proinflammatory pathways. For example, these genes were enriched in IL-6, IL-8, p38 MAPK, TLR, HMGB1, and inflammasome signaling (Fig. 5A). p38 inhibitors have shown modest efficacy in Ph2 RA clinical trials, suggesting that this pathway, although somewhat effective, may not have the required therapeutic effect as a standalone target for sustained remission of RA symptoms. However, it may show enough therapeutic efficacy in a combination regimen with other antirheumatic agents (43). Anakinra, an IL-1βR antagonist, is used in clinic for the treatment of RA and various other autoimmune and rheumatic diseases (44). HMGB1 protein is highly expressed in RA synovial fluid and synovial tissue and appears to play a key role in the pathogenesis of RA. Intra-articular administration of recombinant HMGB1 induced arthritis in mice, which resembled human disease in terms of synovial hyperplasia, pannus formation, and synovitis. Furthermore, treatment of murine CIA with anti-HMGB1 polyclonal Abs or the A-box of HMGB1 reduced disease severity (45). Thus, HMGB1 appears to be a therapeutic target for the treatment of RA. TLRs (TLR2, 4, 5, and 7) are also involved in RA pathogenesis, in which TLR signaling has been proposed to perpetuate and maintain the chronicity of inflammation in the joint. TLR signaling can transform RA myeloid cells to M1 macrophages and thus exacerbate inflammation (46). Interestingly, the expression of the anchor molecules representing these pathways (IL-1β, IL-6, IL-8, HMGB1, IL-18, and p38 MAPK) was increased during RA disease progression, thus indicating their intricate involvement in the pathogenesis and progression of RA (Fig. 5B). In addition, TLR, IL-1β, and p38 MAPK pathway genes were also enriched in early RA and established RA synovial tissue samples (Fig. 5C, 5D).

In this study, we demonstrate that JQ1 displaces BRD2 and BRD4 from upstream regions of key superenhancer and other relevant genes involved in proinflammatory pathways. Removal of BRD2 and BRD4 from the chromatin regions prevents the recruitment of the key transcription factors (AP-1 family, NF-κB, and ETS1) to promoters, thus leading to downregulation of the expression of genes that are critically involved in the pathogenesis and progression of RA. Interestingly, JQ1 targets various genes and pathways that have individually shown efficacy in clinic for the treatment of RA or others that have shown potential for treatment by their action in RA in vitro and in vivo models. Although JQ1 is a potent BET protein inhibitor, further experiments are needed to prove that this class of compounds have the required safety profile. Our results suggest that a safe and effective BET inhibitor may show efficacy not only in the treatment of RA but also in halting its progression to clinically classifiable RA.

We thank Drs. Anuk Das and Joshua Friedman for critical analysis of some of the experimental data.

All authors are current or former employees of Janssen Research & Development, Johnson & Johnson.