Genome-wide binding of BET proteins exhibits dose-dependent displacement upon I-BET treatment
To evaluate I-BET resistance in leukemia, we used the pan-BET inhibitor GSK1210151A (I-BET151) [15], in two cell line models: the CML (chronic myeloid leukemia) I-BET-resistant K562 and the AML (acute myeloid leukemia) I-BET-sensitive MV4;11. Proliferation of K562 cells is driven by a BCR-ABL translocation, and these cells exhibit resistance to I-BET151 treatment up to micromolar concentrations [15]. On the contrary, MV4;11 cells are dependent on a MLL-AF4 fusion and are sensitive to I-BET151 anti-proliferative effect in the nanomolar range (Fig. 1a). To characterize the consequences of I-BET151 treatment on BET proteins bound to chromatin, we undertook a comparative approach by incubating MV4;11 and K562 cells with increasing concentrations of the inhibitor, followed by ChIP-seq for BRD2, BRD3 and BRD4 (Fig. 1b, “Materials and methods” section). For each cell line and condition (vehicle, I-BET151 50 nM, 500 nM or 5000 nM), two replicates exhibiting significant enrichment of known BET proteins-bound loci (verified by qPCR, “Materials and methods” section, Fig. 1c and Additional file 1: Table S1) were subjected to high-throughput sequencing.
Quantitative PCR on both cell lines showed general and gradual decrease in BET proteins binding upon treatment with increasing inhibitor concentration (Fig. 1c).
Peak calling based on a MACS2-derived pipeline [38] followed by joining operations was performed to call the final set of peaks and filter low-confidence ones (Additional file 2: Figure S1, “Materials and methods” section).
Interestingly, more BRD4 binding was observed in the I-BET-sensitive MV4;11 cells compared to the I-BET-resistant K562 cells (Fig. 1d, Additional file 3: Figure S2 and Additional file 4: Table S2). For K562, we observed a substantial reduction in peak counts for all three BET proteins at and above 500 nM of I-BET151 treatment. In contrast, a differential pattern was seen in MV4;11: BRD3 and BRD4 showed a clear decrease in peak numbers concomitant with increasing I-BET151 concentrations, while BRD2 showed an unusual binding scheme with more peaks observed at 500 nM, and generally less peaks compared to BRD3 and BRD4 (Additional file 3: Figure S2 and Additional file 4: Table S2) which could also be explained by the poor quality of the BRD2 antibody in ChIP-seq.
In general, the measured enrichment of BRD2, BRD3 and BRD4 peaks at promoters and H3K27ac marked regions (approximately 78% of all peaks, Additional file 3: Figure S2C and Additional file 4: Table S2), is in line with the role of BET proteins in cis and trans transcriptional regulation.
BRD4 is the most sensitive BET paralog to I-BET treatment on enhancers and promoters
Previous studies showed that treatment with I-BET results in partial displacement of BRD2, BRD3 and BRD4 from their chromatin-bound loci [12, 36, 39,40,41,42]. The displacement is observed at the TSS or proximal to it, at extended promoters as well as at distant enhancers and super-enhancers [12]. However, little is known about the correlation between genome-wide displacement of the individual BET protein paralogs and I-BET concentration in resistant and sensitive cell lines. To address this point, we compared BET proteins ChIP-seq signals at the promoters in K562 and MV4;11 (Fig. 2, “Materials and methods” section).
The aggregation of BET proteins ChIP-seq signals around all TSSs in a meta-profile showed that binding of the BET proteins is globally stronger in the I-BET-sensitive MV4;11 compared to the I-BET-resistant K562 (Fig. 2a), also reflected by the higher number of peaks called in MV4;11 cells (Additional file 4: Table S2). All BET proteins showed a marked bimodal distribution with one strong signal peak approximately 150–180 bp downstream of the TSS and a less pronounced peak located at approximately 300 bp upstream of the TSS (Fig. 2a), consistent with established patterns of histone tail acetylation at positioned nucleosomes flanking the nucleosome-depleted region over gene promoters [43]. Notably, the difference in signal intensity between the downstream and the upstream peak was stronger in the sensitive MV4;11 cells than in the resistant K562 cells (Fig. 2a, c). The bimodal distribution of the BET proteins signal around the TSS resembles the enrichment patterns of H4K5ac and H3K27ac at these sites, while Pol II peaks mainly downstream of the TSS (Fig. 2b) [24, 39, 42]. Interestingly, among the three paralogs, BRD4 showed a stronger displacement at the lowest concentration of the inhibitor (50 nM) in MV4;11 reflected by a lower ChIP-seq signal. This suggests that BRD4 is the most I-BET151-sensitive BET paralog (Fig. 2a).
Moreover, since BRD4 recruits the P-TEFb complex to chromatin which in turn binds to Pol II, a co-localization of Pol II with BRD4 is expected. Although Pol II ChIP signal peaks mostly downstream of the TSS, a shoulder in the peak is noticeable upstream of the TSS (Fig. 2b). Consistent with the role of BET proteins in transcriptional elongation, the downstream peak showed a broader signal than the upstream peak in both cell lines for all BET proteins and particularly for BRD4 (Fig. 2a, c).
Similar binding profiles were observed at cell-type-specific intergenic enhancers comprising typical enhancers and super-enhancers (Additional file 5: Figure S3, “Materials and methods” section). BRD2 binding on enhancers was less pronounced than BRD3 and BRD4 for both cell lines with BRD4 exhibiting a more pronounced displacement at 50 nM compared to BRD3 in the sensitive cell line MV4;11 (Additional file 5: Figure S3).
I-BET-mediated displacement of BRD4 around the TSS identifies a small set of hypersensitive genes and defines four clusters of differentially sensitive promoters
Although BET proteins displacement from chromatin after I-BET treatment has been well described [12, 39, 40], spatial resolution of displacement around the TSS has not been assessed in detail. Since BRD4 binding is particularly sensitive to I-BET in MV4;11 (Fig. 2a), we focused our analysis on this paralog.
We performed differential binding (DB) analysis using DESeq2 [44] (“Materials and methods” section) for BRD4 signal at − 1 Kb from the TSS (defined as core promoter) and at + 1 Kb from the TSS (defined as Pol II pause site) of the 32,091 annotated TSSs in the human genome (Fig. 3a).
The number of affected TSSs increased with I-BET151 concentration in both cell lines: In total, 11% TSSs (3531) in the I-BET-sensitive MV4;11 and 1.6% TSSs (519) in the I-BET-resistant K562 showed significant displacement of BRD4 upon treatment with I-BET151 (Fig. 3a, b Additional file 6, 7, 8: Tables S3, S4 and S5) (“Materials and methods” section). Additionally, BRD4 displacement from the core promoter was more than from the pause site (Fig. 3b), in contrast to the higher BRD4 signal observed at pause site (Fig. 2a). Moreover, a considerable number of TSSs showed displacement at both the core promoter and pause site which was higher in MV4;11 than in K562 (Fig. 3b).
Notably, five TSSs in the sensitive MV4;11 cells showed significant BRD4 displacement upon treatment with 50 nM of I-BET151, while no significant binding loss was observed for the resistant K562 under the same condition (Fig. 3b). The five affected TSSs correspond to four non-coding RNAs: LINC01585, LINC-ROR, NRON, LINC02367 and one protein coding gene, RBM38.
To further examine the I-BET151-dependent displacement of BRD4 upstream and downstream of the TSS, we used k-means clustering and split TSSs into four groups based on their ChIP-seq log fold change signal, independently for both cell lines (Fig. 3c and Additional file 9: Table S6). For MV4;11, clusters 1 and 2 (1274 and 450 TSSs, respectively) were characterized by medium and strong BRD4 displacement at + 1 Kb and − 1 Kb from the TSS and were named “sensitive core promoter and pause site” and “highly sensitive core promoter and pause site,” respectively. Clusters 3 and 4 (735 and 1072 TSSs, respectively) showed strong BRD4 binding loss at + 1 Kb or − 1 Kb from the TSS and were named “highly sensitive pause site” and “highly sensitive core promoter,” respectively (Fig. 3c; Additional file 10: Figure S4; Additional file 7, 8: Tables S4 and S5; “Materials and methods” section). Intriguingly, the four non-coding RNAs in which BRD4 shows a DB at 50 nM I-BET treatment (Fig. 3b) clustered in the highly sensitive promoter and pause site group (Fig. 3c,d cluster 2).
The same clustering approach was applied on the set of 519 TSSs in K562 (Fig. 3c, left panel). Interestingly, a similar clustering profile as for MV4;11 was identified in K562 with a balanced distribution of TSS in the four clusters (cluster 1:168 TSSs, cluster 2:108 TSSs, cluster 3: 132 TSSs and cluster 4: 111 TSSs).
By comparing the number of genes common to equivalent clusters from both cell lines, we found that only 55 genes belonging to the sensitive core promoter and pause site clusters were common in MV4;11 and K562, 31 genes belonging to the highly sensitive core promoter and pause site clusters, 19 genes belonging to the highly sensitive pause site clusters and 43 genes belonging to the highly sensitive core promoter clusters. This finding highlights differential sensitivity of promoters to I-BET-related displacement between the two cell lines.
To check whether differential BRD4 displacement is reflected on transcription, we quantified the expression of the five genes which showed BRD4 displacement exclusively in the I-BET-sensitive cells after 50 nM I-BET treatment. All five genes were downregulated in MV4;11 cells and either not detected or less affected in K562 (Fig. 3e). In MV4;11 cells, LINC-ROR showed a decrease in its expression, although not statistically significant (p value of 0.065; paired t-test). Since most long non-coding RNAs including LINC-ROR are known to have very low expression levels at steady state [45,46,47,48], the observed decrease in expression could have profound effects on its downstream targets.
The finding that four long non-coding RNAs with BRD4 binding at their promoters are among the most sensitive transcripts to I-BET in MV4;11 but not in K562 suggests a previously uncharacterized role for non-coding RNAs in conferring sensitivity to I-BET treatment.
Chem-seq is predictive of differential effects of I-BET151 at promoters
In order to map genome-wide BRD4/I-BET151 interactions, we used a chemical affinity capture-based method, Chem-seq [11, 24] (Fig. 4a, “Materials and methods” section). This approach can both define loci where the BET protein BDs are accessible to the inhibitor and potentially actionable, thereby identifying different BET protein populations on chromatin. For this, we used a biotinylated derivative of I-BET121 (pan-BET inhibitor chemically similar to JQ1) [15] to pull down the associated chromatin fragments in cross-linked I-BET-sensitive MV4;11 cells followed by high-throughput sequencing. To control for specificity of binding, the pull down was also performed after addition of excess free I-BET151 (Fig. 4b). Interestingly, Chem-seq showed stronger signal enrichment downstream of the TSS for genes belonging to both the sensitive and highly sensitive core promoter and pause site clusters as well as the highly sensitive pause site cluster (clusters 1, 2 and 3) (Figs. 3c, 4b). In particular, the highly sensitive pause site cluster genes characterized by BRD4 DB only downstream of the TSS showed the highest signal in Chem-seq.
Conversely, TSSs belonging to the highly sensitive core promoter cluster (cluster 4) showed a typical bimodal distribution with no notable enrichment downstream or upstream of the TSS after Chem-seq. Chemical affinity capture of BET proteins bound to chromatin occurs mainly via the BD2 domain and thus is a measure of BD2 occupancy [24]. This observation indicates that BET proteins strongly displaced by I-BET treatment downstream of the TSS are characterized by greater availability of the BD2 domain. Stronger Chem-seq profiles downstream of the TSS indicate higher accessibility to I-BET treatment and might be used to predict gene responsiveness to BET protein inhibition.
I-BET treatment induces apoptotic-related pathway signatures in the sensitive cell line
Next, we were interested to know whether I-BET treatment also affected global gene expression differentially in resistant and sensitive cell lines and if so, whether the differentially affected genes are connected to specific cellular pathways.
RNA-seq was performed in MV4;11 and K562 cells treated with 50, 500 or 5000 nM I-BET151 (“Materials and methods” section), and differentially expressed genes were called using DESeq2 [44]. As expected, the number of deregulated genes increased with increasing I-BET concentrations in both cell lines (Fig. 5a) (Additional file 11, 12: Tables S7 and S8).
To correlate BRD4 displacement from promoter regions with corresponding gene expression at a global level, we compared the number of genes showing differential BRD4 binding at TSS to those exhibiting differential expression. Although both MV4;11 and K562 exhibited a similar number of downregulated genes (Fig. 5a), the degree of overlap between differentially bound and downregulated genes was higher for the sensitive MV4;11 cell line (Fig. 5b), which suggests a higher number of direct effects of the I-BET drug in the sensitive MV4;11 cell line. For instance, at I-BET 5000 nM, 34.3% (542/1581) of downregulated genes in MV4;11 exhibit differential binding within 1 Kb upstream and downstream of the TSS compared to only 8% (112/1389) in K562.
We then identified enriched pathways using gene set enrichment analysis (GSEA) [49] in pre-ranked mode using all genes following I-BET treatment (“Materials and methods” section) in both cell lines. Interestingly, several genes in pathways linked to cell viability were differentially affected at 50 nM I-BET treatment in both cell lines (Fig. 5c, Additional file 13: Figure S5). For instance, genes in the P53 pathway were upregulated in the I-BET-sensitive MV4;11 (normalized enrichment score or NES = 2.6, FDR FWER p value = 0) and conversely downregulated in the I-BET-resistant K562-treated cells (NES = − 2.13, FWER p value = 0.056). Although K562 cells do not express a functional form of TP53, the P53 pathway is known to be functional in this cell line [50]. Similar differences between the two cell lines were observed for TNF-alpha signaling via NF-kappaB, DNA repair and apoptosis pathways (Fig. 5c). Our results suggest that MV4;11 sensitivity to I-BET treatment is manifested at the transcriptional level by an upregulation of gene sets belonging to pathways which drive cells into cell cycle arrest and/or cell death and oxidative phosphorylation and that an opposite response is shown in the resistant cell line K562. At higher inhibitor concentrations, more pathways become similarly affected in both cell lines (Additional file 13: Figure S5), most likely because the resistant K562 cell line also becomes more responsive to I-BET.
BRD4 displacement downstream of the TSS prolongs RNA Pol II pausing and decreases gene expression
BRD4 displacement at promoters has been linked to downregulation of its target genes. Since BRD4 is essential for transcription elongation which starts downstream of the TSS, we postulated that BRD4 displacement at 1 Kb downstream of the TSS, spanning the Pol II pause site, is mostly responsible for affecting gene expression. To test for this hypothesis, we compared gene expression of the different DB clusters identified above (Fig. 3c) between K562 and MV4;11 (Fig. 6a).
Genes with sensitive and highly sensitive core promoter and pause site to I-BET in MV4;11 cells (i.e., belonging to clusters 1 and 2) are highly downregulated by I-BET in this cell line but not in the I-BET-resistant K562 cells (Fig. 6b, Wilcoxon test p = 3.5e−13 and p < 2.2 e−16, respectively). The expression of genes belonging to clusters 1 and 2 defined in K562 is similarly affected by I-BET treatment in both cell lines (Fig. 6b, Wilcoxon test p = 0.64 and p = 0.17, respectively). Most interestingly, genes belonging to the highly sensitive pause site cluster in MV4;11 showed a very significant reduction in expression in this cell line compared to the corresponding category of genes in K562 (Fig. 6b, Wilcoxon test p = 1.7e−11). Moreover, this is in strong contrast to the genes in the highly sensitive core promoter cluster that show no reduction in expression in the sensitive cell line MV4;11 (p = 0.96).
Since BRD4 is important for both transcription initiation and elongation, these results suggest that the strongest loss of expression observed using I-BET151 in the sensitive MV4;11 cell line is due to BRD4 displacement at the Pol II pause site, downstream of the TSS, which likely impairs transcription elongation [51]. This is in agreement with our Chem-seq data (Fig. 4b), where sensitive and highly sensitive core promoter and pause site clusters, as well as the highly sensitive pause site cluster, are most accessible to the drug. In contrast, genes belonging to the highly sensitive core promoter cluster (cluster 4) showed no topological signal bias by Chem-seq and their expression was not affected by I-BET.
If BRD4 displacement in the highly sensitive pause site cluster impairs transcription elongation, we would expect to observe an accumulation of Pol II binding at the promoter region, as opposed to the gene body, following I-BET treatment. This will affect the pause state of many genes, thus leading to an increase in their pause index (PI), also known as travelling ratio (TR). To test this hypothesis, we compared the TR for all four MV4;11 clusters before and after treatment with 500 nM I-BET [52] (“Materials and methods” section). Pol II traveling ratio was affected in all four clusters following I-BET treatment of the sensitive cell line MV4;11 (Fig. 6c). Interestingly, the pause site sensitive cluster showed a more pronounced increase in Pol II TR when compared to other clusters after drug treatment. In fact, 7.3% genes (55/732) gained in TR (TR > 2) in the highly sensitive pause site cluster compared to 3.6% (38/906) in the highly sensitive core promoter cluster, upon I-BET treatment (p value of 0.00031; two-sided proportion test). For the sensitive and highly sensitive core promoter and pause site clusters, 5.4% (68/1265) and 6.3% (28/444) genes displayed an increase in their TR, respectively, when compared to the core promoter sensitive cluster (p values of 0.046 and 0.025, two-sided proportion test) (Fig. 6c).
To summarize, sensitivity of MV4;11 cells to I-BET is characterized by a strong displacement of BRD4 downstream of the TSS which coincides with a longer pausing of RNA Pol II and a subsequent decrease in gene expression.