RB deficient cells are resistant to small molecule BET inhibitors
The RB1 gene is frequently deleted in many human cancers including neuroendocrine PCa (NEPC), an advanced form of PCa which is resistant to most existing treatments in clinic21. We, therefore, sought to investigate how RB loss affects drug resistance in PCa cells with NEPC traits. It has been reported previously that PC-3 PCa cells exhibit certain characteristics of NEPC22. We generated control or RB knockdown (KD) PC-3 PCa cells and treated them with a group of drugs or chemicals that are used in clinical or pre-clinical settings. As expected, PC-3 cells became resistant to the CDK4/6 inhibitor palbociclib when RB was knocked down (Fig. 1a and Supplementary Fig. 1a). Consistent with our previous finding19, RB-deficient cells were resistant to the NF-κB inhibitor JSH23 (Fig. 1a). Notably, these cells were also resistant to bromodomain inhibitors including BET BD1 inhibitors JQ1 and I-BET726 as well as CBP/p300 inhibitor CPI-637 (Fig. 1a and Supplementary Fig. 1b). RB knockdown also caused BET BD1 inhibitor resistance in another aggressive PCa cell line C4-2 (Fig. 1b and Supplementary Fig. 1a). Additionally, RB knockdown conferred resistance to BET inhibitors in LNCaP cells, a hormone-sensitive PCa cell line (Supplementary Fig. 1a, c). These results suggest that resistance to BET inhibitors could be a common phenomenon in RB-deficient PCa cells regardless of neuroendocrine phenotype. Further analysis showed that treatment of both PC-3 and C4-2 parental cells with JQ1 induced apoptotic cell death as evident by the cleavage of PARP (c-PARP) and Caspase-3 (c-Caspase-3), but these effects were abolished by RB KD (Fig. 1c, d). The inhibition of JQ1-induced apoptosis by RB KD was unlikely mediated through RB KD-induced upregulation of NF-κB target genes because expression of these genes was equivalently blocked by JQ1 in both control and RB KD cells (Supplementary Fig. 1d, e). These data suggest that RB loss blocks BET inhibitor-induced cell death through mechanisms other than the NF-κB signaling.
Due to the pivotal role of RB in regulating E2F1 activity, we examined the contribution of aberrantly activated E2F1 to BET BD1 inhibitor resistance in RB-deficient cells. It has been shown recently that RB deficiency induces JQ1 resistance in NUT midline carcinoma cells and such resistance is mediated by cell cycle regulators including E2F123. In PC-3 cells, however, we found that while JQ1 treatment downregulated expression of c-Myc, a known JQ1 inhibitory target24, it had no obvious effect on E2F1 expression (Supplementary Fig. 1f). DU145 is a PCa cell line in which one allele and exon 21 in the other allele of the RB1 gene are deleted25. To further test our hypothesis, we transfected DU145 cells with either wild-type (WT) RB or the E2F1-binding deficient mutant R661W26. As expected, reciprocal co-immunoprecipitation (co-IP) assays showed that R661W mutant failed to bind to E2F1 (Supplementary Fig. 1g). However, we demonstrated that increased expression of both WT RB and R661W equivalently sensitized DU145 cells to BET BD1 inhibitors JQ1 and iBET (Fig. 1e, f). Furthermore, E2F1 KD failed to reverse RB loss-mediated BET inhibitor resistance in PC-3, C4-2, and LNCaP cell lines (Fig. 1g, h and Supplementary Fig. 2a–d). Together, our data suggest that RB loss induces BET inhibitor resistance in a manner independent of E2F1 signaling in PCa cell lines examined.
RB-N interacts with BRD4 and the interaction is diminished by RB phosphorylation
It is known that RB binds to and regulates the activity of its target proteins, such as E2F1 and p65. We hypothesized that RB modulates BET inhibitor sensitivity by binding to and regulating BET protein activity. To test this hypothesis, we first performed co-IP experiments. We demonstrated that the ectopically expressed HA-RB interacted with BRD2, BRD3, and BRD4 in 293T cells (Supplementary Fig. 3a). RB interaction with these BET proteins were confirmed at the endogenous level in both C4-2 and PC-3 cells (Fig. 2a, b). RB protein is highly phosphorylated in proliferating cells. We sought to determine whether RB phosphorylation affects RB-BET protein interaction. 293T cells were transfected with HA-RB and cell lysate was treated with λ protein phosphatase prior to co-IP with anti-HA antibody. We demonstrated that phosphatase treatment largely enhanced RB interaction with BRD4, but not BRD2 and BRD3 (Supplementary Fig. 3b). Similar results were obtained at the endogenous level in both C4-2 and PC-3 cell lines (Fig. 2c, d). Moreover, we found that RBΔCDK mutant, in which fifteen CDK serine/threonine phosphorylation residues were mutated to alanine16, had greater interaction with BRD4 than the WT counterpart (Supplementary Fig. 3c). These data indicate that RB phosphorylation impairs RB-BRD4 interaction.
To define which region in RB mediates its interaction with BET proteins, we constructed three glutathione-S-transferase (GST)-RB recombinant protein vectors (Supplementary Fig. 3d) and purified the recombinant proteins from bacteria. Results from the GST pulldown assay showed that BRD4 specifically bound to the N-terminal portion of RB (RB-N) in vitro (Fig. 2e). In contrast, BRD2 and BRD3 interacted with both N- and C-terminal parts of RB, but the interactions were much weaker compared to the BRD4-RB-N interaction (Fig. 2e). Therefore, we focused on RB regulation of BRD4 in further studies. Co-IP assay not only confirmed that BRD4 interacted with RB-N in cells, but also demonstrated that their interaction is regulated by phosphorylation (Fig. 2f).
RB serine-249/threonine 252 phosphorylation disrupts RB-N-BRD4 interaction
Since dephosphorylation of RB-N enhances its interaction with BRD4, we decided to determine which phosphorylation sites are involved in this process. GST pulldown assay revealed that BRD4 bound to the N-terminal portion of RB-N which contains CDK4/6 phosphorylation sites serine-249 and threonine-252 (S249/T252), but no interaction between BRD4 and the C-terminal portion of RB-N was detected (Fig. 2g, h). We also found that treatment of CDK4/6 inhibitor palbociclib or co-knockdown of CDK4 and CDK6 enhanced RB-N interaction with BRD4 (Supplementary Fig. 3e, f). Moreover, the BRD4-RB-N interaction was largely enhanced by the S249/T252 phosphorylation-resistant mutant S249A/T252A or deletion of the S249/T252-containing motif 249SPRT252 (RB-NΔSPRT) (Fig. 2i and Supplementary Fig. 3g). These results suggest that S249/T252 phosphorylation inhibits BRD4-RB-N interaction. To mimic the RB-N phosphorylation at S249 and T252, we mutated S249 and T252 to aspartic acid (D). By performing GST pulldown assay, we demonstrated that the S249D/T252D mutant had less binding ability to BRD4 compared to the WT counterpart (Fig. 2j). In contrast, phosphorylation-mimicking mutation T356D/T373D had no effect on BRD4-RB-N binding (Fig. 2j), consistent with the observation that the C-terminal portion of RB-N does not interact with BRD4 (Fig. 2h). Next, we performed in vitro kinase assay prior to GST pulldown assay. We demonstrated that CDK4/6 diminished RB-N-BRD4 interaction; however, this effect was abolished by S249A/T252A or RB-NΔSPRT mutant (Fig. 2k and Supplementary Fig. 3h). Collectively, these data suggest that S249/T252 phosphorylation by CDK4/6 impedes RB binding of BRD4.
The 157FLQKI161 motif in BRD4 BD1 domain mediates BRD4-RB-N interaction
To determine the region in BRD4 that is required for RB binding, we generated four FLAG-tagged truncation mutants of BRD4 for co-IP assay. We found that only the BRD4 N-terminal region which contains two bromodomains (amino acids 1 to 470) bound to RB (Fig. 3a, b). Both GST pulldown and co-IP assays further showed that bromodomain 1 (BD1), but not BD2 binds RB (Fig. 3c, d and Supplementary Fig. 4a).
We and others have previously shown that an FXXXV motif in the RB-binding partners is important for their binding with RB19,20, we sought to determine whether there is any functional FXXXV motif in BRD4 that mediates binding to RB. We noticed that there is one typical FXXXV motif (83FQQPV87) and an atypical FXXXI motif (157FLQKI161) in BD1, the RB-binding region in BRD4 (Fig. 3e). To test the functionality of these two FXXXV/I motifs, we generated three alanine mutants (Fig. 3f) and utilized them for in vitro GST pulldown assay. While the 83AQQPA87 mutant alone had no effect on BRD4 BD1 binding of RB, the 157ALQKA161 mutation alone largely diminished and 83AQQPA87/157ALQKA161 double mutant almost completely lost the binding of RB (Fig. 3g). Similar results were obtained in co-IP assay performed in cell lysate (Fig. 3h). Given that RB interacts with BRD4 through BD1, a domain critical for BRD4 recognition of acetylated histone and non-histone proteins27,28, we examined whether inhibition of the bromodomain activity by JQ1 affects the RB-BRD4 interaction. We found that administration of JQ1 to the GST pulldown and co-IP assay buffers had no impact on BRD4 binding of RB (Supplementary Fig. 4b–e).
Among the three members of the pocket protein family, we found that only RB, but not other two related proteins p107 and p130 can bind BRD4 (Fig. 3i), implicating that there is a unique motif in RB for its preferential binding of BRD4. In agreement with this observation, RB protein is the only member of the pocket protein family that harbors an R-linker which contains BRD4-interacting sites S249/T252. Notably, there are a number of positively charged arginine (R) residues within the R-linker in RB-N and several negatively charged glutamic acid (E) residues surrounding the 157FLQKI161 motif in BRD4 (Fig. 3j, top). As described above, phosphorylation of RB-N S249/T252 impedes the BRD4-RB-N interaction. We hypothesized that the negative charges in the proximity of 157FLQKI161 are critical for BRD4 interaction with RB R-linker (Supplementary Fig. 4f, left panel). To test this hypothesis, we made an E-to-A mutant by replacing E residues with alanine (E/A mutants) surrounding 157FLQKI161 and an E-to-R mutant by replacing E residues with positively charged arginine (E/R mutant) (Fig. 3j, bottom) and performed GST pulldown assay. We demonstrated that the interaction between BRD4 BD1 and RB was attenuated by the E/A mutant and the interaction was further reduced by the E/R mutant (Fig. 3k). Together, we provide evidence that the 157FLQKI161 motif in BD1 and the negative charge in the adjacent residues are important for BRD4 binding of RB and that this interaction is diminished by the addition of negative charges in the RB R-linker due to S249/T252 phosphorylation by CDK4/6 (Supplementary Fig. 4f, right). To investigate the effect of RB on the binding ability of BRD4 to chromatin, we carried out the chromatin-binding assay and found that RB loss increased BRD4 binding of chromatin while the restoration of RB in RB-deficient cells reduced BRD4 binding on chromatin (Fig. 3l, m).
RB loss activates the GNB1L-CREB signaling axis through BRD4
BD1 is critical for BRD4 to recognize the acetylation moiety on histones and participates in transcription control of cancer-related genes2,3. Our data support the notion that RB binds to BD1 and limits BRD4 activity in chromatin binding and that RB loss renders resistance to BET inhibitors by increasing BRD4 engagement with chromatin and inducing transcriptional reprogramming. To interrogate this hypothesis, we performed BRD4 ChIP-seq in control and RB KD C4-2 cells. By running the Cistrome pipeline, we revealed that the quality of the BRD4 ChIP-seq data obtained from both control and RB KD C4-2 cells were very comparable to the BRD4 ChIP-seq results of approximately 300 different human cell samples in the Cistrome database (Supplementary Fig. 5a–l and Supplementary Data 1). We demonstrated that RB knockdown increased BRD4 occupancy in the genome (n = 107, Log2 (fold change) > 1, p < 0.05), but only reduced BRD4 binding at 4 loci (Fig. 4a, b, Supplementary Data 2 and 3). The 107 peaks with enhanced occupancy of BRD4 were related to 152 genes, some of which were enriched in the cAMP pathway (Fig. 4c). cAMP production functions downstream of the GPCR signaling. We demonstrated that RB depletion increased BRD4 occupancy at certain region(s) in the loci of genes in the GPCR-cAMP pathways such as GNB1L (Fig. 4d and Supplementary Fig. 6a, b). Using ChIP-qPCR assay, we confirmed the increased BRD4 occupancy in the loci of GPCR-cAMP genes such as EDNRA, GNB1L, GRIN3A, GRM4, and SSTR1 in both PC-3 and C4-2 cells (Fig. 4e, f). Consistent with the ChIP-qPCR data, expression of these five genes was upregulated after RB KD in both PC-3 and C4-2 cells except GRIN3A and GRM4 expression in PC-3 cells (Fig. 4g, h). GNB1L protein was also substantially upregulated in RB KD PC-3 and C4-2 cells (Fig. 4i). We further showed that knockdown of BRD4 alone not only decreased GNB1L expression and phosphorylation of CREB, a downstream effector of the GPCR-cAMP pathway, but also completely abolished RB KD-induced upregulation of GNB1L protein and CREB phosphorylation (Fig. 4j, k), suggesting that RB deficiency-induced upregulation of GNB1L is mediated through BRD4. Notably, similar results were obtained in PCa patient samples from the TCGA database29. We found that GNB1L mRNA expression was much higher in RB deletion samples compared to RB WT samples (Fig. 4l, m), and their expression was inversely correlated (Fig. 4n). Among other four GPCR-cAMP pathway genes expression of two of them (e.g., GRIN3A and GRM4) was also upregulated in RB-deficient tumors compared to Rb WT counterparts in the TCGA cohort and the difference was statistically significant (Supplementary Fig. 6c–f). These data suggest that RB can regulate expression of these GPCR-cAMP genes in cultured PCa cells and patient samples, but the regulation at certain gene loci could be influenced by the cellular contexts, especially in PCa samples from patients.
The high-level expression of GNB1L mRNA also significantly associated with poor survival of PCa patients (Fig. 4o). Accordingly, we showed that ectopic overexpression of GNB1L caused resistance to BET inhibitors in both PC-3 and C4-2 cells (Supplementary Fig. 7a–d). Moreover, knockout of GNB1L re-sensitized RB-deficient PC-3 and C4-2 cells to BET inhibitors (Supplementary Fig. 7e–h). Similar to the findings in LNCaP/AR cell line30, RB KD not only decreased AR mRNA expression, but also induced resistance to the antiandrogen enzalutamide in C4-2 cells (Supplementary Fig. 7i, j). However, co-knockdown of GNB1L failed to abolish RB KD-induced AR downregulation and enzalutamide resistance (Supplementary Fig. 7i, j). Moreover, similar to the findings in LNCaP/AR cell line30, we found that RB KD in C4-2 cells induced upregulation of neuron-specific enolase (NSE), a neuroendocrine cell marker (Supplementary Fig. 7i); however, this effect was not affected by co-knockdown of GNB1L (Supplementary Fig. 7i). These data support the observation reported previously30,31 that loss of RB promotes antiandrogen resistance and neuroendocrine phenotype. However, none of these effects was mediated through RB loss-induced upregulation of GNB1L. Of course, our data cannot completely rule out the possibility that other GPCR-cAMP-related genes upregulated by RB loss-induced aberrant activation of BRD4 may contribute to antiandrogen resistance and neuroendocrine phenotype in RB-deficient cells. E2F1 knockdown did not alter GNB1L expression in both RB-proficient and -deficient PC-3 cells (Supplementary Fig. 8a, b), indicating that its expression is not regulated by the RB-E2F1 axis. Together, our data show that RB deficiency induces upregulation of GNB1L and activation of the GPCR-cAMP-CREB axis in PCa cells and that this effect is mediated through BRD4.
RB S249/T252 phosphorylation promotes GNB1L expression and CREB activation
Since RB S249/T252 phosphorylation inhibits RB-BRD4 interaction, we sought to determine whether S249/T252 phosphorylation regulates GNB1L expression via BRD4. Consistent with the regulation of GNB1L expression by BRD4 knockdown, we found that in a dose-dependent manner treatment of PC-3 cells with the BRD4 inhibitor JQ1 inhibited GNB1L expression at both mRNA and protein levels (Supplementary Fig. 8c, d). As RB-BRD4 interaction can be enhanced by RB S249/T252 dephosphorylation, we examined whether the expression of GNB1L and activation of CREB were influenced by RB dephosphorylation. We demonstrated that activation (dephosphorylation) of RB by treatment with CDK4/6 inhibitor palbociclib decreased GNB1L expression at protein and mRNA levels and inhibited CREB phosphorylation in a dose-dependent fashion in both PC-3 and C4-2 cell lines (Fig. 5a, b and Supplementary Fig. 8e, f). In agreement with this observation, ChIP-qPCR assay showed that BRD4 binding on chromatin at the GNB1L gene locus was attenuated after palbociclib treatment in both cell l lines (Fig. 5c and Supplementary Fig. 8g). Furthermore, we showed that ectopic expression of WT RB suppressed GNB1L mRNA and protein expression in DU145 cells in which endogenous RB is malfunctional (Fig. 5d, e). Compared to the WT RB, the inhibitory effect of RB S249/T252 phosphorylation-resistant mutant RBΔCDK was much greater as evident by the reduced expression of GNB1L protein and CREB phosphorylation of CREB (Fig. 5d, e).
To verify these findings, we knocked down endogenous RB and performed rescue experiments. We found that RB knockdown-induced upregulation of GNB1L expression and CREB phosphorylation was reversed by restored expression of shRNA-resistant WT RB and this effect was largely enhanced by expression of phosphorylation-resistant mutant RBΔCDK (Fig. 5f, g). In line with these observations, enhanced expression of CDK4 and Cyclin D1 increased GNB1L expression and induced BET inhibitor resistance in C4-2 and LNCaP cells (Fig. 5h–k). In various PCa cell lines with different levels of total and phosphorylated RB, we further demonstrated that PC-3 and DU145 cell lines (in which little/no RB protein or high level of phosphorylated RB was expressed) were much insensitive to BET inhibitors compared to C4-2 and LNCaP (in which low level of phosphorylated RB was expressed) (Supplementary Fig. 8h–j). Notably, these results are consistent with the previous report that DU145 and PC-3 cells were relatively insensitive to JQ1 compared to LNCaP cells32. We also performed immunochemistry (IHC) analysis using a tissue microarray (TMA) containing 115 cores resulting from 51 patients with advanced PCa to investigate the correlation between RB phosphorylation and GNB1L expression. We found that high expression of GNB1L positively correlated with elevated RB S249/T252 phosphorylation and their correlation was statistically significant (Fig. 5l, m). Collectively, our data suggest that RB S249/T252 phosphorylation can promote GNB1L expression in PCa cells in culture and patient samples.
CREB inhibitor 666-15 overcomes JQ1 resistance in RB-deficient PCa cells in vitro and in vivo
Next, we sought to determine whether targeting GPCR-cAMP signaling enables to overcome BET inhibitor resistance in RB-deficient PCa cells. To this end, we treated control or RB KD PC-3 cells with JQ1 or CREB inhibitor 666-15 individually or both prior to MTS assays. We demonstrated that while RB KD cells were highly resistant to JQ1, both control and RB KD cells were equivalently sensitive to 666-15 (Fig. 6a). Importantly, 666-15 co-treatment completely abolished JQ1 resistance in RB-deficient cells (Fig. 6a). Similar results were obtained in C4-2 cells (Supplementary Fig. 8k). Using colony formation assay we further validated that co-treatment of 666-15 completely abolished JQ1 resistance in these two cell lines (Fig. 6b–d and Supplementary Fig. 8l, m).
We also examined the anti-cancer effect of JQ1 and the CREB inhibitor in RB-deficient tumors in vivo. As expected, RB depletion largely enhanced PC-3 tumor growth in mice (Fig. 6e–g). Importantly, RB-deficient PC-3 tumors were resistant to JQ1 in comparison to control tumors (Fig. 6e–g), consistent with the results obtained in PC-3 cells cultured in vitro (Fig. 6a). However, co-administration of 666-15 abolished JQ1 resistance of RB-deficient PC-3 tumors (Fig. 6e–g), suggesting that there was seemingly no obvious drug-drug interaction between these two compounds under these conditions. Analysis of Ki67 and cleaved Caspase-3 IHC in tumor tissues showed that JQ1 treatment only resulted in a minimal effect on proliferation of RB-deficient tumors, but Ki67 expression was largely inhibited by the combined treatment of JQ1 and 666-15 (Fig. 6h, i). Although JQ1 treatment alone had no obvious effect on the expression of cleaved Caspase-3, an indication of apoptosis, the combination of JQ1 and 666-15 drastically induced expression of cleaved Caspase-3 (Fig. 6j, k). In agreement with the result of cell death, there was almost no effect of JQ1 treatment on expression of BCL-2, a known transcription target of CREB and a well-studied anti-apoptotic protein33 (Fig. 6d). In contrast, treatment with 666-15 in combination with JQ1 completely abolished RB depletion-induced upregulation of BCL-2 (Fig. 6d). Together, these data indicate that activation of CREB plays an essential role in mediating BET inhibitor resistance in RB-deficient PCa cells and that inhibition of CREB overcomes BET inhibitor resistance.