CDK7 Inhibition Suppresses Castration-Resistant Prostate Cancer through MED1 Inactivation

Prostate cancer is the most common noncutaneous malignancy and the second leading cause of cancer-related mortality in men of the western world (1). Although effective surgical, radiation, and androgen-ablation therapies exist for clinically localized prostate cancer, progression to metastatic castration-resistant prostate cancer (CRPC) is essentially incurable. The androgen receptor (AR) is a major driver alteration in CRPC (2–4), and most prostate tumors remain addicted to AR-mediated transcription (5). Despite recent advances in antiandrogen therapy, 20% to 40% of patients with metastatic CRPC demonstrate de novo resistance to the FDA-approved drugs abiraterone and enzalutamide, and many of the responders acquire resistance, resulting in a limited survival benefit (6–8). Therefore, the identification and therapeutic targeting of chromatin-associated coactivators or mediators of AR function should be considered as alternative or complementary strategies to block AR transcriptional addiction that defines a large majority of CRPC.

The evolutionarily conserved multi-subunit Mediator complex plays a central role in the regulation of gene transcription by virtue of its ability to functionally bridge gene-specific transcription factors with the RNA polymerase II–associated basal transcription machinery (9). MED1 (also known as TRAP220, PBP, and DRIP205) is a key component of the Mediator complex, and is responsible for targeting and anchoring the complex to both cell type–specific transcription factors and a broad range of nuclear receptors, including AR (10). Despite its pivotal role in transcription, very little is known about the molecular determinants that regulate the formation of a functional MED1–AR complex. In this study, we deciphered that ligand-dependent AR transcriptional signaling is activated through a “phosphoswitch” catalyzed by CDK7. Specifically, CDK7 phosphorylates the transcriptional coactivator MED1 at threonine 1457 to promote the formation of the AR-transcriptional complex. Furthermore, inhibition of CDK7 with the recently developed covalent inhibitor THZ1 (11) attenuates AR signaling and eliminates naïve as well as enzalutamide-refractory AR-addicted prostate cancer cells. Finally, THZ1 demonstrated in vivo efficacy in CRPC xenograft models, suggesting that the CDK7 inhibitors being tested in clinical trials for other malignancies could also be evaluated in refractory CRPC either as monotherapy or in combination with other targeted agents.

To systematically evaluate the significance of MED1 in establishing the ligand-dependent AR cistrome and subsequent transcriptional amplification in prostate cancer cells, we determined by chromatin immunoprecipitation sequencing (ChIP-seq) the genome-wide enrichment of MED1 and AR upon androgen [dihydrotestosterone (DHT)] stimulation in VCaP cells. We utilized VCaP cells because this prostate cancer cell line harbors AR genomic amplification found in more than 85% of patients with metastatic CRPC (4). As we reported previously (12), the average ChIP-seq signal for AR was highly enriched in DHT-treated cells, and this signal was significantly reversed with enzalutamide (Fig. 1A). Interestingly, MED1 also displayed increased chromatin association upon DHT stimulation, which was completely abolished by enzalutamide (Fig. 1A). This apparent trend may be attributed to the corecruitment of AR and MED1 (Supplementary Fig. S1A) to 5,014 regions that are highly enriched in androgen-responsive elements (Supplementary Fig. S1B). To examine the nature of AR- and MED1-bound sites, we first identified transcriptionally active regions using H3K27ac ChIP-seq in VCaP cells (13). H3K27ac was enriched at 93,004 regions, of which 37,974 were defined as enhancers and 1,487 were characterized as superenhancers (SE; Fig. 1B). The latter refers to clusters of enhancers that are densely occupied by transcription factors, cofactors, and chromatin regulators and display a wide variety of chromatin modifications. SEs have been found to play a significant role in the maintenance of cell type–specific gene-expression patterns (14). Interestingly, these SE regions in VCaP cells are positioned within 100 kb of many AR-regulated coding genes and long noncoding RNAs (lncRNA), including canonical AR targets such as TMPRSS2, KLK3, ZBTB16, SLC45A3, PCAT1, and PRNCR1, thus implicating these distal regulatory regions in AR transcriptional output. Our integrative analysis of H3K27ac, AR, and MED1 ChIP-seq data further exemplifies this point, wherein 92% of the SEs, but only 43% of the enhancers, were enriched in AR (AR+H3K27ac), while more strikingly 65% of the SE, compared with 11% of the enhancers, were coenriched for both AR and MED1 (AR+MED1+H3K27ac; Fig. 1C). We also found AR and MED1 enrichment to be significantly higher in the 968 SE regions when compared with the 4,136 enhancer regions (Fig. 1D). These observations together clearly establish that ligand-dependent AR activation results in robust corecruitment of AR and MED1 to SEs, resulting in amplified expression of associated target genes, as exemplified by ChIP-seq tracks shown for FKBP5 (Fig. 1E). We next analyzed AR ChIP-seq data from normal and primary prostate tumors (15) to ascertain whether AR recruitment to SEs is elevated during tumor development. Although no considerable difference was observed in the genome-wide AR-binding levels between the tumor and normal tissues (Supplementary Fig. S1C), the tumor tissues displayed a pronounced enrichment of AR in the SE regions (Fig. 1F). In addition, 70% to 90% of the AR+MED1+H3K27ac SE in VCaP, a cell line originally derived from vertebral metastasis, overlapped with AR-enriched regions in primary tumors (Supplementary Fig. S1D–S1G). These data suggest that a large number of AR-bound SEs in the primary tumor may continue to be enlisted upon progression to metastasis via MED1 association to support SE/promoter long-range interactions required for transcriptional amplification.

Because AR and MED1 interactions are highly localized to the SE regions, we hypothesized that MED1 recruitment to AR-target gene loci is pivotal for AR-transcriptional activity and that modulation of MED1 expression or activity could provide a novel route to resolve AR-dependent transcriptional addiction in prostate cancer. To test this hypothesis, we performed MED1 knockdown in LNCaP and VCaP cells and observed a complete reversal of DHT-induced transcriptional changes without affecting AR expression (Supplementary Fig. S2A–S2D). In addition, to address the effects of MED1 depletion on SE-associated genes, we attempted to integrate the ChIP-seq and RNA-sequencing (RNA-seq) data from VCaP cells. First, we analyzed and found that the magnitude of expression of genes (FPKM values from RNA-seq) associated with AR+MED1+H3K27ac–containing enhancers and SEs was higher than that for the MED1+H3K27ac–containing enhancers and SE genes (Supplementary Fig. S2E). Second, we found a significantly higher expression of genes associated with AR+MED1+H3K27ac–containing SEs compared with AR+MED1+H3K27ac enhancers (P < 1e−05), suggesting that the MED1–AR interaction at SEs elicits much higher transcription of associated genes (Supplementary Fig. S2E). Third, MED1 depletion led to a selective, significant decrease in the expression of genes associated with SEs containing AR+MED1+H3K27ac (P < 6e−133), whereas genes marked with MED1+H3K27ac SE did not display any significant difference in expression (P < 1e−01; Supplementary Fig. S2E). Importantly, gene set enrichment analysis (GSEA; ref. 16) of the differentially expressed genes showed a highly significant negative enrichment of the Hallmark Androgen Response and growth-associated transcriptional signatures (ref. 17; Fig. 1G; Supplementary Fig. S2F). Together, these findings provide clear evidence for the integral role of MED1 in enhancer- and SE-associated AR target gene expression in prostate cancer.

Next, we investigated whether ligand-dependent AR activation leads to MED1 phosphorylation, which might promote its recruitment into the Mediator complex, thereby facilitating chromatin binding (18). Using the Phos-tag immunoblot assay, in which the dinuclear metal complex 1,3-bis[bis(pyridin-2-ylmethyl)amino]propan-2-olato dizinc(II) specifically binds to phosphorylated amino acids and generates a mobility shift on acrylamide gels proportional to the number of incorporated phosphates (19), we observed two additional slow-migrating bands for MED1 in nuclear extracts from cells stimulated with DHT, compared with vehicle control, which were eliminated upon phosphatase [calf intestinal alkaline phosphatase (CIAP)] treatment (Fig. 2A). Phosphorylation of Threonine1032 (T1032) and Threonine1457 (T1457) of MED1 has been shown to promote its association with RNA PolII and Mediator subunits, inducing chromatin looping by increasing recruitment of transcription factors and coactivators on chromatin (18, 20, 21). We confirmed that the DHT-induced MED1 phosphorylation sites are indeed these two previously observed threonine residues by performing HA pulldown followed by phospho-threonine (pThr) immunoblotting and a Phos-tag assay in LNCaP cells overexpressing HA-tagged wild-type (WT), T1032A, T1457A, or double-mutant (DM) MED1 (Fig. 2B and C). Interestingly, Phos-tag analysis of the T1457A mutant displayed a complete loss of phosphorylation on T1032 (Fig. 2C), and both the T1457A and DM MED1 mutants showed impaired association with AR, suggesting that T1457-directed phosphorylation is a major step required for the assembly/stability of the AR–MED1 complex (Fig. 2B). Next, using the pT1457-specific antibody that does not detect the T1457A mutant (Supplementary Fig. S3A), we observed a time-dependent increase in pMED1 levels upon DHT stimulation with a concomitant increase in the expression of phosphoSer81-AR (pAR), an active chromatin-bound form of AR (22, 23), suggesting a ligand-dependent phosphorylation of MED1 in AR-positive prostate cancer (Fig. 2D). In corroboration, treatment with the AR degrader ARD-69 (24) led to a reduction in pMED1 levels in a time-dependent manner in LNCaP cells (Supplementary Fig. S3B). More importantly, post–DHT stimulation, degradation of AR led to a concomitant reduction in pMED1 levels and a subsequent shift of MED1 from the chromatin to the soluble nuclear fraction (Fig. 2E), clearly demonstrating the AR activation–dependent MED1 phosphorylation recruit MED1 to the chromatin. In agreement, AR-null LNCaP cells stimulated with DHT displayed no increase in pMED1 levels (Supplementary Fig. S3C). Next, reciprocal immunoprecipitation for MED1 and AR confirmed their ligand-induced phosphorylation-dependent interaction as shown by the complete loss of interaction when nuclear lysates were treated with CIAP (Fig. 2F). Our finding that pMED1 is a key regulator of AR transcription led us to reason that pMED1 might also be associated with prostate cancer progression defined by increased AR signaling. To test this hypothesis and access the feasibility of pMED1 as a prognostic marker, we analyzed prostate cancer tissue microarrays containing benign adjacent (n = 48), primary (n = 228), and metastatic (n = 96) prostate tumor samples and found a progressive increase in nuclear pMED1 levels from benign, localized, and metastatic sites (P < 3e−12, P < 1e−15, and P < 4e−06; Fig. 2G and H; Supplementary Fig. S3D and S3E). These results agree with our in vitro findings and provide a molecular basis for the use of pMED1 both as a prognostic marker for advanced prostate cancer and as a therapeutically targetable activator of AR signaling for the treatment of prostate cancer.

Next, to identify the kinase involved in the phosphorylation of MED1 in AR-driven prostate cancer, we tested a series of transcriptional cyclin-dependent kinases (CDK), including CDK7, CDK9, CDK12, and CDK13 (25). We reasoned that transcriptional CDKs might be involved in MED1 phosphorylation, as they have been previously shown to regulate processes associated with MED1 functions such as facilitation of gene-specific chromatin looping, coordination of chromatin modification events with preinitiation complex assembly, and regulation of transcriptional elongation by RNA PolII (21, 25, 26). We treated VCaP, LNCaP, and 22RV1 metastatic CRPC cell lines, all of which express AR but harbor three different clinically relevant AR structural variations, namely genomic copy-number amplification, gain-of-function mutation (T877A), and constitutively active splice variant (AR-v7), respectively, with a panel of CDK inhibitors (12, 27). The pan-CDK inhibitor dinaciclib completely abolished pMED1 levels in all three tested cell lines grown in standard culture conditions (Fig. 3A). Contrary to previous reports that suggest ERK involvement in MED1 phosphorylation (18), we found that MEK inhibition with trametinib did not result in loss of pMED1. Treatment with THZ1, a novel highly specific covalent inhibitor of CDK7/12 and CDK7/13 reduced pMED1 levels at baseline and when cells were stimulated with DHT, whereas the CDK9-specific inhibitor LDC had no such effect on pMED1 levels (ref. 11; Fig. 3A; Supplementary Fig. S4A). THZ1 is a phenylaminopyrimidine bearing a cysteine-reactive acrylamide moiety, and, unlike most CDK inhibitors identified before, the THZ1-targeting site on CDK7/12/13 lies outside the kinase domain, providing possible explanations for the increased specificity of inhibition. To identify the specific CDK responsible for the phosphorylation of MED1, using a mixture of four siRNAs, we performed knockdown of CDK7, CDK12, and CDK13 individually, along with CDK9 as a negative control. Interestingly, only CDK7 knockdown led to the loss of DHT-induced pMED1 with a parallel reduction in prostate-specific antigen (PSA) levels, which served as a direct readout of AR transcriptional activity (Fig. 3B; Supplementary Fig. S4B). The reduction in pMED1 was further confirmed using individual siRNA against CDK7 as well as THZ1-R, a non–cysteine reactive analogue of THZ1, as a negative control (ref. 11; Supplementary Fig. S4C). However, CDK7 rescue was able to restore pMED1 to the preknockdown levels (Fig. 3C). In addition, mitogen-induced pMED1 was strongly inhibited upon CDK7 knockdown, whereas knockdown of the genes encoding the previously reported MED1 kinase ERK (18) or AKT (20) displayed only a marginal reduction in pMED1 levels, suggesting that these kinases could potentially be upstream effectors of CDK7 activity (Supplementary Fig. S4D). CDK7 is a serine/threonine kinase that shows a preference for a proline at the +1 position relative to the phosphoacceptor residue. The RNA PolII CTD heptad sequence repeat, 1-YSPTSPS-7, in which CDK7 preferentially phosphorylates Ser5, obeys this rule (25, 28, 29). In addition, CDK7 has been reported to phosphorylate sites containing Ser–Pro or Thr–Pro dipeptides within many DNA-binding transcription factors, including SPT5, E2F1, OCT1, and the tumor suppressor TP53 (25). Interestingly, the evolutionarily conserved T1457 phosphoacceptor residue of MED1 located in its intrinsically disordered region (IDR) appears remarkably similar to the SPT5 site (ref. 30; Supplementary Fig. S4E and S4F). We next sought to determine whether CDK7 can directly phosphorylate T1457 on MED1 by performing in vitro kinase assays. Using purified recombinant components of the CDK-activating kinase (CAK) complex comprising CDK7, Cyclin H, and MAT1, and recombinant GST-tagged MED1 protein (amino acid 1,391–1,490) we observed threonine phosphorylation, specifically at T1457, which was completely inhibited by THZ1 (Fig. 3D). This was further confirmed by assessing the ability of CDK7 to phosphorylate a WT or T1457A-mutant peptide. Although CDK7-mediated phosphorylation of the T1457 WT peptide was inhibited by THZ1 treatment in a dose-dependent manner, CDK7 exhibited no activity toward the T1457A-mutant peptide (Fig. 3E). CDK7 phosphorylation of T1457 was further confirmed by Phos-tag analysis where we observed the loss of slow-migrating pMED1 bands upon CDK7 inhibition by THZ1 or siRNA-mediated knockdown (Supplementary Fig. S4G and S4H). Furthermore, VCaP and LNCaP cells showed a dose- and time-dependent decrease in MED1 pT1457 levels that parallel-reduced phosphorylation at Ser5 of RNA PolII upon THZ1 treatment (Supplementary Fig. S4I and S4J). Intriguingly, AR-negative RWPE and DU145 cells demonstrated no such decrease in pMED1 or pS5-PolII levels even at the highest tested concentration of THZ1 (Supplementary Fig. S4I). Reciprocal immunoprecipitation experiments confirmed an endogenous association between MED1 and members of the CAK complex (25) along with RNA PolII and AR (Fig. 3F). Subsequently, we observed the loss of endogenous MED1–AR complex formation upon THZ1 treatment in VCaP and LNCaP cells (Fig. 3G). We next sought to address the functional significance of phosphorylation of T1457 toward chromatin binding and AR interaction. Compared with the THZ1-sensitive WT MED1, the chromatin fraction was devoid of the T1457A mutant, whereas not only was the T1457D phosphomimic found in the chromatin fraction but its association to chromatin was also insensitive to THZ1 (Supplementary Fig. S5A). Notably, unlike WT MED1, the T1457D phosphomimic retained the interaction with AR in the presence of THZ1 (Fig. 3H). Furthermore, exogenous expression of phosphorylation mutants of MED1 after knockdown of endogenous MED1 demonstrated a reversal of the ligand-activated gene expression and cell proliferation by the T1457D phosphomimic, but not the T1457A mutant, even in the presence of THZ1 (Fig. 3I; Supplementary Fig. S5B and S5C). Finally, compared with WT MED1, overexpression of T1457A displayed a dominant-negative phenotype, whereas T1457D expression led to a further increase in the proliferation of LNCaP cells (Supplementary Fig. S5D). Together, these data identify MED1 T1457 as a novel substrate of CDK7 kinase and demonstrate its significance in AR signaling and prostate cancer cell growth.

Next, to test whether inhibition of CDK7 would result in the loss of MED1 chromatin accessibility, we isolated cytoplasmic, nuclear, and chromatin protein fractions from DHT-stimulated (6 hours) LNCaP cells in the presence or absence of THZ1 (Fig. 4A). As expected, DHT stimulation led to increased phosphorylation and recruitment of MED1 and AR to the chromatin from nuclear and cytoplasmic compartments, respectively, whereas THZ1 triggered a complete loss in the chromatin-bound levels of pMED1/MED1 and confined the entire cellular fraction of MED1 to the nucleus (Fig. 4A). In contrast, the antiandrogen enzalutamide was unable to completely dissociate pMED1 from the chromatin (Fig. 4A). These observations were further confirmed using ChIP-seq analysis of genome-wide MED1 and AR binding in VCaP and LNCaP cells. As expected, the average ChIP-seq signal for MED1 and AR was highly enriched in DHT-treated cells (Fig. 4B), and THZ1 entirely reversed the MED1 enrichment that in turn led to significant derecruitment of AR genome-wide as a consequent disruption of MED1–AR complex in both VCaP and LNCaP cells. Examples of gene tracks of AR-specific genes co-occupied by MED1 and AR on upstream enhancers and SE corroborate the genome-wide findings (Fig. 4C). Next, focusing our evaluation specifically on the MED1- and AR-bound regions (see Supplementary Fig. S6A), DHT-stimulated VCaP cells displayed increased MED1 levels in the AR-occupied enhancers and SEs, defined by H3K27ac, whereas enhancers and SEs devoid of AR, but positive for MED1, did not show any such increase, further establishing ligand-dependent recruitment of MED1 to the AR-bound regulatory regions (Fig. 4D). Notably, AR was enriched at a much higher level at both the enhancers and SEs co-occupied by MED1 (Supplementary Fig. S6B), suggesting the binding of MED1 following AR recruitment to the chromatin stabilizes the AR complex. As expected, THZ1-treated cells demonstrated more efficient loss of MED1 enrichment from the AR-bound enhancers and SEs when compared with the AR-deficient MED1-containing enhancers and SEs (Fig. 4D). Similar observations were made in the LNCaP cells (Supplementary Fig. S6C–S6E). Next, integration of the ChIP-seq data with RNA-seq transcriptome demonstrated that more than 90% of DHT-induced genes common in VCaP and LNCaP cells were associated with AR and MED1 co-occupied enhancers and SEs (Fig. 4E). Interestingly, upon THZ1 treatment, the SE-associated genes displayed a significant reduction in expression (P < 7e−18) compared with enhancer-associated genes (P < 6e−14; Fig. 4E). In addition, GSEA demonstrated a complete reversal in the expression of DHT-induced differentially expressed AR target genes upon THZ1 treatment in VCaP and LNCaP cells (Supplementary Fig. S6F–S6H). Next, we sought to compare the potency of THZ1 and enzalutamide in blocking DHT-induced AR target gene expression. LNCaP and VCaP cells were treated with THZ1 or enzalutamide followed by DHT treatment and analyzed for canonical AR target gene expression. THZ1, even at 100 nmol/L, phenocopied the reduction in DHT-induced gene expression observed in the 10 μmol/L enzalutamide treatment condition (Supplementary Fig. S6I). We also performed CDK7 knockdown to orthogonally validate the specificity of THZ1 in modulating AR transcriptional program and found a complete reversal of DHT-induced transcriptional changes, including loss of hallmark androgen response signature gene expression, without affecting AR protein expression (Supplementary Fig. S7A–S7D). Interestingly, the gene sets that were significantly affected by CDK7 knockdown were analogous to those obtained by MED1 knockdown (Supplementary Fig. S2D). These data establish the CDK7 inhibitor THZ1 as a potent agent for antagonizing AR signaling through its ability to inhibit phosphorylation-dependent recruitment of MED1 to the AR cistrome in prostate cancer cells.

On the basis of our observations that genetic and pharmacologic inhibition of CDK7 affected AR signaling through the disruption of MED1–AR chromatin interaction (Figs. 3 and 4), we hypothesized that AR-dependent prostate cancer cells would be more sensitive to THZ1 treatment. To test this hypothesis, we treated a panel of seven prostate cancer lines and one benign prostate cell line with THZ1. Using a cell viability assay, we found that the five AR-dependent prostate cancer lines tested were highly sensitive to THZ1 with a GI₅₀ concentration ranging between 50 and 170 nmol/L, whereas the remaining three AR-negative cells displayed a GI₅₀ ranging from 400 to 600 nmol/L (Fig. 5A). Interestingly, though all the cell lines expressed equal amounts of CDK7, AR-positive cells displayed higher pMED1 levels (Fig. 5B). In addition, even at a relatively low 50 nmol/L concentration, THZ1 severely inhibited long-term colony formation of AR-positive cells (Fig. 5C; Supplementary Fig. S8A). Importantly, THZ1 treatment induced robust apoptosis in the AR-positive LNCaP and VCaP cells, a response not observed in the AR-negative PC3 and DU145 cells (Fig. 5D and E), suggesting that AR status is a major determinant of THZ1 sensitivity in prostate cancer cells. Subsequently, treatment with THZ1 or siRNA knockdown of MED1 or CDK7 in LNCaP and VCaP cells decreased AR target protein expression, resulting in the induction of apoptosis, and preferential growth inhibition as compared with PC3 and DU145 cells (Supplementary Fig. S8B–S8E). To determine the effect of THZ1 on global transcription, we treated LNCaP, VCaP, and DU145 cells with 100 nmol/L THZ1, a concentration at which substantial inhibition of MED1 phosphorylation was achieved (Supplementary Fig. S4), and performed gene-expression profiling by RNA-seq. Reflecting the observed phenotypic response, treatment with THZ1 led to a massive change in the gene expression in both the AR-positive VCaP and LNCaP cells, (1,325 up/1,203 down in VCaP and 2,612 up/1,695 down in LNCaP), whereas DU145 cells displayed minimal change in gene expression (191 up/190 down; Fig. 5F; Supplementary Fig. S9A). Gene Ontology terms for biological processes (GO:BP) for the 710 genes commonly downregulated between VCaP and LNCaP strongly reiterates the dominant role of THZ1 in regulating key cellular processes, as shown by the enrichment for RNA PolII–dependent transcription, cell cycle, DNA repair, and chromatin organization (Fig. 5G). Moreover, GSEA in VCaP and LNCaP showed negative enrichment of the various hallmark signatures, including androgen response and MYC and E2F targets, known to be deregulated in AR-driven CRPC (refs. 31, 32; Fig. 5H; Supplementary Fig. S9B). Interestingly DU145 cells instead had positive enrichment for MYC and E2F target signatures (Supplementary Fig. S9C), which could explain the reduced sensitivity to THZ1 observed in AR-negative prostate cancer cells. Furthermore, VCaP cells, which harbor the AR-responsive TMPRSS2–ERG gene fusion found in more than 55% of prostate cancers (33), displayed a negative enrichment of ERG signature and a reduction in ERG levels upon THZ1 treatment in a dose- and time-dependent manner (Supplementary Fig. S9D–S9E). This efficient downregulation of ERG, an ETS family transcription factor, is compelling as it is known to drive a unique transitional program and induce DNA damage, invasion, and metastasis, and is difficult to target by small-molecule inhibitors (34). Next, following a study that suggests direct phosphorylation of AR at S515 and transactivation by CDK7 kinase (35), we sought to address whether the observed downregulation of AR signature by THZ1 is due to a reduction in the AR pS515 levels. Unlike pMED1, THZ1 treatment even at 200 nmol/L for 12 hours did not reduce AR pS515 levels (Supplementary Fig. S9F and S9G), suggesting the effect of THZ1 observed in AR signaling is primarily driven by its effect on pMED1 and not on the presumed CDK7-mediated phosphorylation of AR at S515. However, the influence of THZ1 on AR phosphorylation at S515 via long-term treatment with THZ1 could not be ruled out. Together, these results show that THZ1 treatment of AR-positive cells directly interferes with AR and associated transcription factor activity via MED1 inactivation and inhibits key oncogenic pathways required for prostate cancer survival.

One of the current first-line therapies for metastatic CRPC is enzalutamide, a second-generation antiandrogen (36), to which resistance invariably develops (6, 8). The major mechanism of acquired resistance to enzalutamide is restoration of AR signaling, which presents challenges for long-term disease control (37). Therefore, we sought to investigate whether amplification of AR signaling is mediated through CDK7-dependent MED1 phosphorylation in enzalutamide-refractory prostate cancer. Toward this, we first established resistant derivatives of LNCaP, LNCaP-AR, LAPC4, and VCaP cells by long-term exposure to enzalutamide. As expected, the enzalutamide-resistant cells displayed increased PSA expression suggestive of restored AR signaling, surprisingly accompanied by increased pMED1 levels (Fig. 6A). In addition, compared with parental control, increased stability of MED1 was observed in enzalutamide-resistant cells resulting from its hyperphosphorylated state (Fig. 6B). The increased AR stability observed in enzalutamide-resistant cells could be due to increased phosphorylation at S515 and S81 (23), and an indirect consequence of its more stable binding to pMED1. Next, we sought to confirm the role of pT1457 in MED1 stability by conducting additional cycloheximide chase experiments. As expected, compared with the WT MED1, the phospho-dead T1457A-mutant protein exhibited decreased stability and lower half-life (Fig. 6C). Pathologic phosphorylation results from both aberrant activation of kinases and inactivation of phosphatases and is a hallmark of many diseases including cancer. Protein phosphatase 2A (PP2A) is a heterotrimeric serine/threonine phosphatase that prevents pathogenic signaling by dephosphorylating substrates that are key signaling effectors. PP2A is a multiprotein complex that is minimally composed of a scaffolding A subunit (PP2A-A), a catalytic C subunit (PP2A-C), and a regulatory B subunit (PP2A-B). Substrate specificity is dictated by the binding of one of at least 15 regulatory B subunits to form an active PP2A heterotrimer (38). PP2A inactivation has been associated with therapy resistance in multiple malignancies including CRPC (39). Although no change in CDK7 was observed, enzalutamide-resistant cells displayed significantly reduced PP2A (catalytic subunit) expression compared with their parental controls (Fig. 6A). Consistent with this cell-line data, metastatic biopsies (brain and bone) from two enzalutamide-refractory patients displayed increased pMED1 with an associated loss in PP2A expression (Supplementary Fig. S10A). Knockdown of PPP2CA, the catalytic subunit of PP2A, which has been shown to significantly abrogate PP2A enzymatic activity, led to enhanced pMED1 levels (Fig. 6D), confirming PP2A-mediated regulation of MED1 phosphorylation. In support of these findings, treatment with THZ1 led to the loss of pMED1, reduced PSA, increased apoptosis, and growth inhibition in enzalutamide-resistant cells (Fig. 6E and F; Supplementary Fig. S10B). Together, these data demonstrate that enzalutamide-resistant cells with elevated levels of pMED1 are highly susceptible to CDK7 inhibition due in part to restored AR signaling and loss of PP2A.

On the basis of our in vitro data identifying THZ1 as a potent inhibitor of AR-mediated transcription through MED1 inactivation (Figs. 3–5), we sought to study its efficacy in blocking CRPC growth in an in vivo mouse xenograft model. We utilized the VCaP model, as it harbors the TMPRSS2–ERG gene fusion and AR amplification, both of which are frequent molecular aberrations observed in patients with CRPC (3). Although the VCaP cell line was originally derived from a patient with CRPC (40), these cells require constant androgen supplementation for in vitro growth. As a result of this phenomenon, the VCaP tumor xenograft responds to castration initially, but eventually relapses in mouse models (12). VCaP tumor–bearing mice were established and then castrated, leading to tumor regression. Upon regrowth and once the tumors reached their original precastrated size (∼150 mm³), mice were randomized into two groups and treated intraperitoneally with vehicle (n = 8) or THZ1 (10 mg/kg/twice daily; n = 8) for four weeks (Fig. 7A). Treatment with THZ1 led to a dramatic tumor regression compared with vehicle control (Fig. 7B; Supplementary Fig. S11A). In addition, THZ1 treatment was well tolerated with no treatment-related systemic toxicity observed in mice (Supplementary Fig. S11B), suggesting that THZ1 treatment allows a broader therapeutic window in cancer versus normal cells. Tumors from vehicle-treated mice displayed higher PSA expression, a high proliferative activity, and lower apoptosis (Fig. 7C; Supplementary Fig. S11C). In contrast, the vast majority of the residual tumors from THZ1-treated animals demonstrated loss of PSA expression, necrosis, reduced proliferative activity, and increased apoptosis. Importantly, there was a significant reduction in pMED1 levels in THZ1-treated tumors, confirming target engagement in vivo. The plasma PSA level in THZ1-treated mice was significantly reduced compared with vehicle-treated mice (Fig. 7D), suggesting that THZ1 was able to potently inhibit AR signaling in vivo. Interestingly, unlike the tumor regression phenotype observed for the VCaP xenografts, treatment with THZ1 at 10 mg/kg twice daily for four weeks led to tumor growth inhibition (TGI) of AR-negative DU145 xenografts in the castrated mice (n = 8; Supplementary Fig. S11D), which was associated with reduced pMED1 levels, lower Ki-67, and increased apoptosis when compared with vehicle-treated controls (Supplementary Fig. S11D–S11G). Together, these data clearly demonstrate the in vivo efficacy of THZ1 in prostate cancer xenograft models. In addition, the more marked effect of CDK7 inhibition in the castration-resistant VCaP xenograft model compared with the AR-negative DU145 model supports the in vitro results showing increased dependence in AR-driven tumors.

Maintenance of AR signaling is the most frequent resistance mechanism in patients with CRPC that develops after conventional hormone deprivation and newer-generation antiandrogen therapies (41). Genetic and epigenetic events lead to AR amplification, mutation, and alternative splicing, resulting in the observed AR-driven transcriptional addiction seen in CRPC (11). Here, we demonstrate that the CDK7-directed MED1 phosphorylation at T1457 is a key regulator of AR function, and inhibition of the CDK7–MED1 axis results in significant TGI in AR-positive prostate cancer cells in both cell culture and in vivo models. In addition to blocking AR signaling and ligand-activated MED1 recruitment to chromatin, THZ1 also negatively regulates the expression and oncogenic activity of TMPRSS2–ETS gene fusion products. This is very exciting, as the oncogenic ETS transcription factors have been notoriously difficult to target therapeutically (34). Collectively, these data suggest a model in which CDK7 inhibition simultaneously blocks the functional activity and/or expression of multiple oncogenic transcriptional drivers in prostate cancer including AR, ETS, MYC, and E2F, all of which require MED1 for their functions. Moreover, our data elucidating ligand-activated phosphorylation of MED1 by CDK7 and its recruitment to the AR-bound SEs support the recent studies that revealed the phase-separation property of MED1 through its IDRs that result in the formation of high-density assembly of transcription apparatus at SEs (42, 43). The mechanistic basis for the phase separation of MED1 is not well understood, and we hypothesize that this may be due to the phosphorylation of MED1 by CDK7 in response to growth stimuli or nuclear translocation of steroid hormone receptors such as AR. Additional structural, biophysical, and in cellulo phase-separation studies will be needed to test this hypothesis.

CDK7 is a ubiquitously expressed kinase (Supplementary Fig. S12A) that contributes to control of the cell cycle through phosphorylation of other CDKs and plays a crucial role in transcription as part of the transcription factor TFIIH complex (44). A Pan-Cancer analysis for CDK7 in TCGA revealed a comparable level of expression (Supplementary Fig. S12B). However, little is known about CDK7 expression and activity in advanced prostate cancer. Analysis of datasets comprising benign, primary, and metastatic prostate cancer showed no significant change in the expression of CDK7 mRNA between the groups (Supplementary Fig. S12C), which is in line with equal CDK7 protein levels observed in benign and prostate cancer cells (Fig. 5B). Nevertheless, a potential increase in CDK7 activity through T-loop Thr170 phosphorylation (45) by dysregulated MAPK and PI3K/AKT pathway in advanced prostate cancer cells could not be ruled out.

CDK7 inhibitors have demonstrated efficacy in T-cell acute lymphoblastic leukemia, small-cell lung cancer, glioblastoma, and triple-negative breast cancer (11, 46–48). The mechanism for the observed antitumor activity of CDK7-specific inhibitors is suggested to be the attenuation of transcriptional addiction required for the growth and survival of cancer cells. Although CDK7 is involved in normal transcription as an essential component of TFIIH (25), it is not clear why normal cells are less sensitive to CDK7 inhibition. One possible explanation is that cancer cells are addicted to transcription driven by SEs and thus highly dependent on CDK7 activity as exemplified by the identification of a so-called “Achilles' cluster” of genes in TNBC, as well as MYCN-amplified glioblastoma (46, 48). These observations have led to the clinical development of CDK7-specific inhibitors. Currently, SY1365, a covalent CDK7-selective inhibitor, is being assessed in a phase I trial in adult patients with advanced solid cancer as a single agent or in combination with standard-of-care therapies (https://clinicaltrials.gov/ct2/show/NCT03134638). In addition, a second inhibitor, CT7001 (Carrick Therapeutics) is also in phase I expansion for multiple malignancies.

Although there has been much excitement around the development and successful clinical implementation of second-generation therapeutics targeting the androgen axis, such as abiraterone (7, 49, 50) and enzalutamide (8, 51), the responses to these agents are often not durable. Resistance mechanisms to these drugs include those that develop with conventional hormonal treatment (e.g., AR amplification/mutation/intratumor androgen synthesis) as well as novel mechanisms involving compensatory steroid hormone signaling (e.g., estrogen receptor or glucocorticoid receptor; refs. 41, 52, 53). As CDK7 inhibition functions “downstream” of AR itself through MED1 inactivation (Fig. 7E), our data suggest that CDK7-directed therapies may be effective in the context of a wide-range AR-directed resistance mechanisms (54, 55). Our identification of MED1 T1457 as a novel CDK7 substrate, which is essential for driving AR-mediated transcription, makes CDK7 a potential “nononcogene dependency” in advanced prostate cancer. Taken together, the data presented here strongly support the clinical evaluation of CDK7-specific inhibitors as a monotherapy or in combination with second-generation antiandrogens in refractory CRPC.

See Supplementary Materials for additional methods and Supplementary Tables S1–S3 for the RNA-seq and ChIP-seq libraries and the complete list of reagents.

Prostate parental cell lines used in this study were obtained from ATCC, LNCaP-AR cells were a gift from Dr. Charles Sawyers (Memorial Sloan Kettering Cancer Center, New York, NY), and LNCaP AR-KO cells were a gift from Dr. Dean Tang (Roswell Park Cancer Institute, Buffalo, NY; ref. 56). LNCaP, LNCaP-AR, 22RV1, LAPC4, DU145, PC3, and LNCaP AR-KO prostate cancer cell lines were grown in RPMI-1640 (Gibco, 11875093), and the VCaP prostate cancer cell line was grown in DMEM with Glutamax (Gibco, 21013024). The medium was supplemented with 10% of FBS (HYC, SH30910.03) and 1% of penicillin–streptomycin solution (Invitrogen, 15140122). The immortalized benign prostate cell line RWPE-1 was grown in keratinocyte media with supplements (Thermo Fisher Scientific, 17005042). LNCaP-AR and VCaP enzalutamide-resistant cell lines were derived from an enzalutamide-resistant prostate cancer tumor xenograft as described previously (57). LNCaP and LAPC4 enzalutamide-resistant cell lines were generated by culturing the cells in the presence of 20 μmol/L enzalutamide for 3 to 4 months in vitro. Polyclonal pools of enzalutamide-resistant cells were cultured and maintained in media containing enzalutamide throughout. All experiments for enzalutamide-resistant pools were done in the presence of enzalutamide throughout the remainder of all experiments with these cells. The cell lines tested negative for Mycoplasma contamination and were maintained in a humidified incubator at 37°C and 5% CO₂.

The CDK7 inhibitor THZ1 (MedChem Express, HY-80013A/CS-3168), THZ1-R (MedChem Express, HY-19988), LDC000067 (Selleckchem, S7461), trametinib (Selleckchem, S2673), enzalutamide (Selleckchem, S1250), dinaciclib (Selleckchem, S2768), and PROTAC-ARD-69 (gift from Dr. Shaomeng Wang, University of Michigan, Ann Arbor, MI; ref. 24) were dissolved and aliquoted in DMSO (Sigma-Aldrich, D2650). 5α-Dihydrotestosterone (Cerilliant, D073) was dissolved and aliquoted in methanol. EGF (Thermo Fisher Scientific, PHG0311L) was dissolved in water.

The electrophoretic mobility shift in phosphorylated proteins was studied by following the manganese(II)–Phos-tag SDS-PAGE (FUJIFILM Wako Pure Chemical Corporation, 30493521) method as described previously (58). Following the specific treatment, the cells were subjected to lysate preparation and analyzed through Phos-tag SDS-PAGE (see Supplementary Methods). Briefly, a 3.5% (wt/vol) separating gel solution containing 50 μmol/L polyacrylamide-bound Mn2⁺–Phos-tag and 0.5% (wt/vol) agarose and a stacking gel solution of 3.5% (wt/vol) containing 0.5% (wt/vol) agarose was prepared. The gel was run under constant-current conditions at room temperature for 3 to 4 hours. Upon completion of the run the gel was removed and soaked in 100 mL of blotting buffer containing 1 mmol/L EDTA for 10 minutes. Wet tank transfer apparatus equipment was used to transfer the proteins onto polyvinylidene difluoride.

pWZL hygro FLAG-HA TRAP220 wt plasmid was purchased from Addgene (plasmid 17433) and used for site-directed mutagenesis. Primers were designed and site-directed mutagenesis was performed by the QuikChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies, 200521). Three single mutants with the point mutations T1032A, T1457A, and T1457D as well as a double mutant containing T1032A and T1457A mutations were generated using the manufacturer's protocol. All plasmids were verified by DNA sequencing.

Primers used for mutagenesis are:

• MED1 (T1032A) Forward 5′ – GTT CTT CTA ACA GAC CTT TTG CCC CAC CTA CCA GTA C-3′

• MED1 (T1032A) Reverse 5′ – GTA CTG GTA GGT GGG GCA AAA GGT CTG TTA GAA GAA C-3′

• MED1 (T1457A) Forward 5′ – GCC ATA GTA AGT CAC CAG CAT ATG CCC CCC AGA ATC TG-3′

• MED1 (T1457A) Reverse 5′ – CAG ATT CTG GGG GGC ATA TGC TGG TGA CTT ACT ATG GC-3′

• MED1 (T1457D) Forward 5′ – GCC ATA GTA AGT CAC CAG CAT ATG ACC CCC AGA ATC TG-3′

• MED1 (T1457D) Reverse 5′ – CAG ATT CTG GGG GTC ATA TGC TGG TGA CTT ACT ATG GC-3′

For ChIP-seq experiments, LNCaP and VCaP cells were grown in charcoal-stripped serum–containing media for 72 hours, followed by stimulation with 10 nmol/L DHT in the presence or absence of 100 nmol/L THZ1 (6-hour treatment) or 10 μmol/L enzalutamide (12-hour treatment). ChIP was performed using the iDeal ChIP-seq Kit for Transcription Factors (Diagenode, C01010170) according to the manufacturer's protocol. In brief, the cells were cross-linked with 1% formaldehyde in culture medium for 10 minutes at room temperature. Cross-linking was terminated by the addition of 1/10 volume 1.25 mol/L glycine for 5 minutes at room temperature followed by cell lysis and sonication (Bioruptor, Diagenode), resulting in an average chromatin fragment size of 200 bp. Chromatin equivalent to 5 × 10⁶ cells was isolated and incubated with 10 μg antibody overnight at 4°C [MED1, H3-acetyl K27, AR, and IgG (Diagenode)]. ChIP-seq libraries (Supplementary Table S2) were prepared from the ChIP-enriched DNA samples using the TruSeq ChIP Library Preparation Kit (Illumina, IP-202-1012, IP-202-1024) and protocol. In brief, ChIP-enriched DNA (1–10 ng) was converted to blunt-ended fragments. A single A-base was added to fragment ends followed by ligation of Illumina adaptors. The adaptor-modified DNA fragments were enriched by PCR using the Illumina Barcode primers and Phusion DNA polymerase. PCR products were size-selected using 3% NuSieve agarose gels (Lonza) followed by gel extraction using QIAEX II reagents (Qiagen). Libraries were quantified with the 2100 Bioanalyzer (Agilent) and sequenced on the Illumina NextSeq 500 Sequencer (75 nucleotide read length).

In vivo efficacy studies were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania and in compliance with all regulatory standards. Four- to 5-week-old male NOD/SCID gamma mice were procured from Jackson Laboratory (005557). In the VCaP experiment, 2 × 10⁶ VCaP prostate cancer cells suspended in 80 μL of RPMI-1640 with 50% Matrigel (BD Biosciences) were implanted subcutaneously into the dorsal flank of the mice. The animals were castrated when the tumor volumes reached approximately 150 mm³. After castration, once the tumor relapsed to 150 mm³, mice were randomized into two groups, treated with either 10 mg/kg body weight THZ1 or 10% of DMSO in vehicle (D5W) intraperitonially twice daily for 4 weeks. In the DU145 xenograft experiments, animals were castrated 1 week before the subcutaneous implantation of DU145 cells (2 × 10⁶). Treatments of THZ1 at 10 mg/kg or 10% of DMSO in vehicle (D5W) twice daily started for 4 weeks when the tumors reached 150 mm³. Tumor growth in all studies was recorded using digital calipers, and tumor volumes were estimated using the formula (π/6) (L × W²), where L = length and W = width of tumor. Body weight during the study was also monitored. At the end of the treatment regimen the mice were sacrificed and were extracted for further analysis.

RNA-seq and ChIP-seq data have been deposited in the NCBI GEO database with the accession number GSE125245.

D.C. Brady is a co-founder at Merlon Inc. and has ownership interest in US 20150017261 A1. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.U. Rasool, R. Natesan, I.A. Asangani

Development of methodology: R.U. Rasool, R. Natesan, Q. Deng, D.C. Brady, I.A. Asangani

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Natesan, Q. Deng, S. Aras, P. Lal, J.M. Posimo, S. Carskadon, M.M. Pomerantz, J. Siddiqui, L.E. Schwartz, N. Palanisamy, R.B. Den, I.A. Asangani

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.U. Rasool, R. Natesan, Q. Deng, S. Sander Effron, S.C. Baca, J. Siddiqui, L.E. Schwartz, G. Narla, R.B. Den, M.L. Freedman, D.C. Brady, I.A. Asangani

Writing, review, and/or revision of the manuscript: R.U. Rasool, R. Natesan, Q. Deng, S. Aras, S. Sander Effron, J. Siddiqui, L.E. Schwartz, D.J. Lee, G. Narla, R.B. Den, M.L. Freedman, D.C. Brady, I.A. Asangani

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Q. Deng, E. Mitchell-Velasquez, M.M. Pomerantz, D.J. Lee, I.A. Asangani

Study supervision: I.A. Asangani

We thank Drs. A. Chinnaiyan and S. Wang from University of Michigan for the reagent support. Research in the I.A. Asangani laboratory is funded by the Department of Defense Idea Development Award (W81XWH-17-1-0404 to I.A. Asangani). J.M. Posimo is supported by American Cancer Society Postdoctoral Fellowship 131203PF1714701CCG.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.