CDK7 inhibition suppresses aberrant hedgehog pathway and overcomes resistance to smoothened antagonists

Significance

There have been FDA-approved anti-hedgehog drugs for treating hedgehog-driven cancers, which act by antagonizing the upstream component smoothened; however, resistance to smoothened inhibitor (SMOi) drugs have been described. Our previous study demonstrates that epigenetic or transcriptional targeted therapy represents an anti-hedgehog therapeutic strategy that can effectively overcome SMOi resistance. Here we report that transcriptional inhibition through targeting CDK7 suppresses the aberrant hedgehog pathway and growth of hedgehog-driven cancers either responsive or resistant to SMOi drugs, supporting CDK7 inhibition as a promising therapeutic strategy for overcoming SMOi resistance. Since multiple CDK7-targeted drugs have recently entered phase I trial for tumor therapy, our study provides the preclinical rationale for enrolling hedgehog-driven cancers into those clinical trials in the near future.

Abstract

The aberrant hedgehog (Hh) pathway plays important roles in multiple cancer types, therefore serving as a promising drug target. Current clinically available hedgehog-targeted drugs act mostly by antagonizing the upstream component smoothened; however, both primary and acquired resistance to FDA-approved smoothened inhibitor (SMOi) drugs have been described. We have recently demonstrated that the BET inhibitor effectively suppresses SMOi-resistant Hh-driven cancers through antagonizing transcription of GLI1 and GLI2, the core transcriptional factors of Hh pathway, suggesting epigenetic or transcriptional targeted therapy represents an anti-Hh therapeutic strategy that can overcome SMOi resistance. Here we performed an unbiased screening of epigenetic or transcriptional targeted small molecules to test their inhibitory effects on GLI1 and GLI2 transcription or cell viability of Hh-driven tumor lines. THZ1, a covalent inhibitor of cyclin-dependent kinase 7 (CDK7), is identified as the top hit in our screening. We then confirmed that antagonizing CDK7 by either small-molecule inhibitors or the CRISPR-Cas9 approach causes substantial suppression of GLI1 and GLI2 transcription, resulting in effective inhibition of Hh-driven cancers in vitro and in vivo. More importantly, antagonizing CDK7 retains inhibitory activity against Hh-driven cancers with almost all so-far described primary or acquired SMOi resistance. Furthermore, we reveal a synergy between CDK7 inhibition and BET inhibition on antagonizing aberrant Hh pathway and Hh-driven cancers that are either responsive or resistant to SMOi. Our results illustrate transcriptional inhibition through targeting CDK7 as a promising therapeutic strategy for treating Hh-driven cancers, especially those with primary or acquired resistance to SMOi drugs.

The hedgehog (Hh) pathway plays important roles in governing tissue proliferation, patterning, and homeostasis within normal development and lifespan (1). Canonical Hh signaling is initiated upon the binding of a Hh ligand (2, 3) to the cell-surface 12-pass transmembrane receptor patched (PTCH1) (4, 5), which relieves its suppression of the cell-surface 7-pass transmembrane protein smoothened (SMO) (6, 7). SMO activation promotes dissociation of transcription factor GLI2 from its negative regulator SUFU and subsequent translocation from primary cilia into nucleus, where it transactivates expression of transcription factor GLI1 as well as other Hh-dependent target genes including PTCH1 (8).

Aberrant activation of Hh pathway has been implicated in initiation and maintenance of a wide range of tumor types (9). Basal cell carcinoma (BCC), medulloblastoma (MB), and more rarely rhabdomyosarcoma are the major tumor types arising from patients with germline mutations of PTCH1 or SUFU that cause activation of the Hh pathway (Gorlin syndrome) (10–12). These tumors also activate the Hh pathway through somatic genetic changes in various Hh pathway component genes, including loss of function mutations in PTCH1 or SUFU, gain of function mutations in SMO, and genomic amplification of GLI2 or MYCN (13–17). Additionally, noncanonical activation of GLI1 expression have been described to occur in malignant rhabdoid tumors and Ewing sarcoma through loss of chromatin remodeling gene SMARCB1 or transactivation with EWS-FLI fusion oncogene, respectively (18, 19).

There have been extensive efforts by pharmaceutical companies to develop Hh pathway antagonists for cancer therapeutic purposes, mostly focusing on the upstream component SMO due to the discovery of its natural compound inhibitor cyclopamine (20–22). Many cyclopamine-based SMO inhibitor (SMOi) drugs have entered human clinical trials against various cancers with Hh pathway activation (23–26), and two of them (GDC-0449 and LDE225) have already been FDA-approved for treating BCC. However, both primary and acquired resistance to SMOi drugs have been found to occur through SMO or SUFU mutation, GLI2 or MYCN amplification, noncanonical activation of the Hh pathway, or a cancer dependency shift to other signaling pathways (17, 24, 27–31). Accordingly, alternative anti-Hh therapeutic strategies that can overcome the abovementioned drug resistance have been reported, such as a new generation of SMOi and GLI (GLI1 and GLI2) inhibitors (32–35).

Of note, we and others have recently identified that antagonizing GLI expression and the downstream transcriptional output by BET inhibitor (BETi) effectively overcomes most if not all so-far-described SMOi resistance (36, 37), indicating targeting GLI dependency of Hh-driven cancers at the epigenetic or transcriptional level may serve as a promising direction for developing an anti-Hh therapeutic strategy. In this study, we identified THZ1, a covalent small-molecule inhibitor of cyclin-dependent kinase 7 (CDK7), as the top potent inhibitor of both GLI transcription and cell viability of Hh-driven mouse medulloblastoma cells through an unbiased screening of a collection of epigenetic or transcriptional targeted small molecules. CDK7 is a member of the cyclin-dependent kinase protein family. In addition to cell cycle regulation, CDK7 also plays critical roles in RNA polymerase II (RNAPII)-mediated gene transcription (38–41). It controls transcription initiation or elongation through direct or indirect phosphorylation of serine 2 (S2), serine 5 (S5), and serine 7 (S7) at the C-terminal domain (CTD) of RNAPII (42, 43). CDK7 blockade by the selective covalent inhibitor THZ1 has been recently demonstrated to effectively treat multiple cancer types in preclinical models through targeting their oncogenic transcriptional dependencies, such as T-cell acute lymphoblastic leukemia (44), small-cell lung carcinoma (45), neuroblastoma (46), triple-negative breast cancer (47), esophageal squamous cell carcinoma (48), diffuse intrinsic pontine glioma (49), et al. However, the role of CDK7 in the Hh pathway and the efficacy of CDK7 inhibition against Hh-driven cancers remain unclear so far.

Results

Epigenetic/Transcriptional Targeted Compound Screening Identifies THZ1 as a Potent Inhibitor of GLI Transcription and Growth of Ptch1-Deficient Mouse Medulloblastoma.

To search for an anti-Hh strategy that acts by antagonizing GLI transcription, we performed an unbiased screening of a collection of 94 epigenetic or transcriptional targeted small-molecule compounds by measuring their effects (two doses, 0.1 and 1 μM) on tumor cell viability of SMB21 and SMB56, two recently reported Hh-driven mouse MB cell lines derived from spontaneous medulloblastoma of Ptch1^+/− mice (29) (SI Appendix, Fig. S1 A–D). Then, we performed validation tests by remeasuring their inhibition against the two mouse MB lines, as well as normal mouse astrocyte cells for the top 10 ranked compounds, which contained three HDAC inhibitors (HC toxin, panobinostat, and Trichostatin A), two BET bromodomain inhibitors (JQ1 and OTX015), two histone methyltransferase (HMT) inhibitors (Chaetocin and MI-2), two CDK inhibitors (Flavopiridol and THZ1), and a chemotherapy drug with inhibitory activity against Gadd45a-mediated DNA demethylation (Gemcitabine) (50). The validation results demonstrated that all of the other nine tested compounds exhibited potent and selective suppression against SMB21 and SMB56 cells except the pan-CDK inhibitor flavopiridol (SI Appendix, Fig. S1B). Moreover, we measured their effects on GLI transcription, and our results revealed that CDK7 inhibitor THZ1 and HDAC inhibitor panobinostat exhibited the most potent inhibitory effects in both SMB21 and SMB56 lines among the tested compounds (SI Appendix, Fig. S1E). Notably, HDAC inhibitors have been previously reported to effectively treat Hh-driven malignancy through targeting the transcriptional activity of Gli1 (51). Therefore, we chose to further study the role of CDK7 in the normal and the aberrant Hh pathway as well as CDK7 inhibition therapy against Hh-driven cancer.

First, we measured the efficacy of THZ1 in parallel with BETi (JQ1) and SMOi (GDC-0449 and LDE225) in inhibiting an aberrant Hh pathway by using an expanded panel of Ptch1-deficient mouse MB lines derived from spontaneous medulloblastomas of Ptch^+/− (SMB21, SMB55, and SMB56) or Ptch^+/−, Trp53^−/− (SmoWT-MB) mice (29, 33). As expected, the deregulated Hh pathway in these mouse MB lines was sensitive to SMOi and BETi as presented by the decrease of Gli1 and Gli2 levels upon treatment (Fig. 1A). THZ1 induced similar if not stronger down-regulation of Gli1 and Gli2 expression at both mRNA and protein levels in all tested mouse MB lines (Fig. 1 A and B). In contrast, the reversible THZ1 analog THZ1-R, which cannot form a covalent bond with CDK7 (44), did not show any inhibitory effect (Fig. 1A). Time-course tracking of mRNA levels revealed a drastic reduction of Gli1 and Gli2 starting from as early as 2 h post-THZ1 treatment, suggesting a direct role of CDK7 in regulating GLI transcription (SI Appendix, Fig. S1F).

Fig. 1.

THZ1 inhibits Gli transcription and growth of Ptch1-deficient mouse medulloblastoma; (A) qRT-PCR of Gli1 and Gli2 in four Ptch1-deficient mouse MB lines treated as indicated for 8 h. Data are shown as mean ±SD. (B) Immunoblot detecting Gli1 expression in cell extracts from mouse MB cells treated with 0.1 μM THZ1 for 24 h. Anti–β-tubulin immunoblots are shown as loading control. (C) Dose–response curves of mouse MB lines and control cells in response to THZ1. Cells were plated in 96-well plate in triplicate and treated with THZ1 at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD. (D) Mouse MB lines were treated with THZ1 at indicated concentrations. Cell viabilities were measured at day-0/1/2/3 posttreatment and normalized to day-0 value. Data are shown as mean ±SD. (E and F) Cell proliferation and apoptosis analyses of mouse MB lines treated as indicated by EdU incorporation or Annexin-V staining FACS assay. Percentages of EdU+ (E) or Annexin-V+ (F) cells are presented.

Next, we examined the dose-dependent inhibitory effect of THZ1 on cell viability of these Hh-driven mouse MB lines. NIH 3T3, Sufu^−/− MEF, as well as primary mouse neural stem cells, astrocytes, granule neuron precursors (GNPs), and differentiated granule neurons were used as normal controls. In line with our screening results, THZ1 resulted in much more potent inhibition of cell viability in mouse MB lines (IC50 ranging from ∼15 to 25 nM) compared with control cells (IC50 ranging from ∼80 to 400 nM; Fig. 1C). The inhibitory effects of THZ1 at 0.1 μM were equivalent to those of JQ1 at 1 μM in these Hh-driven MB lines; whereas, THZ1-R at 0.1 μM had almost no inhibition at all (SI Appendix, Fig. S1G). Time-course tracking of tumor cell growth in response to THZ1 in vitro revealed a cytocidal effect (Fig. 1D), which was accompanied by reduction of cell proliferation (Fig. 1E and SI Appendix, Fig. S2A) and induction of cell apoptosis (Fig. 1F and SI Appendix, Fig. S2B).

THZ1 Inhibits Gli Transcription in Hh-Activated Normal Cells.

To better understand the role of Cdk7 in regulating the Hh pathway, we measured the inhibitory effects of THZ1 on the Hh pathway transcriptional output in Hh-activated normal cell models. In mouse NIH 3T3 cells, smoothened agonist (SAG) compound was used to stimulate the activation of the Hh pathway as presented by up-regulation of Gli1, Gli2, and Ptch1 (SI Appendix, Fig. S3A). As expected, the induction of these Hh target genes was significantly inhibited by SMOi (GDC-0449 and LDE225) and BETi (JQ1) (SI Appendix, Fig. S3A). Even though THZ1 also exhibited a dose-dependent inhibitory effect on transcriptional of Hh target genes in SAG-treated NIH 3T3 cells, much higher concentrations (∼1 μM) of THZ1 was needed to reach similar potency as SMOi or BETi (SI Appendix, Fig. S3 A and B). Moreover, we measured the inhibition of THZ1 on the Hh pathway in primary mouse GNP cells, which also rely on SAG to maintain its Hh pathway activation in culture. Even though both 0.1 and 1 μM of THZ1 exhibited marked inhibition on transcription of Gli1, Gli2, and Ptch1, its inhibition on Gli1 protein level was much less effective compared with SMOi or BETi (SI Appendix, Fig. S3 C and D).

Since CDK7 controls RNAPII-mediated gene transcription, CDK7 inhibition may overcome SMOi resistance through directly antagonizing transcription of GLI and other downstream Hh target genes as BET inhibition. Therefore, we further tested the inhibitory effect of THZ1 on the Hh pathway in Sufu^−/− MEF and NIH 3T3-GLI2-deltaN cells, which display constitutive Hh pathway activation downstream of SMO through loss of Sufu (SI Appendix, Fig. S3E) or stably ectopic expression of an N-terminal truncated active form of human GLI2 (SI Appendix, Fig. S3 F and G). In both cell lines, THZ1 exhibited similar if not stronger suppressive effects on transcription of Gli1, Gli2, and Ptch1 as JQ1 when they were both used at 1 μM, whereas SMOi or THZ1-R had very little or no inhibitory effects (SI Appendix, Fig. S3 H–K).

THZ1 Suppresses Hh Pathway and Hh-Driven Medulloblastoma through Targeting CDK7.

To confirm the on-target inhibitory effect of THZ1 against Cdk7, we measured the changes of Cdk7-associated RNAPII CTD phosphorylation in Hh-activating cells in response to THZ1. As shown in Fig. 2A, THZ1 treatment resulted in dose-dependent reduction of phosphorylation at S2, S5, and S7 of RNAPII CTD in both Hh-activating normal (GNP and Sufu^−/− MEF) and tumor (SMB21 and SMB56) cells. Of note, Cdk7-associated RNAPII CTD phosphorylation of GNP or Sufu^−/− MEF cells exhibited less sensitivity to THZ1 than those of SMB21 or SMB56 cells (Fig. 2A). The matched dose-responsive patterns of THZ1 on inhibiting RNAPII CTD phosphorylation and Gli transcription in Hh-activating cells supports that THZ1 inhibits Gli transcription through targeting CDK7.

Fig. 2.

THZ1 targets CDK7 to inhibit Gli transcription and growth of the Hh-driven mouse medulloblastoma. (A) Immunoblot-detecting phosphorylation levels at Ser2, Ser5, and Ser7 of RNAPII CTD as well as total RNAPII from GNP, Sufu^−/− MEF, SMB21, or SMB56 cells treated with THZ1 at indicated concentrations for 8 h. Anti–β-tubulin immunoblot are shown as loading control. (B) qRT-PCR of Gli1, Gli2 in SMB21 (Top) or SMB56 cells (Bottom) stably expressing Cas9 together with sgCdk7-1, sgCdk7-2, sgLacZ, or sgGFP. Data are shown as mean ±SD. (C) Immunoblot detecting Gli1 and Cdk7 in cell extracts from SMB21 (Top) or SMB56 cells (Bottom) stably expressing Cas9 together with sgCdk7-1, sgCdk7-2, sgLacZ, or sgGFP. Anti–β-tubulin immunoblot are shown as loading control. (D) In vitro growth of SMB21 (Top) or SMB56 cells (Bottom) stably expressing Cas9 together with sgCdk7-1, sgCdk7-2, sgLacZ, or sgGFP were measured at day-0/2/4 and normalized to day-0 value. Data are shown as mean ±SD.

To further prove that THZ1 suppresses Hh pathway through targeting CDK7, we tested if genetically targeting Cdk7 through RNAi and CRISPR-Cas9 approaches could phenocopy the inhibitory effect of THZ1 on the Hh pathway. Indeed, expression of Gli1, Gli2, and Ptch1 were significantly down-regulated upon loss of Cdk7 by either RNAi or CRISPR-Cas9 in Sufu^−/− MEF cells, indicating that Cdk7 was necessary for Gli transcription and downstream Hh transcriptional output (SI Appendix, Fig. S4 A–D). Similarly, knockout of Cdk7 by CRISPR-Cas9 also resulted in significant reduction of Gli transcription and tumor cell growth in cultured SMB21 and SMB56 lines, indicating that Cdk7 was essential for aberrant Hh pathway and in vitro growth of Hh-driven mouse MB lines and therefore a valid anti-Hh therapeutic target (Fig. 2 B–D).

THZ1 Inhibits SMOi-Resistant Hh-Driven Mouse Medulloblastoma.

Given the retained anti-Hh activity of THZ1 in Sufu^−/− MEF and NIH 3T3-GLI2-deltaN cells, we further investigated its efficacy in treating Hh-driven mouse MB lines with SMOi resistance. We first examined the inhibitory effects of THZ1 on Gli transcription and tumor cell viability in a Hh-driven mouse MB line carrying a D477G mutation in Smo (SmoD477G-MB), which was derived from a flank tumor of SmoWT-MB cells that was treated with SMO inhibitor until the emergence of acquired resistance (33). As expected, SmoD477G-MB cells were much less sensitive to GDC-0449 compared with SmoWT-MB (Fig. 3 A and B). Nonetheless, they exhibited similar sensitivity to THZ1 or JQ1, indicating CDK7 inhibition could overcome Smo-mutant mediated acquired SMOi resistance as BET inhibition (Fig. 3 A and B). Moreover, we generated SMB21 or SMB56 cells stably expressing shRNA against Sufu (SMB21-shSufu or SMB56-shSufu) or GLI2-deltaN (SMB21-GLI2-deltaN or SMB56-GLI2-deltaN) to mimic two other types of SMOi resistance. These engineered tumor lines were confirmed to develop resistance to GDC-0449, but still retain similar sensitivity to JQ1 (Fig. 3 C–F and SI Appendix, Fig. S4 E–H). Again, we observed similar suppressive effects of THZ1 on both Gli transcription and tumor cell viability between these engineered tumor lines and their parental lines (Fig. 3 C–F and SI Appendix, Fig. S4 E–H). Furthermore, a recent study identified a new mechanism of SMOi resistance mediated by alternative activation of RAS–MAPK pathway that leads to sustained tumor growth and enhanced metastatic behavior (29). To build such model, we cultured SmoWT-MB cells with GF-containing medium (normal medium plus EGF and bFGF) for 3 wk. Consistent with the recent study, the Hh pathway was shutoff in SmoWT-MB-GF line (SI Appendix, Fig. S4 I and J), and it became much less sensitive to GDC-0449, but much more sensitive to a MEK inhibitor PD325901 (SI Appendix, Fig. S4K), confirming that its cancer dependency had shifted from the Hh pathway to the MAPK pathway. Importantly, THZ1 still exhibited potent inhibitory effect on tumor cell growth of the SmoWT-MB-GF line, suggesting the induced RAS-MAPK dependency still relied on CDK7 (SI Appendix, Fig. S4K).

Fig. 3.

THZ1 suppresses Gli transcription and growth of Hh-driven mouse medulloblastoma with SMOi resistance. (A) qRT-PCR of Gli1, Gli2 in SmoWT-MB or SmoD477G-MB cells treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 8 h. Data are shown as mean ±SD. (B) SmoWT-MB and SmoD477G-MB cells were treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD. (C) qRT-PCR of Gli1, Gli2 in SMB21-Mock, SMB21 cells stably expressing shCtrl (SMB21-shCtrl), or shSufu (SMB21-shSufu) treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 8 h. Data are shown as mean ±SD. (D) SMB21-Mock, SMB21-shCtrl, and SMB21-shSufu cells were treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD. (E) qRT-PCR of Gli1, Gli2 in SMB21-Mock, SMB21 cells stably expressing empty vector (SMB21-EV) or GLI2-deltaN (SMB21-GLI2-deltaN) treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 8 h. Data are shown as mean ±SD. (F) SMB21-Mock, SMB21-EV, and SMB21-GLI2-deltaN cells were treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD.

To assess THZ1-induced global changes of transcriptome of Hh-driven cancer, we performed RNAseq analysis of SMB56 or SMB56-shSufu cells treated with THZ1 or DMSO for 8 h. GDC-0449-treated SMB56 cells were included as control (52). Compared with GDC-0449 treatment in SMB56 cells, THZ1 induced a robust transcriptional down-regulation in both SMB56 and SMB56-shSufu cells (SI Appendix, Fig. S5A). When we compared the top 2,000 significantly down-regulated genes [logtwofold change (FC) ≤ −1, false discovery rate (FDR) ≤ 0.05, ranked by FC] between THZ1-treated SMB56 and SMB56-shSufu cells, we found they had ∼70% in common, suggesting CDK7 inhibition targeted very similar gene signatures between these two SMOi-responsive and -resistant Hh-driven tumor lines (SI Appendix, Fig. S5B). Gene ontology and pathway analyses of the shared 1,343 significantly down-regulated genes revealed cell cycle and DNA repair were the top two enriched biological processes and Kyoto Encyclopedia of Genes and Genomes pathways (SI Appendix, Fig. S5 C and D). Gene set enrichment analysis further demonstrated that gene sets containing Hh-dependent genes of SMB56 or SmoWT-MB (37) (significantly down-regulated by GDC-0449, log2 FC ≤ −1, FDR ≤ 0.05) as well as SmoM2-driven MB (significantly up-regulated genes of SmoM2-driven mouse MB tissue versus control cerebellum, log2 FC ≥ 1, FDR ≤ 0.05, from GEO dataset GSE104623) were significantly enriched in THZ1-down-regulated genes of both tumor lines, confirming its preferential targeting of the Hh pathway transcriptional output (SI Appendix, Fig. S5 E and F).

Up-Regulation of ABC Transporter Contributes to Acquired THZ1 Resistance in Hh-Driven Cancer.

To investigate the potential emergence of THZ1 resistance in Hh-driven cancer, we tried to build a THZ1-resistant Hh-driven mouse MB line by culturing SMB56 cells in increasingly sublethal doses of the compound. After 4 mo of culturing, we successfully generated SMB56R cells that exhibited almost 10-fold increase of THZ1 IC50 compared with the parental SMB56 cells (SI Appendix, Fig. S6A). Up to 0.1 μM of THZ1 had very little inhibitory effects on SMB56R cells as presented by almost unchanged levels of RNAPII phosphorylation, Gli expression, cell proliferation, or apoptosis in THZ1-treated cells (SI Appendix, Fig. S6 B–E). In contrast, these cells retained similar sensitivity to JQ1 or GDC-0449 as the parental SMB56 cells (SI Appendix, Fig. S6 B and F). A recent study has found that induction of ATP-binding cassette (ABC) transporters could cause THZ1 resistance in neuroblastoma and small-cell lung cancer cell lines (53), so we measured the mRNA levels of three major mouse drug resistance–associated ABC transporter genes (Abcb1a, Abcg2, Abcc1) in SMB56R and the parental SMB56 cells. Our data showed that only Abcg2 was dramatically up-regulated in SMB56R cells (SI Appendix, Fig. S6G) and Abcg2 inhibitor Ko-143 was able to restore its THZ1 sensitivity (SI Appendix, Fig. S6 H–J), thus proving the existence of a similar THZ1 resistance mechanism mediated by up-regulated ABC transporter in SMB56R cells. Furthermore, we compared the mRNA levels of Abcb1a, Abcg2, and Abcc1 between THZ1-sensitive Hh-driven mouse MB cells and THZ1-insensitive normal control cells and no significant differences were observed, suggesting these three ABC transporters did not contribute to the primary THZ1 resistance in our tested normal control cells (SI Appendix, Fig. S7A).

THZ1 Works Synergistically with JQ1 on Inhibiting Gli Transcription and the Growth of Hh-Driven Mouse Medulloblastoma.

Combination therapy has been proven as an effective approach to overcoming drug resistance. Given that SMB56R cells retained similar sensitivity to GDC-0449 or JQ1, we first tested the combinatory effects of THZ1 with GDC-0449 or JQ1 on inhibiting the cell viability of SmoWT-MB cells. Synergistic inhibitory effect (combination index < 1) was detected between THZ1 and JQ1 but not THZ1 and GDC-0449 (Fig. 4A and SI Appendix, Fig. S7B). We further tested the combinatory effect of THZ1 and JQ1 in three other Hh-driven mouse MB lines with or without SMOi resistance (SmoD477G-MB, SMB21, and SMB21-shSufu). Our results showed that THZ1 also worked synergistically with JQ1 in suppressing the cell viability of these three lines (Fig. 4 B–D). Their combination resulted in stronger down-regulation of Gli transcription (Fig. 4 E and F and SI Appendix, Fig. S7C), more reduction of cell proliferation (SI Appendix, Fig. S8 A and B), and higher induction of cell apoptosis (SI Appendix, Fig. S8 C and D), eventually leading to more severe cytocidal effect (SI Appendix, Fig. S7 D and E) in SMB21 and SMB21-shSufu cells.

Fig. 4.

THZ1 and JQ1 works synergistically together to suppress Gli transcription and growth of Hh-driven mouse medulloblastoma. (A–D) Cell viability and synergy measurement in combinatory drug treatments. (Top) SmoWT-MB (A), SmoD477G-MB (B), SMB21 (C), SMB21-shSufu (D) cells were treated with THZ1 and JQ1 individually or in combination at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD. (Bottom) CI value of each drug combination condition was calculated using CalcuSyn software and a CI value of less than 1 indicates synergy. qRT-PCR of Gli1, Gli2 in SMB21 (E) or SMB21-shSufu (F) cells treated with THZ1 and JQ1 individually or in combination at indicated concentrations for 8 h. Data are shown as mean ±SD.

CDK7 Inhibition Suppresses Hh-Driven Mouse Medulloblastoma In Vivo.

To further demonstrate the therapeutic potential of CDK7 inhibition in treating Hh-driven cancers, we investigated its efficacy against Hh-driven mouse MB allografts in vivo. First, we compared the growth of flank allografts derived from Cdk7-knockout SMB21 or SMB56 cells with the control ones. Our results demonstrated the loss of Cdk7 led to significant tumor growth defect of both flank allograft models of Hh-driven mouse MB in vivo (Fig. 5 A and B). Of note, Cdk7-knockout allografts somehow regained their growth potential ∼3 wk post allografting (Fig. 5 A and B). To understand why that happened, we obtained the Cdk7-knockout allograft tissues and found they had similar Cdk7 protein levels as the control ones (SI Appendix, Fig. S9A). Moreover, they also had significantly detectable Flag-Cas9 protein levels, suggesting the growing tumor cells in Cdk7-knockout allografts were probably the ones that failed to knock out Cdk7 due to in-frame insertion/deletion after nonhomologous end joining (NHEJ) repair process (SI Appendix, Fig. S9A).

Fig. 5.

CDK7 inhibition effectively suppresses Gli transcription and growth of SMOi-responsive or -resistant Hh-driven mouse medulloblastoma in vivo. Tumor growth of allografts established by injecting SMB21 (A) or SMB56 cells (B) stably expressing Cas9 together with sgCdk7-1, sgCdk7-2, sgLacZ, or sgGFP as indicated into hind flanks of nude mice (five mice with 10 allografted tumors in each group) was assessed by caliper measurement. Data are shown as mean ±SEM. P values were determined by two-way ANOVA. Nude mice with SMB21 (C) or SMB21-shSufu (D) flank allografts (each mouse carried two allografts) were treated with THZ1 at 10 mg/kg or DMSO twice daily by i.v. injection. Tumor growth assessed by caliper measurement is presented as mean ±SEM. P value was determined by two-way ANOVA. (E) Nude mice with SMB21-shSufu intracranial allografts were treated with THZ1 at 10 mg/kg (n = 8) or DMSO (n = 6) twice daily by i.v. injection as indicated. Mice survival was monitored and Kaplan–Meier survival curve was presented. P value was determined by log-rank test (Mantel–Cox test). (F) Nude mice with SMB21-shSufu flank allografts (each mouse carried two allografts) were treated with THZ1 (10 mg/kg, twice a day, i.v.) and JQ1 (25 mg/kg, twice a day, i.p.) individually or in combination as indicated. Tumor growth assessed by caliper measurement is presented as mean ±SEM. P values were determined by two-way ANOVA. qRT-PCR of Gli1, Gli2 in SMB21 (G) or SMB21-shSufu (H) allografts treated with THZ1 or DMSO (n = 4). Data are shown as mean ±SEM. (I) qRT-PCR of Gli1, Gli2 in SMB21-shSufu flank allografts treated with THZ1 and JQ1 individually or in combination (n = 4). Data are shown as mean ±SEM.

Next, we treated flank allografts of SMB21 or SMB21-shSufu cells as well as intracranial allografts of SMB21-shSufu cells with either THZ1 (10 mg/kg, twice a day, i.v.) or vehicle control. Our results showed that THZ1 treatment caused significant reduction of flank tumor growth and increase in overall survival of intracranial allografted mice without obviously affecting mice body weight (Fig. 5 C–E and SI Appendix, Fig. S9 B–D). Moreover, we assessed the in vivo efficacy of the combinatory treatment of CDK7i and BETi against flank allografts of SMB21-shSufu cells. Our results revealed that the combination of THZ1 and JQ1 resulted in significantly more reduction of flank tumor growth compared with either of them individually (Fig. 5F). Nevertheless, we also noticed that the combination caused reduction of mice body weight starting from 6 d posttreatment, suggesting drug doses and treatment logistics should be more carefully tested in future to minimize toxicity (SI Appendix, Fig. S9E). Furthermore, we measured the inhibition of THZ1 on the aberrant Hh pathway of these mouse MB allografts in vivo. Both Gli1 and Gli2 mRNA levels were found to be down-regulated in THZ1-treated flank allografts of SMB21 and SMB21-shSufu cells, proving an on-target anti-Hh activity of THZ1 in vivo (Fig. 5 G and H). THZ1 and JQ1 combination also resulted in stronger reduction of transcription of Gli1 and Gli2 in vivo (Fig. 5I). Together, these results revealed the in vivo efficacy of CDK7 inhibition in targeting the aberrant Hh pathway and Hh-driven tumors, especially those with SMOi resistance.

THZ1 Inhibits Hh-Driven Human Cancers with SMOi Resistance.

Hh-driven mouse MB cell lines and allografts were used as representative Hh-driven cancer models in our study here. Previously, we chose to further confirm the inhibitory effects of THZ1 in a panel of SMOi-resistant Hh-driven human tumor lines, including RCMB018 (a patient-derived SHH-subtype MB line with MYCN amplification) (37), A673 (a patient-derived Ewing sarcoma line with EWS/FLI1 fusion gene), and ATRT-03 (a patient-derived atypical teratoid rhabdoid tumor line with SMARCB1 loss). Our data showed that GLI transcription and cell viability were markedly decreased in response to THZ1 or JQ1 in all tested tumor lines, whereas only subtle or no inhibitory effects were detected with GDC-0449 treatment (Fig. 6 A–D and SI Appendix, Fig. S10 A and B). MYCN expression in the MYCN-amplified RCMB018 cells was also markedly inhibited by THZ1 or JQ1 (SI Appendix, Fig. S10B). Moreover, we examined the combinatory therapeutic effect of CDK7i and BETi against A673 and ATRT-03 cells. Combination index analyses revealed a robust synergy between THZ1 and JQ1 in treating A673 and ATRT-03 (Fig. 6 E and F). As in mouse MB cells, the combination also resulted in stronger down-regulation of GLI1 transcription (Fig. 6 G and H), more reduction of cell proliferation (SI Appendix, Fig. S10C), higher induction of cell apoptosis (SI Appendix, Fig. S10D), leading to more severe cytocidal effects (SI Appendix, Fig. S10 E and F). Together these data demonstrated that THZ1 alone and its synergistic combination with JQ1 also worked effectively against Hh-driven human tumors with SMOi resistance.

Fig. 6.

THZ1 suppresses GLI transcription and growth of Hh-driven human tumors with SMOi resistance. A673 (A) or ATRT-03 (B) cells were treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD. qRT-PCR of GLI1 in A673 (C) or ATRT-03 (D) cells treated with THZ1, JQ1, or GDC-0449 at indicated concentrations for 8 h. Data are shown as mean ±SD. A673 (E) and ATRT-03 (F) cells were treated with THZ1 and JQ1 individually or in combination at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD on the top. CI value of each drug combination condition was calculated using CalcuSyn software and presented on the bottom. A CI value of less than 1 indicates synergy. qRT-PCR of GLI1 in A673 (G) and ATRT-03 (H) cells treated with THZ1 and JQ1 individually or in combination at indicated concentrations for 8 h. Data are shown as mean ±SD.

CDK7 Inhibitor Drug Works Effectively Against GLI Transcription and Growth of Hh-Driven Tumors Alone or in Synergy with BET Inhibitor Drug.

Both CDK7 inhibitor (CT7001) and BET inhibitor (OTX015) drugs have entered early-phase clinical trials of treating cancer, so we examined if these clinical trial drugs would also work effectively as THZ1 and JQ1 against GLI transcription and growth of Hh-driven tumors, especially those with SMOi resistance. We examined the dose-dependent inhibitory effect of CT7001 on cell viability of the Hh-driven mouse and human tumor lines (SmoWT-MB, SmoD477G-MB, A673, and ATRT-03). Primary mouse astrocytes and granule neurons were used as normal controls. Our results showed that CT7001 exhibited more potent inhibitory effect against Hh-driven tumor lines (IC50 ranging from ∼300 to 800 nM) compared with normal control neural cells (IC50 ranging from ∼1,300 to 2,100 nM) (Fig. 7A). CT7001 treatment also induced dose-dependent reduction of phosphorylation at S2, S5, and S7 of RNAPII CTD (Fig. 7B) as well as GLI transcription (Fig. 7 C–F). Moreover, we demonstrated that CT7001 worked synergistically with OTX015 to inhibit cell viability (Fig. 7 G–J) and GLI transcription (Fig. 7 K–N) of these Hh-driven tumor lines. These results further supported the future clinical application of CDK7 inhibition against Hh-driven cancers alone or in combination with BET inhibition.

Fig. 7.

CDK7 inhibitor drug suppresses GLI transcription and growth of Hh-driven tumors alone or in synergy with BET inhibitor drug. (A) Dose–response curves of Hh-driven tumor lines and control cells in response to clinical trial used CDK7i drug CT7001. Cells were plated in 96-well plate in triplicate and treated with CT7001 at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD. (B) Immunoblot detecting phosphorylation levels at Ser2, Ser5, Ser7 of RNAPII C-terminal domain as well as Cdk7 and total RNAPII from Hh-driven cancer cell lines treated with CT7001 at indicated concentrations for 8 h. An anti–β-tubulin immunoblot is shown as loading control. qRT-PCR of Gli1 and Gli2 in SmoWT-MB (C) or SmoD477G-MB (D) cells treated with DMSO or CT7001 at indicated concentrations for 8 h. Data are shown as mean ±SD. qRT-PCR of GLI1 in A673 (E) or ATRT-03 (F) cells treated with DMSO or CT7001 at indicated concentrations for 8 h. Data are shown as mean ±SD. SmoWT-MB (G), SmoD477G-MB (H), A673 (I), or ATRT-03 (J) cells were treated with CT7001 and OTX015 individually or in combination at indicated concentrations for 72 h before cell viabilities were assessed. Data are shown as mean ±SD on the top. CI value of each drug combination condition was calculated using CalcuSyn software and presented on the bottom. A CI value of less than 1 indicates synergy. qRT-PCR of Gli1 and Gli2 in SmoWT-MB (K) or SmoD477G-MB (L) cells treated with CT7001 and OTX015 individually or in combination at indicated concentrations for 8 h. Data are shown as mean ±SD. qRT-PCR of GLI1 in A673 (M) or ATRT-03 (N) cells treated with CT7001 and OTX015 individually or in combination at indicated concentrations for 8 h. Data are shown as mean ±SD.

Discussion

The efficacy of current clinical available anti-Hh therapy with SMO inhibitor drugs is limited by primary and acquired drug resistance. Our previous finding about BET inhibition in treating Hh-driven cancer suggests that targeting epigenetic or transcriptional regulators of the Hh pathway is a promising approach for overcoming SMOi resistance. Here we have shown that CDK7, another component of RNAPII transcriptional complex, also plays a critical role in regulating transcriptional output of the Hh pathway and CDK7 inhibition provides another effective therapeutic strategy for treating Hh-driven cancer. Like BRD4, CDK7 directly controls the transcription of Gli1 and Gli2, therefore, CDK7 blockade should retain therapeutic efficacy against Hh-driven tumors with either primary or acquired SMOi resistance as BET inhibition. Indeed, our study has proven that CDK7 inhibition preferentially targets Hh pathway transcriptional output in both SMOi-responsive and -resistant Hh-driven tumors; therefore, it is able to circumvent almost all reported SMOi-resistance mechanisms so far, including mutations of SMO, loss of SUFU or SMARCB1, amplifications in GLI2 or MYCN, as well as expression of EWS-FLI fusion oncogene.

Given the broad involvement of CDK7 in RNAPII mediated gene transcription and the importance of Hh signaling in normal development, the potential toxicities of CDK7 inhibition therapy should be carefully evaluated. Our results demonstrate that both RNAPII CTD phosphorylation and cell viability of tested normal neural cells or fibroblast cells were less sensitive to THZ1 compared with Hh-driven mouse MB cells. Given our finding of markedly up-regulated ABC transporter in mediating acquired THZ1 resistance in SMB56R cells, we have measured the expression of the three major ABC transporter genes in our tested cell lines and no significant differences were identified between Hh-driven tumor and control normal cell lines (SI Appendix, Fig. S7A). Even though we cannot rule out the possibility that other members of ABC transporter genes may mediated the resistance, our results suggest that those normal cells carry intrinsic THZ1 resistance that may help them tolerate CDK7 inhibition therapy. In line with this, our in vivo treatment of THZ1 by i.v. injection at 10 mg/kg twice daily did not generate systemic toxicity problem as reported in previous studies of THZ1 treatment of cancer (SI Appendix, Fig. S9 B–D) (44–49).

We have shown before that there are BRD4-independent GLI target genes collectively mediating part of Hh-driven oncogenic phenotype in Hh-driven mouse MB models given that ectopic GLI2-deltaN expression can partially restore the proliferation defect induced by JQ1 in SmoWT-MB cells (37). It raises concerns about the potential limited efficacy of BET inhibitor in treating GLI2-amplified MB if expressions of some of the amplified GLI2 genes are not dependent on BRD4. In this study, we found that ectopic GLI2-deltaN induced Ptch1 transcription was inhibited by THZ1 but not JQ1 (SI Appendix, Fig. S3J). Moreover, GLI2-deltaN stably expressed SMB21 or SMB56 cells both exhibited slight resistance to JQ1 but retained almost the same response to THZ1 as the parental cells (Fig. 3F and SI Appendix, Fig. S4H). These data suggest that CDK7 inhibition may work better for overcoming SMOi resistance than BET inhibition probably due to targeting further downstream transcriptional process. Additionally, we have shown that the combination of CDK7 inhibition and BET inhibition works in synergy to inhibit the aberrant Hh pathway and Hh-driven tumors in vitro and in vivo. Even though the current combination condition could result in some toxicity (SI Appendix, Fig. S9E), it is still worthy of further testing with a more careful design of drug treatment doses and logistics, because other than the synergistic inhibitory effect, the combination may also help suppress the adaptive resistance to CDK7 inhibition (SI Appendix, Fig. S6F) or BET inhibition (54).

In summary, our study demonstrates that targeting transcriptional output of the aberrant hedgehog pathway through CDK7 inhibition alone or in combination with BET inhibition effectively overcomes resistance to smoothened antagonists, further supporting targeting epigenetic or transcriptional process of Hh signaling as a promising approach of overcoming SMOi resistance. Since both CDK7i and BETi drugs have already entered early-phase clinical trial for tumor therapy, our study provides the preclinical rationale for enrolling Hh-driven cancer types into CDK7i and BETi clinical trials in the near future. On the other hand, even though we have tried our best to put a collection of mostly commercially available epigenetic or transcriptional targeted compounds together, we must recognize that the compound library used in our initial screening only targets a very limited number of epigenetic or transcriptional regulators. Therefore, alternative genetic screening approaches are needed to systemically identify other crucial epigenetic or transcriptional regulators of the aberrant Hh pathway in Hh-driven cancers, which may provide new therapeutic target for drug development.

Methods

Epigenetic/Transcriptional Targeted Compounds.

Ninety of the 94 collected epigenetic or transcriptional targeted small-molecule compounds were from 96-well Epigenetics Screening Library (Item 11076, Cayman Chemical). The rest of four compounds were THZ1 (HY-80013) and OTX015 (HY-15743) from MedChem Express as well as Flavopiridol (S1230) and Panobinostat (S1030) from SelleckChem. Additionally, THZ1-R (HY-19988), JQ1 (HY-13030), Ko-143 (HY-10010), and CT7001 (HY-103712A) were purchased from MedChem Express. Smoothened agonist (SAG, S7779), GDC-0449 (S1082), LDE225 (S2151), and PD325901 (S1036) were purchased from SelleckChem.

Cell Culture.

Mouse NIH 3T3 cell line and human Ewing sarcoma line A673 were purchased from Cell Bank of Chinese Academy of Science. Sufu^−/− MEF cell line were kindly provided by P.-T. Chang (University of California, San Francisco, CA). Patient-derived primary human SHH-subtype MB line RCMB018 and ATRT line ATRT-03 were kindly gifted by Robert J. Wechsler-Reya (Sanford Burnham Institute, La Jolla, CA) and Yoon-Jae Cho (Oregon Health & Science University, Portland, OR), respectively.

SMB21, SMB55, and SMB56 lines were kindly gifted by Rosalind A. Segal (Dana-Farber Cancer Institute, Boston, MA). SmoWT-MB and SmoD477G-MB lines were kindly provided by C. Rudin (Memorial Sloan-Kettering Cancer Center, New York, NY).

Primary cell cultures of mouse neural stem cells, astrocytes, granule neuron precursor, and granule neuron cells were all established from freshly isolated mouse cerebellum.

More detailed cell culture information can be found in SI Appendix.

Lentivirus Preparation and Infection.

To generate lentiviral sgRNA or shRNA plasmids, oligonucleotides were annealed and cloned into LentiCRISPRv2 plasmid (Addgene plasmid, 52961) or pLKO.1 plasmid (Addgene Plasmid, 10878), respectively. The sequence of used oligonucleotides can be found in SI Appendix. pCDH-CMV-GLI2-deltaN-EF1-puro plasmid was constructed by subcloning GLI2-deltaN CDS region from pCS2-MT GLI2 delta N (Addgene plasmid, 17649) into pCDH-CMV-MCS-EF1-puro vector. Lentivirus was generated by transfection of HEK293T cells with packaging vectors (pMD2.G and psPAX) and transducing vector and concentrated with PEG6000 (Sigma). Forty-eight hours postinfection, puromycin (1 μg/mL; 60210ES25, YEASEN) was used to select positively infected cells.

Cell Viability Assays.

Cells were plated into 96-well in triplicates or 384-well plates in duplicates and then treated as indicated. Cell viability was measured using CellTiter-Glo 2.0 Assay (G9243, Promega) and data were collected on Synergy H4 Hybrid Reader (BioTek). For testing combinatory effects of two drugs, CalcuSyn software (Biosoft) was used to calculate the combination index (CI) value. CI less than 1.0 indicates synergy.

Cell Proliferation and Apoptosis Assays.

Cell proliferation and apoptosis were measured using Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (C10640, Invitrogen) and Annexin V-FITC Apoptosis Detection Kit (556547, BD Biosciences) respectively. Fluorescence-activated cell sorting (FACS) analyses were performed using BD Fortessa FACS instrument (BD Biosciences) and FlowJo software (FlowJo).

Western Blot Analysis.

Cells were homogenized in RIPA buffer and the protein concentration was quantified using a BCA Protein Assay Kit (Pierce, 23227). The following primary antibodies were used: Gli1 (1:1,000; Cell Signaling Technology, 2643S), Sufu (1:1,000; Cell Signaling Technology, 2522S), Cdk7 (1:2,000; Bethyl Laboratories, A300-405A), RNAPII (1:1,000; Santa Cruz Biotechnologies, sc-899), RNAPII CTD (p-Ser2) (1:1,000, Cell Signaling Technology, 13499S), RNAPII CTD (p-Ser5) (1:1,000; Cell Signaling Technology, 13523S), RNAPII CTD (p-Ser7) (1:1,000; Cell Signaling Technology, 13780S), GLI2 (1:500; Santa Cruz Biotechnologies, sc-28674), Flag (1:500, Sigma, F1804), and β-tubulin (1:5,000; Abcam, ab6046). Secondary antibodies were goat anti-rabbit IgG (0.2 μg/mL; Pierce, 31460) or goat anti-mouse IgG (0.2 μg/mL; Pierce, 31430).

RNA Isolation and Quantitative RT-PCR (qRT-PCR) Analysis.

RNA isolation and reverse transcription were performed using TRIzol Reagent (TR118, MRC) and High Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, 4368813) according to the manufacturer’s instructions. Quantitative PCR (qPCR) analysis was performed with Fast Real-time PCR System (ABI, 7900HT) using SYBR Green Master (ROX) (Roche, 24759100). The qPCR primer sequences can be found in SI Appendix.

Tumor Allograft Models and Drug Treatment In Vivo.

BALB/c nude female mice of 6–8 wk were purchased from Shanghai Experimental Animal Center, Chinese Academy of Sciences. All animal procedures were performed according to the protocols approved by the Medical Experimental Animal Administrative Committee of Shanghai. For flank allograft model, cell suspension was mixed with equal volume of Matrigel (BD Biosciences, 354230) and s.c. injected into each side of dorsal flanks. Tumor volumes were measured twice per week with a caliper and calculated as length × width² × 0.5. For intracranial allograft model, cell suspension was injected stereotactically into the cerebellum (2 mm posterior to lambda suture, 1–2 mm lateral to the midline, and 2.5 mm deep). Mice survival was analyzed by Kaplan–Meier method with Graphpad Prism 6 software. For drug treatment in vivo, THZ1 solution at 30 mg/mL was diluted to 2 mg/mL with dextrose (5%) for i.v. administration (10 mg/kg, twice a day). JQ1 solution at 50 mg/mL was diluted to 5 mg/mL with cyclodextrin (10%) (Sigma, H107) for i.p. administration (25 mg/kg, twice a day).

RNAseq Analysis.

RNAseq service was provided by SMARTQUERIER BIOTECHNOLOGY and GENERGY BIO, Shanghai, China. The RNAseq reads were mapped to the hg19 reference genome using STAR (v2.5.3a) (55). Fragments per kilobase of transcript per million fragments mapped (FPKM) was calculated using Cufflinks (v2.2.1) (56), and active genes was defined as mean FPKM ≥ 1 in either DMSO or treatment group. Differential expression analysis was carried out using R package DESeq2 (v1.20.0) (57). Gene ontology and pathway analyses were performed using Investigate Gene Sets tool on the website http://software.broadinstitute.org/gsea/msigdb/annotate.jsp. Gene set enrichment analysis was performed according to the instructions on the website http://www.broadinstitute.org/gsea/index.jsp.

Statistical Analysis.

GraphPad prism was applied for the statistical analysis. Comparisons of in vivo tumor growth curves were done by two-way ANOVA. Log-rank (Mantel–Cox) test was used for analyzing survival data.