Western blotting was performed as previously described²⁸. Antibodies used were as follows, prepared according to manufacturer recommendations: CDK9 (C12F7) Rabbit mAb Cell Signaling #2316, c-Myc (D84C12) Rabbit mAb Cell Signaling #5605, β-Actin (13E5) Rabbit mAb Cell Signaling #4970, Phospho-Rpb1 CTD (Ser2) (E1Z3G) Rabbit mAb Cell Signaling #13499, Phospho-Rpb1 CTD (Ser5) (D9N5I) Rabbit mAb Cell Signaling #13523, Rpb1 CTD (4H8) Mouse mAb Cell Signaling #2629, α-Tubulin (DM1A) Mouse mAb Cell Signaling #3873, Cleaved Caspase-3 (Asp175) Antibody Cell Signaling #9661, Anti-rabbit IgG, HRP-linked Antibody Cell Signaling #7074.

Click-iT reaction for nascent RNA (Click-iT RNA HCS Assay, C10327, Invitrogen) was performed according to manufacturer instructions following zotiraciclib treatment or DMSO for 24 hrs as previously described²⁹. Cells were then incubated with Myc antibody (Cell Signalingl, 5605S) and secondary antibody (Goat anti-Rabbit IgG Alexa Fluor 568, Invitrogen A-11011) prior to staining with HCS NuclearMask Blue stain solution. Images obtained using a confocal microscope (Keyence).

Lentiviral packaging was performed as previously described²⁹. Control and shCDK9 plasmids were obtained from the University of Colorado Functional Genomics Core Facility (CDK9 #1: TRCN0000000494, CDK9 #2: TRCN0000000497, CDK9 #3: TRCN0000000498).

Cells were seeded 1,000 cells/well in 96-well ultra-low-attachment round-bottom tissue culture plates (Corning) and centrifuged to promote neurosphere formation. Neurospheres were then treated with indicated concentrations of AZD4573 and SY5102 and imaged for one week using an IncuCyte S3 Live Cell Analysis System. Neurosphere measurements were then normalized to the control wells and analyzed per Bliss and HSA synergy models using Combenefit software (v2.021)³⁰.

Athymic nude mice aged 4 to 8 weeks were anesthetized, immobilized, and prepared for intracranial injection as previously described²⁸. To target the cerebellum, a 1.0 mm diameter burr was drilled in the cranium with a dental drill bit at 1.500 mm to the right and 2.000 mm posterior to lambda. A suspension of D458 cells was prepared by suspending a concentration of 20,000 cells/3 μl in serum free media per injection, and slowly injected 3.000 mm ventral to the surface of the skull using an UltraMicroPump III and a Micro4 Controller (World Precision Instruments) equipped with a Hamilton syringe and a 26-gauge needle. The burr hole was sealed, incision closed, and topical antibiotic applied as previously described. Post-surgical pain was controlled with SQ carprofen (5 mg/kg/daily) on the day of and for 2 days following the procedure. Flank models generated as previously described²⁹. Protocol-defined endpoints included ataxia, >15% weight loss, tumor size ≥2 cm, skin ulceration, or significant distress. University of Colorado Institutional Animal Care and Use Committee (IACUC) approval was obtained and maintained throughout the conduct of the study.

Cleavage Under Targets & Release Using Nuclease assays were performed as previously described²⁹ using H3K4me3 (EpiChypher 13–0041) and Myc (Cell Signaling, 13987S) antibody in accordance with the manufacturer’s protocol (CUTANA ChIC/CUT&RUN EpiCypher 14–1048). The NEBNext Ultra DNA Library Prep Kit for Illumina (NEB #E7645S) with Dual Index Primers (NEB #E7600) were used for library preparation prior to sequencing on the NovaSeq6000 platform.

Hi-C reactions and library preparations were performed by Arima Genomics. FFPE samples were sectioned at 5–10 μm thickness, de-waxed, and rehydrated prior to sample preparation using the Arima-HiC+ for FFPE kit (#A311038). Sequencing libraries were prepared by shearing proximally ligated DNA followed by size-selection of DNA fragments using SPRI beads. Size-selected fragments were then enriched for fragments containing ligation junctions using streptavidin beads before library construction. Illumina-compatible sequencing libraries were prepared using the Arima Library Prep Module (#A303011) according to manufacturer instructions and sequenced on the NovaSeq6000 platform.

After sample demultiplexing, raw read pairs were mapped to the reference genome (hg19) and de-duplicated using HiCUP³¹. Processed alignments are then converted into Hi-C matrices using Juicer v1.6³² and visualized using Juicebox³³. Chromatin loop calls were identified using HiCCUPS algorithm with KR normalization. TADs were identified using Arrowhead algorithm at multiple resolutions.

We have previously performed and reported CRISPR-Cas9 screening of 1,140 druggable targets across three established models of Myc-MB³⁴. In order to strengthen the reproducibility of this approach and further enrich for conserved dependencies to prioritize for translational development, we intersected these data with published CRISPR-Cas9 screens performed specifically in Myc-MB by two additional independent labs (eight cell lines total)^35,36 (Figure 1A and Supplemental Table 1). As the statistical cutoffs applied and the detail of raw data reported varied between publications, we included significant hits as originally defined by the source manuscript authors. This resulted in 35 genes consistently identified as significant dependencies across all three screens (Figure 1B). Conserved hits included genes with well-characterized roles in cell cycle regulation (CDK1/6, CHEK1, WEE1, AURKB, PLK4, CDC7), nucleotide synthesis (DHODH, RRM1/2, TYMS), and DNA damage response (POLE, MCL1, RAD51, BCL2L1) (Supplemental Table 1). Also notable were hits for the transcription-associated cyclin-dependent kinases CDK9 and CDK8, with the closely related CDK7 scoring in 7 of 8 cell lines^34–36.

In order to validate CDK9 as a relevant therapeutic target in Myc-MB, we next transduced two established Myc-driven cell lines (D425 and D458) with non-overlapping shRNA constructs targeting CDK9 or scrambled control. Successful knockdown of CDK9 protein expression resulted in a proportional decrease in expression of Myc and a loss of proliferative potential when cells were cultured in neurosphere conditions (Figure 1C–E). We then engrafted the shNull-D458 or shCDK9-D458 cells orthotopically in the cerebellum of athymic nude mice. CDK9 knockdown led to a delay in tumor development and significant prolongation in animal survival (median survival 32 days shNull vs 41 days shCDK9, p=0.03) (Figure 1F–G). When tumors were collected at endpoint, shCDK9-D458 tumors showed considerably lower Myc expression by immunohistochemistry (Figure 1H). In total, these data identify CDK9 as a conserved dependency in Myc-driven medulloblastoma and suggest it may be amenable to targeting for anti-tumor effect.

There has been growing interest in CDK9 as a promising target for cancer therapy, leading to the formulation of increasingly specific inhibitory compounds now in preclinical and early clinical development^37,38. We therefor sought to assess whether the above anti-tumor effects could be replicated using small molecule inhibitors targeting CDK9 (CDK9i). We tested a range of available CDK9i against a panel of Myc-driven (D425, D458) and non-Myc-driven (ONS76, UW228) medulloblastoma lines. We broadly observed nanomolar (atuveciclib, AZD4573, flavopiridol, PHA767491) or low-micromolar (LDC00067) susceptibility to CDK9i across the models tested, with a trend towards lower mean inhibitory concentrations in the Myc-MB (Figure 2A). We selected two compounds highly specific for CDK9 (vs other CDKs or kinases), atuveciclib³⁹ and AZD4573⁴⁰, for further testing (Figure 2B). Treatment of Myc-MB lines with either atuveciclib (2 uM) or AZD4573 (40 nM) for 24 hours led to a marked reduction in Myc mRNA and protein expression (Figure 2C–D). When tested in neurosphere culture, a dose-dependent reduction in fitness was observed (Figure 2E), supporting the efficacy of CDK9-selective agents as anti-tumor therapies for Myc-MB. To more specifically interrogate the relationship between Myc amplification status and CDK9i sensitivity, we utilized a system in which the non-Myc-driven ONS76 line is modified to overexpress either Myc (ONS76-MycOE) or vector control (ONS76-Ctrl)(Figure 2F and ⁴¹). When exposed to either AZD4573 or atuveciclib, ONS76-MycOE cells demonstrated cytotoxic sensitivity at significantly lower concentrations than did ONS76-Ctrl (Figure 2F). This supports the model that transcriptional amplification in Myc-driven systems may create a relative dependency on transcriptional machinery such as CDK9. However, we note cytotoxic effect from these inhibitors in both ONS76-Ctrl and the non-Myc-driven MB lines more broadly at therapeutically relevant concentrations, suggesting the anti-tumor effect of CDK9i is not restricted solely to Myc-driven disease.

To better understand the mechanistic consequences of selective CDK9 inhibition in Myc-MB, we then performed RNA sequencing (RNA-seq) after 72 hours of atuveciclib treatment and compared to untreated controls in both D425 and D458 cell lines. This identified 532 – 1728 differentially expressed genes with pval <0.01 and LFC >1.2 (Figure 2F). Atuveciclib treatment resulted in a significant downregulation of both transcription by RNA Pol II and hallmark Myc target programs, as well as gene sets related to cell division and cell cycle progression (Figure 2G). Functional programs demonstrated strong overlap between the two models (Supplemental Figure 1). Taken together, these support CDK9 inhibition as an effective means of drugging transcriptional circuitry at the heart of Myc-driven programs.

Regulation of processive transcription by RNA Pol II is not a single-step checkpoint but instead is governed by inputs from multiple transcriptional CDKs across the transcription cycle^19–21. Our team has previously identified and characterized the role of the TFIIH-member CDK7 in Myc-driven circuitry within MB³⁴. Given the closely related biological functions of these two kinases, we asked whether they could be targeted in parallel, both to augment anti-tumor effect and expand the repertoire of available small molecule inhibitors available for study. Again using AZD4573 as a CDK9-specific tool compound, we tested it at increasing concentrations with SY5102, a highly-selective noncovalent inhibitor of CDK7⁴². When assessing synergy using the Bliss model, we observed a remarkable synergistic effect at very low nanomolar concentrations (synergy defined as Bliss score ≥10, observed scores >30–40) (Figure 3A). Analysis using HSA modeling of synergy yielded similar positive scores (Supplemental Figure 2), supporting the utility of targeting these transcriptional cofactors in parallel.

Following this, we turned our attention to the drug zotiraciclib. Zotiraciclib (ZTR) is an orally-available, pyrimidine-based inhibitor that inhibits CDK9 at an IC50 of 3 nM and CDK7 at 37 nM (Cothera Bioscience, Investigator Brochure), and it is currently in phase I/II trials for adults with relapsed anaplastic astrocytoma or glioblastoma (⁴³ and NCT03224104, NCT02942264). When tested against the D425 and D458 models of Myc-MB, it demonstrated anti-tumor effect with mean inhibitory concentrations in the low nanomolar range and inhibition of CDK-catalyzed Pol II CTD phosphorylation (Figure 3B–C). We then tested dose-dependent effects with Myc-MB cells cultured in neurosphere conditions. At concentrations tested as low as 25 nM, all zotiraciclib dose exposure resulted in a complete cessation of growth in neurosphere media (Figure 3D). Together, this supported combined targeting of CDK9 with CDK7 as a promising means of disrupting transcriptional circuitry in Myc-MB.

The MYC promoter is regulated by inputs from distinct, evolutionarily conserved super enhancers, and disruption of this enhancer activity leads to a loss of MYC expression comparable to deletion of the MYC locus itself^12,15,44. Given the comparatively high levels of transcriptional output at enhancer or super enhancer loci, inhibition of transcriptional machinery has been shown to preferentially deplete enhancer-driven oncogenes^13,15. Active enhancer loci or specific enhancer-promoter (E-P) regulatory looping are tightly regulated across development and cell fate commitment, but they can be altered in states of neoplastic transformation, including medulloblastoma^45–47. This complicates experimental assessments of enhancer perturbation, as E-P inferences based on genomic distance or transcriptional correlation may be confounded by complex chromatin rearrangements.

To address this, we performed chromatin conformation capture (Hi-C) from two primary FFPE patient samples of Myc-MB⁴⁸. Each patient sample was clinically annotated as a Group 3 tumor with Myc amplification in double minute pattern by FISH (Figure 4A and Supplemental Figure 3). Coverage maps delineated quite distinct regions of amplification between patients (approximately 5.7 Mb upstream in UPN1355, 885 Kb downstream in UPN1433), though both encompassed the entirety of the MYC gene body (Figure 4B). When examining loops anchored at the MYC promoter, we called 14 unique loops in one sample and 15 in the second. Of these, 8 were present in both patient samples, while the remainder were patient-specific (Figure 4B and Supplemental Table 2). To expand this characterization of MYC- promoter-anchored looping in MB, we next reanalyzed a recently published dataset of 3D topology in MB⁴⁹ applying the same filtering criteria. We observed considerable inter-tumor variability of looping around the MYC promoter. Though not annotated as Myc-amplified or not within this dataset, group 3 tumors had, on average, a greater number of loops anchored at the MYC promoter than group 4, SHH, or WNT tumors (mean 5.6 [range 1–13], vs 2.2 [1–5], 3.6 [1–5] or 3 [3–3], respectively). Of the MYC-anchored loops in group 3 tumors, only 5 loops were present in more than one group 3 sample, and only two appeared unique to group 3 (Supplemental Table 3). This is consistent with prior reports that while promoters of core oncogenic drivers maintain similar active epigenetic signatures across a diagnostic entity, non-promoter regulatory inputs may vary widely from patient to patient⁵⁰.

We then focused on the topologically associated domain (TAD) surrounding the MYC promoter, intersecting our Hi-C data with H3K27ac ChIP-seq³⁵, H3K4me1 CUT&RUN, and RNA-seq from D458 cells to refine active E-P contacts (Figure 4C, top). Of the 8 MYC-anchored loops conserved between both patient samples, 7 cognate anchors showed H3K4me1 deposition consistent with enhancer elements. Two of these lacked convincing H3K27ac or RNA-seq reads and thus may have been inactive in the cell line, while the remainder showed strong H3K27ac and active transcription. This included an enhancer locus ~90kb upstream of the MYC promoter that showed exceptionally strong contact frequency and has been previously validated as a group 3 subgroup-specific enhancer⁴⁵ (Figure 4C, ENSG00000249375 left). When we examined Myc binding by CUT&RUN, we found each of the active putative enhancers to themselves be bound by Myc, consistent with the feed-forward mechanism commonly described in transcriptional core regulatory circuitry (CRC)⁵¹. Inspection of the patient-specific loop anchors showed variability in co-localized activating chromatin marks and RNA-seq reads, suggesting that many of these may also reflect functionally active if variable E-P loops (Figure 4C, LINC00824 right).

Next, we tested whether zotiraciclib treatment could be used to disrupt transcriptional activity at these enhancers and, by extension, their collective amplification of transcription at the MYC promoter itself. We treated D458 cells with zotiraciclib (200 nM for 24 hours) before repeating the Myc CUT&RUN and RNA-seq experiments. At each of the loci identified with active enhancer features, treatment with zotiraciclib led to a loss of Myc occupancy and loss of transcriptional activity as measured by RNA-seq (Figure 4C, bottom). Consequently, at the MYC promoter we observed a dramatic loss of Myc binding and corresponding abrogation of transcriptional output (Figure 4D). This zotiraciclib-induced loss of Myc expression was consistent when validated by qPCR and Western blot across multiple cell lines (Figure 4E–F). To assess whether this transcriptional depletion was a global phenomenon, we examined the same chromatin features at several housekeeping genes (ACTB, GAPDH). At baseline, these loci notably had neither E-P loops identified by Hi-C nor significant Myc binding by CUT&RUN, and when exposed to zotiraciclib, transcriptional output was comparatively unaffected (Figure 4G). In total, these data support a model in which CDK9/7i treatment potently inhibits E-P transcriptional activity within the MYC-amplified TAD, resulting in a marked loss of overall Myc expression (Figure 4H). The limitation in sample numbers subjected to Hi-C means that the characterization of E-P looping shown here is illustrative rather than comprehensive, and a much larger cohort would be required to robustly map the spectrum of enhancer structure for Myc-MB. However, the conserved role of CDK7 and CDK9 in modulating transcriptional output means that this therapeutic dependency could be reasonably anticipated to hold true irrespective of the specific structural arrangement within a given patient tumor.

Myc acts broadly as transcriptional amplifier, stimulating Pol II pause-release in both neoplastic and non-transformed cells^11,17. Aberrant Myc activity drives oncogenic transcriptional programs involved in stemness, proliferation, and DNA repair across a range of cancers, including Myc-MB^34,52,53. Given the depletion of Myc expression we observed following CDK9/7i treatment, we predicted a concordant suppression of downstream Myc-driven oncogenic functions. We first visualized this directly by combining immunofluorescence for Myc and nascent RNA synthesis. This demonstrated high levels of nuclear Myc localization uniformly across the Myc-MB cells, with puncta of nascent transcriptional activity clearly identifiable. Treatment with zotiraciclib led to a complete loss of appreciable Myc expression and significant downregulation of nascent transcription, though some foci of transcriptional activity persisted (Figure 5A). To characterize the consequences of this on Myc-driven transcriptional program, we examined changes to Myc binding (CUT&RUN, n=2) and gene expression (RNA-seq, n=3) genome-wide. This demonstrated a marked loss of Myc occupancy on chromatin following zotiraciclib treatment (Figure 5B). Differentially lost peaks were enriched for gene programs involved in chromatin remodeling and transcriptional regulation, cell cycle regulation, and DNA repair (Figure 5C and Supplemental Table 4). We replicated the RNA-seq experiment in an additional cell line (D425, n=3) and sorted for gene sets that were differentially enriched in both cell lines (n=3 replicates per model, p adj <0.05). Reproducibly enriched programs included downregulation of hallmark Myc signaling, cell cycle checkpoint regulation, and DNA repair, with a reciprocal upregulation of apoptosis (Figure 5D–E and Supplemental Figure 4). TRRUST analysis⁵⁴ of transcriptional regulatory networks implicated in the observed expression changes clearly inferred Myc as the strongest candidate TF driving these programs (Figure 5F and Supplemental Table 5).

We then validated these transcriptomic results via functional assays. Quantification of direct microscopy demonstrated a significant decrease in mitotic figures with a corresponding increase in apoptotic bodies visible following zotiraciclib treatment (Figure 5G). Synchronized cell cycle analysis by flow cytometry showed a G1 arrest following zotiraciclib exposure (Supplemental Figure 5A), with a significant induction of apoptosis as assessed by caspase 3 cleavage by immunoblot (Supplemental Figure 5B). Given the central role of ionizing radiation (IR) in standard-of-care multimodal therapy for Myc-MB⁷, we assessed IR-induced DNA double strand breaks (DSBs) as a surrogate for DNA damage response efficiency. Zotiraciclib treatment alone resulted in a significant accumulation in DSBs as measured by yH2AX, increased further when applied in combination with IR (Supplemental Figure 6). Collectively, these data establish that CDK9/7i-mediated Myc depletion leads to a loss of Myc binding on chromatin and a consequent loss of Myc-driven transcriptional programs. These programs represent core oncogenic functions in Myc-MB, including proliferation, DNA repair, and control of apoptosis. CDK9/7i therefore presents a novel, clinically applicable means of disrupting the core TF circuitry underpinning Myc-MB growth and survival.

As these data supported the utility of selective CDK9 inhibition as a therapeutic strategy for Myc-MB, we finally tested whether CDK9i would maintain anti-tumor effect in vivo. The agents utilized in this study have primarily been developed in the context of hematologic malignancies, and subsequent data has suggested that CNS exposure across the blood-brain barrier may be poor (^28,29 and AstraZeneca, Investigator Brochure). Athymic nude mice were injected orthotopically into the cerebellum with D458 cells and randomized to control or AZD4573 treatment (15/15 mg/kg intraperitoneal biweekly⁴⁰), atuveciclib (30 mg/kg/dose daily oral gavage; 3 days on / 2 days off²⁸), or zotiraciclib (50 mg/kg oral gavage three times weekly) (Figure 6A). Unfortunately, mice treated with CDK9i showed no benefit in survival (Figure 6B and Supplemental Figure 7).

To assess anti-tumor effect of CDK9 inhibition in vivo independent of CNS penetration, we repeated this testing of CDK9i using subcutaneous models. Animals were injected in the flank and randomized to the same AZD4573, atuveciclib, or zotiraciclib dosing regimens(). In contrast to what was observed with the intracranial model, treatment with CDK9i led to a statistically significant improvement in xenograft survival (median survival 41 vs 63 days AZD4573, p=0.03; vs 60 days atuveciclib, p=0.03; vs 63 days zotiraciclib, p=0.008) (Figure 6C and Supplemental Figure 7). Tumor volumes were decreased in CDK9i-treated animals, with a concordant decrease in Myc intensity by IHC (Figure 6D–F). Our data here is largely consistent with prior evidence of limited CNS exposures for this class of agents, though zotiraciclib has been reported as CNS penetrant in some studies^55–57. This highlights that while CDK9 is a relevant pharmacologic target for Myc-MB, consideration of pharmacokinetic properties will remain important for effective application to primary CNS disease.