Targeting CDK9 Reactivates Epigenetically Silenced Genes in Cancer

Abstract

Cyclin-dependent kinase 9 (CDK9) promotes transcriptional elongation through RNAPII pause release. We now report that CDK9 is also essential for maintaining gene silencing at heterochromatic loci. Through a live cell drug screen with genetic confirmation, we discovered that CDK9 inhibition reactivates epigenetically silenced genes in cancer, leading to restored tumor suppressor gene expression, cell differentiation, and activation of endogenous retrovirus genes. CDK9 inhibition dephosphorylates the SWI/SNF protein BRG1, which contributes to gene reactivation. By optimization through gene expression, we developed a highly selective CDK9 inhibitor (MC180295, IC50 = 5 nM) that has broad anti-cancer activity in vitro and is effective in in vivo cancer models. Additionally, CDK9 inhibition sensitizes to the immune checkpoint inhibitor α-PD-1 in vivo, making it an excellent target for epigenetic therapy of cancer.

Keywords: BRG1; CDK9; DNA methylation; SMARCA4; SWI/SNF; drug development; epigenetic therapy; gene silencing; immunosensitization; kinase inhibitors.

Figure 1:

CDK9 inhibition reactivates epigenetically silenced genes. See also Figure S1 and Table S3. (A) Expression of GFP (measured by FACS) after 24hr treatment of YB5 with HH0 (structure at the top). Depsipeptide (Depsi), an HDACi, was used as a control. Insert: representative fluorescent microscopy of YB5 treated with DMSO (left) and HH0 (right). (B) GFP expression 24hr after treatment with CDK inhibitors measured by FACS. Corresponding structures are shown on top. (C)Reactivation of GFP, SYNE1 and MGMT 24hr after treatment with CDK9 inhibitors as detected by qPCR. (D) Reactivation of GFP, SYNE1 and MGMT 72hr after dominant-negative CDK9 (dnCDK9) overexpression (TET-off) detected by qPCR. HH1 (25μM, 24hr) was used as a positive control. The Western Blot (right) shows that dnCDK9 is overexpressed when cells are grown in the absence of tetracycline, together with decreased phosphorylation of RNA Pol ll at Ser2 (pSer2). (E) Overexpression of P-TEFb (CDK9 and Cyclin T1) for 72hr abolished the effect of CDK9 inhibitors (24hr) on the activation of GFP, SYNE1 and MGMT as detected by qPCR. Depsipeptide was used as a negative control (uninhibited by CDK9 overexpression). The Western Blot (right) shows the overexpression of CDK9 and Cyclin T1 72hr after viral transduction. Single transduction of CDK9 or Cyclin T1 was used as a size control.

Figure 2:

Structure activity optimization identifies MC180295, a novel potent and selective CDK9 inhibitor. See also Figure S2 and Table S4. (A) GFP expression measured by flow cytometry four days after single dose treatment with aminothiazole analogs in YB5. Corresponding structures are shown on top. Data are shown as mean+SD, N=3. *p<0.05, ***p < 0.001. (B) Kinase phylogenetic tree showing the distribution of human kinases inhibited by 1μM MC180295. (C) IC50 of MC180295 against 10 CDK/Cyclins. (D) Quantification of Western Blots 2hr after MC180295 treatment against pSer2 (CDK9 target), phosphor-Rb at T870/811, phosphor-Rb at T826, p130 (CDK4/6 targets), phosphor-CDK Substrate Motif (K/H)pSP and phosphor-PRC1 (CDK1/2 targets). At 500nM (top), there is specific inhibition of CDK9 only while at 5μM, CDK9, 4 and 6 are inhibited. Data on other doses are in Figure S2D. (E) In our model (green), the aminothiazole core of MC180295 engaged the CDK9 hinge region (left panel) with interactions that mimic that of dasatinib (shown here bound to cSrc, PDB ID 3G5D, pink). MC180295 engaged the conserved Lys48-Glu66 hydrogen bond (green) (middle panel); the multi-CDK inhibitor flavopiridol also made a similar interaction (PDB ID 3BLR, pink). The norbornyl group from MC180295 requires that the C-terminus of the hinge region adopts a slightly lower conformation; this conformation is shared amongst the many crystal structures of CDK9 (right panel) (yellow: structures of CDK9 bound to ATP and to another Type I inhibitor (PDB IDs 3BLQ and 4BCJ), and our model of MC180295), but this loop conformation is rarely observed in structures of other CDK kinases (blue: representative structures of CDK1/2/5/6/7, each bound to ATP or a Type I inhibitor (PDB IDs 5LQF/1HCK/1UNH/2EUF/1UA2)).

Figure 3:

Inhibition of CDK9 leads to global reactivation of epigenetically silenced genes. See also Figure S3. (A) Time course of gene expression after HH1 treatment as measured by RNA-seq. Data show mean fold change + SEM for each gene group. There is overall gene induction (black line) but silenced genes (green line) have more profound gene induction. The blue dotted line represents one-fold change. (B) Number of silenced genes (baseline RPKM < 0.31) up- and downregulated by 10μM HH1 at each time point as measured by RNA-seq (N=3, FC>2 or <0.5, FDR<0.1) after excluding genes downregulated at 2hr/4hr. dnCDK9 (72hr), DAC (100nM daily for 48hr) and combinatorial treatment (100nM DAC daily for 48hr followed by a single10μM HH1 exposure for four days) were also included. (C) Percentage of genes upregulated by HH1, dnCDK9, DAC and combinatorial treatment that have low (0–10%), moderate (10–50%), or high (50–100%) promoter DNA methylation measured by RRBS. (D) Dynamics of gene expression for silenced genes (baseline RPKM < 0.31) that were significantly upregulated (FC>2, FDR<0.1) four days after HH1. The yellow dotted lines represent two-fold change. (E) 3D principal component analysis of RNA-seq data upon DMSO (in blue) or HH1 10μM treatment (in green) (N=3). Also shown are dnCDK9 (72hr) (in gold (dnCDK9-on) and orange (dnCDK9-off)), 48hr daily DAC treatment at 100nM (in pink) and sequential combinatorial treatment (100nM DAC daily for 48hr followed by HH1 at 10μM (in red)) (n=3). DMSO and dnCDK9-off clustered together and are circled in red (baseline). DAC, dnCDK9-on and 4-day HH1 also clustered together and are circled in blue (long-term). Different time points are shown in different shapes and labeled in the legend. (F) Gene expression changes after HH1 (y-axis) recapitulate the effect of dominant negative CDK9 (x-axis). Concordant changes are in orange, discordant changes in blue, changes smaller than two-fold are in grey. The numbers in each quadrant show genes with > 2fold expression change and % of total genes. (G) Density plots show the distribution of differentially expressed genes (log₂ fold change on y-axis and log₂RPKM on x-axis) after HH1 treatment at 10μM or dnCDK9 overexpression (72hr). The red lines represent no change. (H) Ingenuity Pathway Analysis of upstream regulators of genes activated by HH1 based on genes in (D).

Figure 4:

CDK9 regulates BRG1 to de-repress silenced genes. See also Figure S4. (A) Enrichment of ATAC-seq signal around transcription start sites (TSS). Data show merged triplicates. On the left is aggregated enrichment of all genes around TSS for cells treated with DMSO (blue) or HH1 (green). On the right is aggregated enrichment around TSS of genes that are induced by HH1 treatment (green) compared to DMSO (blue). (B) Representative traces of ATAC-seq after HH1 treatment. SPOCK2 and CYP1B1 are significantly (FC>2) upregulated by HH1 and are methylated in their promoter regions. Arrows indicate peaks gained in promoter region. “Normalized HH1” refers to ATAC-seq reads after normalization based on invariantly expressed genes. (C) (C–F) CDK9 immunoprecipitates with BRG1. In each panel, the immunoprecipitation antibody is shown on top while the Western Blot antibody is shown on the right. In (C), BRG1 immunoprecipitates with transfected FLAG-tagged CDK9 in HEK293T. In (D), (top) CDK9 immunoprecipitates with transfected FLAG-tagged BRG1 and (bottom) BRG1 immunoprecipitates with transfected GFP-tagged CDK9 in SW48. (E) shows endogenous Co-IPs in HEK293T while (F) shows endogenous Co-IPs in SW48. IgG was used as a negative control for panels E and F. The IgG heavy chain (Hc) was also shown in panel F. (G) Isotope kinase activity assay using recombinant active full-length CDK9 and BRG1 with or without CDK9 inhibitors (flavopiridol (FVP) and MC180295 (295)) in the presence of ³²γ-ATP. RNA-Pol-ll, CTD, BRG1 and CDK9 itself were all phosphorylated by CDK9 and unphosphorylated after CDK9 inhibition. Coomassie blue staining is shown on the bottom of the graph to verify equal loading. (H) BRG1 overexpression overcomes gene silencing and synergizes with CDK9 inhibition. YB5 cells were transfected with different V5-tagged BRG1 constructs (WT (wild-type), 5STOA (five serine residues substituted by alanine residues) and NO5S (five serine residues were deleted)) for 48hr prior to drug treatment for 48hr. The number of GFP+ cells was measured using confocal microscopy. Data are shown as mean+SD, N=3. *p<0.05, **p < 0.01, ***p < 0.001.

Figure 5:

In vitro and in vivo efficacy of CDK9 inhibition. See also Figure S5. (A) Proliferation responses of a normal lung fibroblast (IMR90) and cancer cell lines treated with a single-dose of 5μM HH1 or 0.1μM MC180295 and counted four days after treatment by trypan blue exclusion. (B) (C) Soft agar assays of SW48 (B) and HCT116 (C) cells following a single dose of HH1 or MC180295 for four days. (D) HL60 differentiation measured by expression of CD11b (flow cytometry) after a single dose exposure to HH1 or MC180295 for four days. 1μM ATRA and high concentrations of DMSO (1.25%) were used as positive controls. (E) Anti-tumoral effect of MC180295 in vivo. (Left) NSG mice were inoculated (s.c.) with 2×10⁶ SW48 cells. Eleven days later, when tumors were palpable, 20mg/kg MC180295 or vehicle was administered (i.p.) qod. Tumor sizes were measured using a caliper. Data are shown as mean+SEM (Student's t-test). (Right) Survival of the mice in days. Significance was calculated using a log-rank (Mantel-Cox) test. (F) Efficacy of CDK9 inhibition in a syngeneic cancer model. Measurement of ascites fluid in the VEGF-DEF ID8 ovarian cancer mouse model is an indicator of tumor burden. In vivo treatment with the CDK9 inhibitor SNS-032 every 3 days demonstrated a decrease in tumor burden at weeks 4 and 5 (left). Addition of α-PD-1 led to a further decrease in tumor burden at week 5. Data are shown as mean+SEM (Mann Whitney test). Survival of NSG mice in (G) was calculated using a log-rank (Mantel-Cox) test. MC180295 significantly extended survival of the mice in this model (right). In all panels, *p<0.05, **p < 0.01, ***p < 0.001. Bar graphs represent mean+SD of at least biological triplicates.

Figure 6:

CDK9 inhibition triggers upregulation of endogenous retroviruses (ERV) and synergizes with immune checkpoint inhibitors in vivo. See also Figure S6. (A) ATAC-seq reads mapping to repetitive elements enriched or depleted after 4-day HH1 treatment are shown on the left. There is a high preponderance of enrichment, consistent with broad chromatin decompaction. Differentially expressed repetitive elements as measured by RNA-seq after 4-day HH1 are shown on the right. The majority of these are activated repeats, consistent with the broad epigenetic effects of CDK9 inhibition. (B) ERV activation measured by qPCR four-days after single dose CDK9 inhibitor treatment in YB5 cells (n=3). DAC was used as a positive control. Data are shown as mean+SD. *p<0.05, **p < 0.01, ***p < 0.001. (C) CDK9 IMmune Signature (CIM) gene expression panel clustered TCGA melanoma patients into high and low immune signatures. CIM-high patients have a longer survival than CIMlow patients. (D) In vivo treatment of mice with SNS-032 resulted in increased populations of immune cells (CD45+) and T cells (CD3+) in the tumor microenvironment. *p<0.05, **p < 0.01, ***p < 0.00). (E) MC180295 did not inhibit human T-cell growth in vivo. 20 million human PBMCs from a healthy donor were injected i.p. on day 0 into NSG mice. 20mg/kg MC180295 was injected (i.p.) qod for 12 days and whole blood was collected on day 14. Flow cytometry was performed using anti-CD45, anti-CD4 and anti-CD8 antibodies.