Synthetic Lethal Targeting of CDK12-Deficient Prostate Cancer with PARP Inhibitors

Abstract

Purpose: The cyclin-dependent kinase (CDK), CDK12, is mutated or amplified in multiple cancers. We previously described a subtype of prostate cancer characterized predominantly by frameshift, loss-of-function mutations in CDK12. This subtype exhibits aggressive clinical features.

Experimental design: Using isogenic prostate cancer models generated by CRISPR/Cas9-mediated inactivation of CDK12, we conducted a chemical library screen of ∼1,800 FDA-approved drugs. We inhibited cyclin K and CDK13 and evaluated the effects on PARP inhibitor (PARPi) sensitivity. CDK12 truncation and kinase domain mutations were expressed in cell lines to determine the effects on PARPi sensitivity. Mice bearing control and CDK12-mutant prostate tumors were treated with rucaparib. Finally, we evaluated PSA responses in patients with CDK12 mutations treated with rucaparib on the TRITON2 trial.

Results: Cancer cells lacking CDK12 are more sensitive to PARPi than isogenic wild-type cells, and sensitivity depends on the degree of CDK12 inhibition. Inhibiting cyclin K, but not CDK13, also led to PARPi sensitivity and suppressed homologous recombination. CDK12 truncation mutants remained sensitive to PARPi, whereas kinase domain mutants exhibited intermediate sensitivity. The PARPi rucaparib suppressed tumor growth in mice bearing CDK12-mutated tumors. Finally, 6 of 11 (55%) patients with prostate cancer with biallelic CDK12 mutations had reductions in serum PSA levels when treated with rucaparib on the TRITON2 clinical trial.

Conclusions: In prostate cancer, sensitivity to PARPi is dependent on the specific type and zygosity of the CDK12 mutation. PARPi monotherapy may have some activity in patients with prostate cancer with biallelic inactivating CDK12 alterations.

Figure 1:. CDK12 deletion in prostate cancer cells is synthetic lethal with PARP inhibition.

(A-B) Western blots (WB) showing CDK12 protein in LNCaP (A) and C42B (B) CDK12 KO clones. Lamin B1 is shown as loading control. (C) Small molecule screen showing the fold change (FC) per drug in the C42B CDK12 KO (y-axis) over the FC in the Control (x-axis). The concentration of each drug is shown in cyan (0.1 μM), green (1 μM) and orange (10 μM), and the relative FC ratio is indicated by the size of the circle. The drug hits are indicated in the dotted square in the bottom right corner. (D) The drug hits are organized by drug class categories and are highly enriched in molecules involved in DNA damage. (E-F) Dose response assays to olaparib in LNCaP (E) and C42B cells (F). A representative assay of n=3 independent experiments performed in technical triplicates is shown for each cell line. (G-H) Clonogenic assay and quantification in C42B and CDK12 KO cells treated with the indicated doses of olaparib. The assay was performed in triplicate, n=3 independent experiments, *** indicates p<0.001, **** indicates p<0.0001. Error bars represent SEM for E-F. ANOVA with multiple comparison correction (Sidak) was performed for H.

Figure 2:. CDK12 loss in multiple cancer cell types results in PARPi sensitivity in a gene dosage dependent manner and CDK12 is necessary for HR repair and gene expression.

(A-E) Dose response assays to talazoparib using CRISPRi to knockdown CDK12 or BRCA2 in LNCaP (A), PC3 (B), CAOV3 (C), OVCAR8 (D), and DLD-1 (E) cells. A representative assay of n=4 independent experiments performed in technical triplicates is shown for each cell line. (F) WB showing protein levels of CDK12 in control and CRISPRi knockdown in LNCaP cells. (G) mRNA levels of CDK12 and BRCA2 in control and CRISPRi knockdown in LNCaP cells, as measured by QPCR. **p<0.01, n=3 independent samples. (H) Dose response assay to talazoparib in LNCaP CDK12 KO versus CDK12 CRISPRi KD cells. (I) Dose response assays to olaparib in LNCaP CDK12 heterozygous (Het1 and Het2) versus KO cells, compared to controls. (J) WB showing protein levels of CDK12 levels in control, KO, Het1 and Het2 cells (K) U2OS-DR-GFP reporter cells were transfected with ISce-I and siRNAs targeting CDK12 or BRCA2 and GFP fluorescence was measured. ***p<0.001, n=3 independent assays performed in triplicate. (L) U2OS-DR-GFP reporter cells were treated with increasing doses of THZ531 (0–400nM), transfected with ISce-I and GFP fluorescence was measured. *p<005, **p<0.01, n=3 independent assays performed in triplicate. (M) Quantification of comet tail:head DNA ratios in C42B control and CDK12 KO cells treated with or without 1μM rucaparib for 5 days. At least 100 nuclei were counted for each cell line and condition. Box and whiskers plot shows a representative experiment of n=3 independent experiments. ns=not significant, ***p<0.001. (N) mRNA levels of indicated DNA repair genes in C42B cells treated with 250 nM of THZ531, as measured by QPCR (n=5 samples at each concentration, graph depicts mean $\pm$ SEM, *p<0.05, **p<0.01, ***p<0.001, p<0.0001, ns=not significant). Error bars represent SEM in A-E, G-I, and K-N. Student’s t test was performed for G. ANOVA with multiple comparison correction (Sidak) was performed for K-N.

Figure 3:. Loss of cyclin K (CCNK) but not CDK13 impairs HR and results in PARPi sensitivity.

(A) mRNA levels of CCNK and CDK13 in C42B CRISPRi cells expressing an sgRNA targeting CCNK or CDK13 (n=3 samples, graph depicts mean $\pm$ SEM, **p<0.01). (B) Dose response assay to rucaparib in C42B CRISPRi cells targeting CCNK or CDK13. (C) WB of CDK13 in C42B CDK13 KO cells. GAPDH is shown as loading control. (D) Dose response assay to rucaparib in control and CDK13 KO cells. A representative assay of n=3 independent experiments performed in technical triplicates is shown. (E) U2OS-DR-GFP reporter cells were transfected with ISce-I and siRNAs targeting CDK13, CCNK, CDK12 or BRCA2 and GFP fluorescence was measured. A representative assay of n=3 independent experiments performed in technical triplicates is shown. Error bars represent SEM in A-B, D-E. Student’s t test was performed for A. ANOVA with multiple comparison correction (Sidak) was performed for E.

Figure 4:. Kinase domain and truncation mutants differentially affect PARPi sensitivity.

(A) Lollipop diagram showing kinase domain point mutations in CDK12, including the two more frequently occurring mutations, R858W and D918G. (B) WB showing expression of LacZ, CDK12 WT, CDK12 R858W and CDK12 D918G in LNCaP control and CDK12 KO cells. GAPDH is shown as the loading control. All proteins were generated with an amino terminus V5 tag. (C) Dose response assay to rucaparib in LNCaP control and CDK12 KO cells expressing WT, R858W or D918G point mutants. (D) Schematic of full-length CDK12 and truncation mutants (KDΔ, PRMΔ, and RSΔ). The number to right represents the last amino acid of full-length CDK12 and each mutant. (E) WB of expression of truncation mutants in LNCaP cells. GAPDH is shown the loading control. (F) Dose response assay to rucaparib in LNCaP control and CDK12 KO cells expressing WT, KDΔ, PRMΔ, and RSΔ truncation mutants. (G) Immunoprecipitation of doxycycline-inducible FLAG-tagged WT, KDΔ, PRMΔ, and RSΔ mutants. Eluate was blotted for FLAG and endogenous cyclin K. The input for cyclin K is shown. For panels C and F, a representative assay of n=3 independent experiments performed in technical triplicates is shown. Error bars represent SEM in C and F.

Figure 5:. CDK12 KO prostate cancer xenografts are more sensitive to PARPi monotherapy.

(A) Tumor growth measurements in C42B control and CDK12 KO tumor xenografts, treated with or without rucaparib (150 mg/kg daily). n = 8 mice per group, graph depicts mean tumor size $\pm$ SEM. * denotes p<0.05 and ** denotes p<0.01. (B) The mean percentage change of tumor volume for C42B control and CDK12 KO tumor xenografts, treated with or without rucaparib. (C) WB of tumor xenografts for CDK12 and phosphoSer2-RNA Pol II. (D) Immunohistochemical images for CDK12 in control and CDK12 KO tumor xenografts after rucaparib treatment. Scale bar = 100μm. (E-F) Representative images of γH2AX (E) and phospho-histone H3 (F) immunohistochemical images and quantification (right). n=4 fields per tumor, 5 tumors per cohort, ns=not significant, **p<0.01, ****p<0.001. Errors bars represent SEM in A-B. Horizontal bars represent the mean in E-F. ANOVA with multiple comparison correction (Sidak) was performed for A-B and E-F.

Figure 6:. Responses to rucaparib in prostate cancer patients varies by the type of CDK12 mutation.

(A) Waterfall plot of best percent change in serum prostate specific antigen (PSA) levels in patients with monoallelic (light shaded gray), gene arrangements (dark gray) or biallelic (black) CDK12 mutations treated with rucaparib monotherapy on the TRITON2 trial. (B) Four of 6 patients with measurable disease showed stable disease (SD), while 2 of 6 patients showed progressive disease (PD) by RECIST. (C-D) PSA values in patient 1 who achieved a PSA50 response (C) and patient 2 who did not achieve a PSA50 response (D) treated with rucaparib monotherapy on the TRITON2 trial. The baseline PSA value for each patient is indicated by the dotted line.