Synthetic Lethal Targeting of Cyclin Dependent Kinase-12-Deficient Prostate Cancer with PARP Inhibitors

Animal studies.

NSG (NOD/SCID/gamma) mice were housed in the UCSF barrier facility under pathogen-free conditions and were obtained through an in-house breeding core. For tumor cell line xenografts, 1× 10⁶ cells were injected subcutaneously into each flank (bilateral) of 7–10 week old male NSG mice. The injected cells were resuspended in 1:1 serum-free media and Matrigel (BD Biosciences). Tumor measurements were performed once per week with digital calipers, and the volume of each tumor was calculated based on the formula: V=0.52*length*width², where the width represents the shorter axis. Mice were enrolled into treatment groups once tumor volumes reached between 50–150 mm³. Rucaparib camsylate (Clovis) was reconstituted in 0.5% methylcellulose (Spectrum) as per the manufacturer’s recommended concentration and administered by oral gavage (per os) at 150 mg/kg daily (Monday – Friday) for 3 weeks.

Cell culture.

LnCAP, PC3, C42B, TRAMP-C2 and DLD-1 cells were obtained from the ATCC. OVCAR8, CAOV3 and MDA-MB-231 cells were obtained from the UCSF Cell Culture Facility. Cells were grown in standard conditions with antibiotics (RPMI-1640 with 10% FBS for LnCAP, PC3, C42B, DLD-1, OVCAR8 and CAOV3 cells; DMEM with 10% FBS for MDA-MB-231 and TRAMP-C2 cells) and routinely tested for mycoplasma every 3 months (Lonza MycoAlert Mycoplasma Detection Kit) throughout the study, and most cell lines were lasted tested in November 2023. Cell lines were all validated by STR DNA profiling (UC Berkeley DNA Sequencing Facility), and were used within 40 passages from the time of STR DNA profiling for all experiments shown.

Plasmids and cloning.

pSpCas9(BB)-2A-GFP (PX458) (Addgene plasmid # 48138) and lentiGuide-Puro (Addgene plasmid # 52963) were gifts from Feng Zhang. The pHR-SFFV-dCas9-KRAB-BFP and pCRISPRia_v2 plasmids (Addgene plasmid #46911 and 84832) were gifts from Drs. Stanley Qi, Luke Gilbert and Jonathan Weissman. The pDONR221 plasmid was from Thermo Fisher Scientific. Cloning was performed using standard protocols. sgRNA sequences to target CDK12, CDK13, and CCNK are provided in Supplementary Table S3. Gel purification was performed using the NucleoSpin Gel and PCR Clean-up kit (Takara Bio Inc.) according to the manufacturer’s protocol. Annealed sgRNAs were ligated into the plasmid using the Quick Ligation Kit (New England BioLabs) according to the manufacturer’s protocol. For Gateway cloning, BP and LR Clonase II (Thermo Fisher Scientific) reactions were performed according to the manufacturer’s protocol.

Generation of CRISPR Interference (CRISPRi) knockdown cells.

CRISPRi cell lines were generated as previously described². Briefly, cells were transduced with dCas9-KRAB-BFP and sorted by FACS (BD Fusion) for BFP positivity. Cells were expanded and then sorted at least once again for BFP to obtain an optimized pure population. The dCas9-KRAB expressing cells were then transduced with lentivirus containing the sgRNA targeting the gene of interest. After 2–3 days, cells were selected in 3–5 μg/ml of puromycin for at least 3 days. Cells were used for experiments up to 8 passages and kept in puromycin to maintain a stable knockdown of the gene of interest.

Generation of CDK12 and CDK13 knockout cells.

Cells were transfected with the PX458 plasmid containing the guide targeting the gene of interest using Lipofectamine 3000 (Thermo Fisher) according to manufacturer’s instructions. 48 – 72 hours after transfection, cells were sorted for green fluorescence protein (GFP) expression, and then into single cells in a 96-well plate format. Clones were expanded and validated by Western blotting and sequencing the target site.

Small molecule compound screening.

50 uL cell suspension/250 cells per well were seeded into 384 well white plates (Greiner) using a WellMate bulk dispenser and incubated overnight prior to addition of 50 nL Bioactive Small Molecule Library (SelleckChem) by 384 pin tool (VP Scientific). Stock compound concentrations were 10, 1 and 0.1 mM in DMSO, yielding final assay concentrations of 10, 1 and 0.1 μM (0.1% DMSO). Compounds were pre-arrayed in columns 3 through 22 (320 per plate), with columns 1, 2, 23 and 24 (untreated negative controls, n=32) receiving DMSO only. Matched plate pairs (control reference and CDK12 KO) were assayed for viability after 72 and 144 hours using Promega CellTiter-Glo according to the manufacturer’s instructions. Luminescence was recorded using an Envision Xcite plate reader (Perkin Elmer).

Data normalization.

Compound luminescence counts were normalized to fold change values using respective plate mean negative controls (n = 32). Per plate significance limits were obtained from mean negative control $\pm$ 3 x SD of the mean, with compound fold change values < Mean-3SD curated as significantly outside control variation (viability hits).

Data reduction.

Fold change ratios (Control/CDK12 KO) were calculated to identify compounds with selectivity relative to reference genetic background. Ratio significance limits were established using aggregate mean +/− 3 x SD of the mean for all normalized results, with ratios > Mean Ratio+/− 3SD indicating significantly higher selectivity in CDK12 KO cells.

Drug dose-response assays.

Cells were seeded in a 96-well plate at a density of 1,000 cells per well (C42B, PC3, DLD-1, MDA-MB-231) or 2,000 cells per well (LNCaP, CAOV3, OVCAR-8). After 24 hours, drug or DMSO was added to the cells. The following drugs and concentrations were used: rucaparib, olaparib and veliparib (4-fold serial dilutions, starting at 50μM to 0.763nM), and talazoparib (4-fold serial dilutions starting from 10μM to 0.153nM). Cells were exposed to drug for seven days. On the last day of drug treatment, cells were incubated with Cell Proliferation Reagent WST-1 solution (Roche) according to the manufacturer’s protocol. Absorbance values were obtained using an Epoch Microplate Spectrophotometer with Gen5 software, normalized to the DMSO control, and values were plotted and analyzed using Prism 9 software (GraphPad).

Lentiviral production.

Lentiviral production was carried out using calcium-phosphate-mediated transfection of HEK293T-Lenti-X cells using standard procedures. Lentivirus was concentrated using the Lenti-X concentrator (Takara Bio Inc.) according to the manufacturer’s protocol. Stably transduced cells were selected with puromycin or blasticidin for at least 5 days or selected by FACS.

Flow cytometry.

Cells were trypsinized using standard procedures. Prior to sorting, cells were passed through a 40μm cell strainer. For single cell cloning, single cells were sorted into individual wells in a 96-well plate and expanded over 3–6 weeks. Analysis and cell sorting were performed on the Attune analyzer or FACS Aria III (Becton Dickinson), and analysed using FlowJo (TreeStar) or FACSDiva software (BD Biosciences).

Quantitative PCR (QPCR).

Total RNA was isolated from cells using the Quick RNA Kit (Zymo Research). cDNA was synthesized using the Superscript III RT First Strand Kit (Invitrogen). QPCR was performed using SYBR Green master mix (Roche) in an Applied Biosystems QuantStudio7 machine. Ct values were normalized to actin and GAPDH, and relative expression was calculated using the 2^ddCt method. Primer sequences for QPCR were found using the Harvard Primer Bank and are detailed in Supplementary Table S3.

Immunostaining and histology.

Tissue samples were fixed in 4% PFA overnight and paraffin embedded. Standard hematoxylin and eosin (H&E) staining was performed for routine histology. Antigen retrieval for immunohistochemistry was performed using the BioSB TintoRetriever and citrate buffer. The TSA Amplification Kit (Perkin Elmer #NEL700001KT) was used according to manufacturer’s specifications. Primary antibodies were incubated overnight and secondary antibodies were incubated for 2 hours. The following antibodies were used at the indicated concentrations: CDK12 (Atlas #HPA008038, 1:100), Phospho-Histone H3 (Cell Signaling #9701, 1:200), AR PG-21 (Millipore #06–680, 1:500), γH2AX (Cell Signaling #9718, 1:500), Normal Rabbit IgG (#3900, 1:100), Biotinylated anti-rabbit (Jackson #111-065-144, 1:1000). The DAB developing kit (Vector Laboratories) was used according to instructions.

Western blotting.

Cells were lysed in RIPA buffer the Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific). Protein concentration was measured using the BCA Protein Assay Kit (Thermo Scientific). Lysates were mixed at a 1:1 ratio with Laemmli Buffer (Bio-Rad) supplemented with beta-mercaptoethanol, boiled at 95°C, and then subjected to SDS-PAGE, transferred to PVDF membranes, blocked in 5% w/v BSA, incubated with primary antibody overnight. The next day, Horseradish Peroxidase (HRP) conjugated secondary antibodies were used and blot was visualized using ECL Detection Reagents (Genesee Scientific). Antibodies used include: CDK12 (Atlas #HPA008038, 1:1000 and Cell Signaling #11973, 1:1000), RNA Pol II pSer2 (Active Motif, 1:5000), Lamin B1 (Cell Signaling, 1:1000), GAPDH (Cell Signaling, 1:2000), CDK13 (Bethyl Laboratories, 1:1000), and cyclin K (Bethyl Laboratories, 1:1000). Uncropped blots with ladder markers are provided as Supplementary Fig. S6.

Neutral comet assay.

Neutral comet assay was performed according to the Trevigen protocol. Briefly, cells were trypsinized, combined with low melting agarose and spread on comet slides. Cells were lysed in Lysis Solution for 1 hour at 4°C in the dark, then immersed in neutral electrophoresis buffer for 30 minutes. Comets were ran in a large electrophoresis tank at 21 volts for 45 minutes at 4°C in the dark, then precipitated for 30 minutes at room temperature. Samples were washed in 70% ethanol and dried until flat, then stained with SYBR Gold. Slides were washed, dried and mounted in Prolong Gold, and imaged on a fluorescence microscope. 100–150 comets were imaged per sample, and Tail Moment was calculated using OpenComet in ImageJ.

Clinical trial data (TRITON2).

TRITON2 is an open-label, phase II study evaluating the safety and efficacy of rucaparib in men with mCRPC associated with DDR deficiency. This trial was conducted at 144 centers in 12 countries. Eligible patients at least 18 years old had histologically or cytologically confirmed adenocarcinoma or poorly differentiated carcinoma of the prostate that was metastatic and that had progressed on second-generation, AR-directed therapy (e.g., abiraterone acetate, enzalutamide, or apalutamide) for prostate cancer and one prior line of taxane-based chemotherapy for mCRPC. Disease progression on prior therapy was based on any of the following criteria: rise in PSA (minimum of two consecutive rising levels, with an interval of 1 week or more between each determination; the most recent screening measurement must have been greater than or equal to 2 ng/mL); transaxial imaging with new or progressive soft tissue masses on CT or MRI as defined by RECIST version 1.1; or radionuclide bone scan with at least two new metastatic lesions. Patients were surgically or medically castrated, with serum testosterone levels of 50 ng/dL or less. Patients with and without measurable visceral or nodal disease per RECIST criteria were eligible for the study; patients without measurable disease were required to have PSA levels of greater than 2 ng/mL on the most recent measurement. Patients had an Eastern Cooperative Oncology Group Performance Status of 0 or 1 and adequate organ function. Patients who had prior treatment with any PARP inhibitor, mitoxantrone, cyclophosphamide, or any platinum-based chemotherapy were excluded.

Patients were screened for the presence of a deleterious somatic or germline alteration in BRCA1, BRCA2, ATM, BARD1, BRIP1, CDK12, CHEK2, FANCA, NBN, PALB2, RAD51, RAD51B, RAD51C, RAD51D, or RAD54L through central genomic testing of plasma or tumor tissue (archival or contemporaneous), or through local testing. Central testing was performed by Foundation Medicine, Inc., and mutations in CDK12 were determined by monoallelic, biallelic or gene rearrangements.

Patients received oral rucaparib 600 mg twice daily until confirmed radiographic disease progression as assessed by investigator on the basis of modified RECIST and/or Prostate Cancer Clinical Trials Working Group 3 (PCWG3) criteria, unequivocal clinical disease progression, death, or other reason for discontinuation. Dose reductions were permitted if a patient had a grade 3 or greater or a persistent grade 2 adverse event (AE).

Study approval.

All animal experiments were performed at UCSF, and reviewed and approved by the UCSF IACUC. The TRITON2 clinical trial was approved by national or local institutional review boards and was carried out in accordance with the Declaration of Helsinki and Good Clinical Practice Guidelines of the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. Patients provided written informed consent before participation. In terms of the Representativeness of Study Participants, the demographics of the study population, which included men with prostate cancer, were previously reported (23,24).

Statistical analysis.

Statistical analysis was performed using Prism 9 software (GraphPad Software). All data are presented as mean +/− SEM, unless otherwise stated. When two groups were compared, the two-tailed Student t -test was used, unless otherwise stated. When three or more groups were compared, the one-way analysis of variance (ANOVA) test was used, followed by multiple testing with correction (Sidak) to determine significance between groups, unless otherwise stated. We considered p <0.05 as significant.

Data availability statement.

The whole exome sequencing (WES) data generated in this study are publicly available in the NIH Sequence Read Archive (SRA) under BioProject ID numbers PRJNA1145981 and PRJNA932332. Data from the TRITON2 clinical trial are available upon request from the corresponding authors. Drug screening data and western blots can be found within the article and its supplementary data files, and raw data for plots are available upon request from the corresponding authors. Plasmids and cell lines described in this study are available upon request from the corresponding authors.

A small molecule screen in CDK12 KO prostate cancer cells identifies multiple PARP inhibitors amongst the top selective hits

To generate pre-clinical prostate cancer models of CDK12-deficiency, we first generated single-cell derived CDK12 knockout (KO) clones using CRISPR/Cas9 targeting exon 1 of CDK12 in multiple prostate cancer cell lines, including LNCaP and C42B cells. These clones were confirmed to be deficient in CDK12 by western blotting (Fig. 1A–1B). These cells were further validated by whole-exome sequencing (WES) to demonstrate that we had successfully generated indels in exon 1 of CDK12 in LNCaP and C42B cells, leading to a frameshift and early truncation, which recapitulate the frameshift mutations found in CRPC patients (Supplementary Fig. S1A–S1C) (12). In cell culture, the CDK12 KO cells grew slightly slower than the isogenic wildtype control cells (Supplementary Fig S2A–S2B).

To identify compounds selectively targeting CDK12-deficient cells, we utilized an isogenic pair of C42B cells to screen a commercially-available library containing approximately 1800 bioactive, FDA-approved compounds at 3 concentrations (10 μM, 1 μM and 0.1 μM). We used CellTiter-Glo and calculated the ratio of the control cell luminescence to CDK12 KO cell luminescence (after subtracting the background) for each compound and concentration and identified candidates that selectively killed the CDK12 KO if the ratio was $\ge$2. We focused on the 168 hour time point which had more hits. We found that among the top hits, multiple compounds belonged to the DNA damage subclass (Supplementary Table 1, Fig. 1C–1D). We were particularly intrigued that among the DNA damage compounds, 2 PARP inhibitors (PARPi), olaparib and rucaparib, were in the top 20 hits; a third PARPi (AG14361), which is chemically related to rucaparib, was also a significant top 40 hit. Given the recent approval of PARPi for CRPC and several ongoing clinical trials investigating the use of PARPi for CDK12-mutated cancers (NCT03012321, NCT04030559 and reviewed in (25)), we validated that CDK12 KO cells were indeed more sensitive to olaparib using dose-response and clonogenic assays (Fig. 1E–1H). We also evaluated the sensitivity to other PARPi, including talazoparib, rucaparib and veliparib, and found that CDK12 KO cells were more sensitive to these other PARPi as well (Supplementary Fig. S2C–S2E), which preferentially inhibited growth in the CDK12 KO cells (Supplementary Fig. S2F). Mouse TRAMP-C2 prostate cancer cells deficient in Cdk12 were also more sensitive to PARPi (Supplementary Fig. S2G–S2H). Taken together, these results demonstrate that CDK12 KO human and mouse prostate cancer cells exhibit enhanced sensitivity to multiple PARP inhibitors.

CDK12 loss in multiple cancer cell types results in PARPi sensitivity in a gene dosage dependent manner

We next utilized catalytically-inactive CRISPR/dCas9 (referred to as CRISPRi)(26) to knockdown (KD) but not completely eliminate CDK12 expression. We tested the silencing efficiency of multiple sgRNAs targeting CDK12 to generate stable KDs (Supplementary Fig. S3A) and chose the most robust sgRNA (sgRNA-2) to express in multiple prostate cancer (e.g., LNCaP and PC3 cells) and a colon cancer (DLD-1) cell line. We used ovarian cancer (OVCAR8 and CAOV3) and breast cancer (MDA-MB-231) cell lines as controls. In all cancer types tested, we found that silencing CDK12 led to PARPi sensitivity, nearly to the same degree as silencing BRCA2 (Fig. 2A–2E, Supplementary Fig. S3B), suggesting that the synthetic lethal relationship between CDK12 and PARPi is observed across multiple tumor cell types. In addition, in prostate cancer lines, PARPi sensitivity was not dependent on expression of androgen receptor (AR), since CDK12 loss in PC3 cells (AR-negative, Fig. 2B), TRAMP-C2 (AR-negative, Supplementary Fig. S2F–S2G), LNCaP (AR-positive) and C42B cells (AR-positive) all showed enhanced sensitivity to PARPi. In the CRISPRi cells, the degree of CDK12 expression inhibition was approximately 70% as measured by western blotting and quantitative PCR (qPCR) (Fig. 2F–2G). Interestingly, when comparing the degree of PARPi sensitization in the LNCaP CDK12 KD versus KO cells, we noticed more pronounced differences in PARPi sensitivity in the complete KO cells compared to the KD and control cells (Fig. 2H). This was also seen with BRCA2, in which DLD1 cells with a complete BRCA2 KO were more sensitive to PARPi than BRCA2 KD cells (Supplementary Fig. S3C). Cells in which only one allele of CDK12 was edited by CRISPR/Cas9 (i.e. a monoallelic deletion), resulting in at least one remaining intact, wildtype, allele of CDK12, were not more sensitive to PARPi, although levels of CDK12 were mildly reduced (Fig. 2I–2J). Taken together, these data demonstrate that CDK12 loss in multiple cancer cell types leads to PARPi sensitivity, which is dependent on the residual gene dosage of CDK12; cells with complete CDK12 KO are more sensitive to PARPi than cells with partial CDK12 loss.

CDK12 is necessary for homologous recombination repair activity and gene expression

We next tested the functional significance of CDK12 loss on homologous recombination (HR) repair activity. We silenced CDK12 in the U2OS-DR-GFP cells, which carry a HR reporter (27), and found that CDK12 inhibition decreased reporter fluorescence after I-SceI transfection, nearly to the same degree as BRCA2 inhibition (Fig. 2K). Interestingly, treating U2OS-DR-GFP cells with THZ531, a covalent CDK12/13 inhibitor (28), also suppressed HR reporter activity in a dose-dependent manner (Fig. 2L). At baseline, we did not detect more endogenous double-stranded breaks in CDK12 KO cells. However, PARPi treatment specifically increased double-stranded breaks in the CDK12 KO cells, as measured by comet assays (Fig. 2M). In accordance with prior studies showing an enrichment of DNA repair pathway genes affected in the setting of CDK12 inhibition (17–19,29–31), treating C42B PC cells with THZ531 (250 nM) also suppressed expression of multiple genes involved in DNA repair, including BRCA1 and BRCA2 (Fig. 2N). Interestingly, BRCA1, BRCA2 and other DNA repair genes were not consistently downregulated in the CDK12 KO cells, suggesting possible compensatory mechanisms to restore their expression in the chronic versus acute CDK12 deficient conditions (Supplementary Fig. S3D–S3E). Taken together, these results demonstrate that CDK12 is necessary for optimal HR activity and that pharmacologic inhibition, but not chronic loss, of CDK12 affects the expression of multiple DNA repair genes.

Loss of cyclin K, but not CDK13, impairs HR and results in PARPi sensitivity

The kinase domains of CDK12 and CDK13 are 92% identical, and both CDK12 and CDK13 bind to same cognate cyclin, cyclin K (32). Although CDK12 and CDK13 have been shown to have distinct functions (33), both depend on interactions with cyclin K to execute their kinase functions (14). We silenced cyclin K (CCNK) and CDK13 using CRISPRi (Fig. 3A) and found that inhibition of cyclin K, but not CDK13, conferred PARPi sensitivity (Fig. 3B). In addition, we generated single-cell derived CDK13 KO clones (Fig.3C) and found that in contrast to the CDK12 KO cells, the CDK13 KO cells were not more sensitive to PARPi and behaved similarly to control cell lines (Fig. 3D). Accordingly, silencing CCNK but not CDK13 suppressed HR reporter activity (Fig. 3E). Taken together, these data demonstrate that despite highly similar kinase domains, CDK13 does not affect HR activity or PARPi sensitivity, suggesting that the CDK12-Cyclin K heterodimer is specifically critical for these functions. Interestingly, mutations and deletions of cyclin K are found in multiple cancer types, including cervical, endometrial and esophageal squamous cell carcinoma (Supplementary Fig. S4), suggesting that additional tumor types may also been sensitive to PARPi.

CDK12 kinase domain and truncation mutants differentially affect PARPi sensitivity

Given the importance of the kinase domain, we next asked whether two recurrent kinase domain point mutations identified in PC patients (R858W and D918G, which are labeled as variants of unknown significance (VUS) on cBioPortal) altered PARPi sensitivity (Fig. 4A). The R858 residue is one of three critical arginine residues located within the active kinase center that stabilizes the T-loop upon phosphorylation of a critical threonine residue T893 (34,35). To test the function of these mutants, we constitutively expressed wildtype (WT) CDK12, and the R858W and D918G kinase domain point mutants in LNCaP CDK12 KO cells (Fig. 4B). Expression of LacZ was used a control. While expression of CDK12 WT (magenta) completely restored PARPi resistance, expression of the R858W (yellow) and D918G (green) point mutants (which likely have reduced kinase activity) resulted in a partial rescue of PARPi sensitivity from an IC₅₀ of 0.003 μM to an IC₅₀ of 0.05–0.10 μM (Fig. 4C). In addition, the R858W and D918G point mutants partially rescued PARPi sensitivity in the C42B CDK12 KO cells (Supplementary Fig. S5A).

Next, we expressed truncated forms of CDK12, which are commonly found in prostate cancer patients (12), and made deletions C-terminal to the known functional protein domains: RSΔ (amino acids 1 – 386), PRMΔ (amino acids 1 – 700) and kinase domain (KDΔ) (amino acids 1 – 1020) (Fig. 4D–4E). None of the truncation mutants were sufficient to fully restore PARPi sensitivity in LNCaP or C42B CDK12 KO cells, with IC₅₀ ranges of 0.002 – 0.004 μM compared to the WT, which shifted the IC₅₀ to 0.3 μM (Fig. 4F, Supplementary Fig. S5B). These data suggest that not only is the kinase domain important in mediating PARPi sensitivity, but that the carboxy terminal of CDK12 beyond the kinase domain is also critical. Indeed, prior studies demonstrated that a carboxy terminal extension from the CDK12 kinase domain which is located between amino acids 1044–1063 was important for kinase activity (34). Interestingly, the majority (96%, 262/271 examined in cBioportal) of truncating mutations in PC reported occur prior to the carboxy terminal extension (amino acid 1063), further suggesting the importance of the carboxy terminal extension. The phosphorylation levels of serine 2 (Ser2) RNA Pol II were slightly reduced or similar in the CDK12 KO, point mutants or truncation mutants cells compared to control, while levels of serine 5 (Ser5) and serine 7 (Ser7) phosphorylation were overall similar (Supplementary Fig. S5C–5D). Finally, we asked whether the truncated forms of CDK12 were able to interact with cyclin K, since the interaction with cyclin K is critical for CDK12 kinase activity. Co-immunoprecipitation (co-IP) experiments using doxycycline-inducible FLAG-tagged CDK12 WT, KDΔ, PRMΔ and RSΔ constructs demonstrated that none of the truncated mutants were able to interact with cyclin K (Fig. 4G). Taken together, these data demonstrate that CDK12 kinase domain mutants and truncation mutants are unable to fully restore PARPi sensitivity. In addition, CDK12 truncation mutants cannot interact with cyclin K, which is a critical partner for the kinase function of CDK12, although there may also be additional unknown kinase-independent functions of CDK12.

CDK12 KO prostate cancer xenografts are sensitive to PARPi monotherapy

We next asked whether CDK12 KO tumor xenografts were sensitive to PARPi monotherapy in vivo. Rucaparib treatment inhibited tumor growth in C42B CDK12 KO xenografts, whereas treatment of control C42B xenografts showed slight growth inhibition that was not statistically significant (Fig. 5A–5B). We confirmed that tumors explanted at the end of the experiment remained null for CDK12 by western blot and immunohistochemistry (Fig. 5C–5D). CDK12 KO tumors also had only slightly reduced levels of phosphorylated Ser2-RNA Pol II (Fig. 5C). Rucaparib treatment also increased γH2AX staining (marker of double-strand DNA breaks) in the CDK12 KO, but not control CDK12 WT tumors and decreased phospho-histone H3 staining, a marker of proliferation, consistent with enhanced sensitivity to PARPi in CDK12 KO tumors in vivo (Fig. 5E–5F). Taken together, our data demonstrate that PARPi monotherapy inhibits CDK12 KO tumor growth in vivo, and is associated with increased markers of DNA damage and decreased proliferation.

Patients with CDK12-mutated prostate cancer exhibit responses to rucaparib monotherapy

Finally, we asked whether prostate cancer patients with CDK12 alterations had any clinical benefit from PARPi monotherapy. TRITON2 (NCT02952534) was a phase 2 clinical trial evaluating rucaparib monotherapy in patients with advanced castrate-resistant prostate cancer (CRPC) who progressed after receiving an androgen pathway inhibitor (e.g., enzalutamide or abiraterone) and taxane chemotherapy (23). A total of 15 patients were included in the CDK12 cohort (Supplementary Table S2) (24). We found that 11 patients had biallelic CDK12 alterations (which included two patients with a single detectable CDK12 mutation but had evidence of loss of heterozygosity (LOH) of the second wildtype allele based on sequencing). Of the biallelic mutated patients, 7 of the 11 patients (63.6%) had a reduction in serum prostate specific antigen (PSA) levels from baseline (range −1.4% to 74.9%, Fig. 6A), and 5 of the 11 patients (45.4%) had a PSA30 response or better (representing a PSA reduction of 30% or more from baseline after starting rucaparib). Of the patients with biallelic CDK12 alterations and evaluable lesions on computed tomography (CT) scans, 4 of 6 patients (66.6%) had radiographic disease control, with stable disease (SD) as their best response per RECIST (response evaluation criteria in solid tumors) (Fig. 6B). We also analyzed the PSA trends for individual patients who responded to rucaparib. Patient 1 had a greater than PSA50 response and the PSA remained below the initial baseline for 32 weeks (Fig. 6C). Meanwhile, Patient 2 had a less than PSA50 response, but the PSA remained below the initial baseline for 39 weeks (Fig. 6D). Although these responses do not meet the clinical response criteria or primary objectives pre-specified in the study design of TRITON2, the data suggest that rucaparib monotherapy may have activity in heavily pre-treated patients with CDK12 alterations, who would otherwise have few remaining therapeutic options at this advanced stage of disease. Interestingly, one patient who had a single CDK12 mutation (whom we classified as monoallelic, since only one mutation was detected and there was no evidence of LOH of the second allele) did not have a PSA response. In addition, 0 of 3 patients with a CDK12 gene arrangement (with indeterminate zygosity) had a PSA30 response (Fig. 6A, Supplementary Table S2). Taken together, these data demonstrate that PARP inhibitors may have modest anti-tumor activity in advanced CRPC patients with CDK12 alterations, with responses observed only in the subset of patients with biallelic alterations.