Abstract
Patients with metastatic pancreatic ductal adenocarcinoma (PDAC) have an average survival of less than 1 year, underscoring the importance of evaluating novel targets with matched targeted agents. We recently identified that poly (ADP) ribose glycohydrolase (PARG) is a strong candidate target due to its dependence on the pro-oncogenic mRNA stability factor HuR (ELAVL1). Here, we evaluated PARG as a target in PDAC models using both genetic silencing of PARG and established small-molecule PARG inhibitors (PARGi), PDDX-01/04. Homologous repair–deficient cells compared with homologous repair–proficient cells were more sensitive to PARGi in vitro. In vivo, silencing of PARG significantly decreased tumor growth. PARGi synergized with DNA-damaging agents (i.e., oxaliplatin and 5-fluorouracil), but not with PARPi therapy. Mechanistically, combined PARGi and oxaliplatin treatment led to persistence of detrimental PARylation, increased expression of cleaved caspase-3, and increased γH2AX foci. In summary, these data validate PARG as a relevant target in PDAC and establish current therapies that synergize with PARGi.
PARG is a potential target in pancreatic cancer as a single-agent anticancer therapy or in combination with current standard of care.
Introduction
Pancreatic ductal adenocarcinoma (PDAC) is estimated to become the second leading cause of cancer-related death in the United States, with a 5-year survival rate of only 9% (1, 2). Despite recent advancements in treatment options for PDAC, the prognosis remains poor. In 2008, global genomic analyses on human pancreatic cancers revealed 12 core signaling pathways that were altered in the majority of PDAC (3–5), and subsequent sequencing studies have supported and expanded upon these initial findings (6, 7). Initial results from the “Know Your Tumor Initiative” for patients with PDAC, as well as several related publications (5–8), have identified certain patient populations with 27% highly actionable mutations. Actionable mutations commonly found were in DNA repair genes (BRCA1/2 or ATM; 8.4%) and cell-cycle genes (CCND1/2/3 or CDK4/6; 8.1%; refs. 6, 7). Germline or sporadic mutations in homologous recombination genes found in PDAC and several other cancers compromise repair of damaged DNA, most likely helping to facilitate tumorigenesis (8–11). Selectively targeting alternative DNA repair pathways in such tumors provides a putative “personalized” therapeutic strategy via a synthetic lethal approach (12, 13).
In patients carrying germline or somatic defects in homologous-repair (HR) genes, DNA repair is heavily dependent on PARPs (e.g., PARP1; refs. 7, 11). Hence, these tumors are highly susceptible to PARP inhibitors (PARPi; refs. 13–15). Recent data are also emerging with the use of PARPi in the treatment of PDAC (16–20). Many clinical trials are ongoing for locally advanced or metastatic PDAC, either with single-agent PARPi (olaparib, veliparib, talazoparib, rucaparib) or in combination with standard chemotherapy, including gemcitabine, FOLFIRINOX, and cisplatin (16–20). A recent dose-escalation, phase I trial with talazoparib as a single agent confirmed responses in patients with PDAC (21). Three PARPi are approved for patients with metastatic breast and ovarian cancer, and are being used routinely as standard of care (22, 23). However, while PARPi can be very effective, resistance to PARPi is an evolving concern (24–26). In addition, it is becoming increasingly apparent that not all tumors that share phenotypic features of BRCA1/2–mutant tumors (BRCAness) respond to PARPi therapies. Besides conventional resistance mechanisms (26), recent studies from our group showed that PDAC cells develop resistance to DNA-damaging agents through post-transcriptional upregulation of WEE1 and poly (ADP-ribose) glycohydrolase (PARG), mediated by the pro-oncogenic, RNA-binding protein, HuR (27, 28).
While WEE1 has been explored as a target in PDAC by us and others (27, 29), PARG has been explored by others as a target in colon cancers, lung cancers, and acute myeloid leukemia (30–32). PARG is an enzyme that breaks down poly (ADP-ribose) (PAR) chains for a number of different cellular processes including the following: release and recycling of repair proteins, break repair resolution, and replication fork progression. PAR chains are added by the PARP enzymes as covalent modifications to acceptor proteins, primarily in response to DNA damage by PARP1 and PARP2 (33, 34). This reversible post-translational modification is critical for signaling and recruitment of repair factors, and has typically been targeted via PARPi to prevent or weaken the DNA damage response (28, 33–35). Other studies have demonstrated that inhibiting PARG could: (i) sensitize cells to radiation (36); (ii) slow down DNA repair and cause mitotic abnormalities (37); and (iii) synergize with silencing of BRCA1, BRCA2, PALB2, FAM175A (ABRAXAS), and BARD1 genes to increase DNA damage and stall replication forks (38, 39). Finally, we recently showed that the pro-oncogenic factor, HuR, upregulates PARG in a cancer-specific manner (28), providing a strong rationale (i.e., a therapeutic window) to explore targeting PARG in PDAC cells. Thus, we hypothesized that targeting PARG could provide an alternative and complementary therapeutic strategy to treating PDAC (38, 40). To date, no one has explored targeting PARG in both homologous repair–deficient and -proficient (HR-D and HR-P) PDAC cells due to a lack of available specific and potent PARGi. Herein, we evaluate for the first time targeting PARG, through both genetic silencing and small-molecule inhibitors (38, 41, 42) in preclinical models of PDAC.
Materials and Methods
Cell lines
HR-P: MIA PaCa-2, PANC-1, and HR-D: Hs 766T PDAC cells were obtained from ATCC. Normal HPNE cells were purchased from ATCC. DLD1.BRCA2 and RKO.FANCC isogenic colorectal cell lines were a generous gift from Dr. Scott Kern (Johns Hopkins University School of Medicine, Baltimore, MD) and were generated by targeted disruption of either BRCA2 or FANCC (Fanconi Anemia pathway) DNA repair genes by homologous recombination and provide excellent cancer models for studying drug sensitivity in a HR-deficient background (43, 44). PDAC PDX–derived cells were obtained from Dr. Golan (Tel Aviv University, Tel Aviv, Israel; Supplementary Table S1A; ref. 45). These cells were derived from pancreatic ascites or pleural effusion cancer cells from patients with PDAC and represent clinically relevant models to study PARG as a target because many of these models recapitulate metastatic PDAC, and have been characterized for BRCA1/2 mutations and KRAS status. KPC BRCA2 wild-type (WT) and BRCA2 null cell lines were a kind gift from Dr. Kenneth P. Olive (Columbia University, New York, NY; refs. 46, 47). PDX cell lines were cultured in RPMI in a humidified incubator at 37°C and 5% CO2, as recommended. All other cell lines were cultured in DMEM medium supplemented with 10% FBS, 1% l-glutamine, and 1% penicillin–streptomycin. All cell lines were short-tandem repeat–authenticated, and were Mycoplasma-tested monthly, using PCR-based Mycoplasma Detection Kit (# MP0035, Sigma Aldrich). We also evaluated cell lines for maintenance of a defining oncogenic driving mutation (i.e., Kras). Cells were passaged at least twice after thawing before experimental use.
Generation of doxycycline-inducible shPARG
The pCW-shPARG plasmid was prepared using Gibson cloning: A pCW-cas9 tetracycline inducible plasmid (# 50661, Addgene) was enzyme-digested with NheI and BamHI (New England BioLabs) as per the manufacturers’ specifications and run through a 0.75% agarose gel for subsequent purification with the QIAquick Gel Extraction Kit (Qiagen) to extract the TET-inducible pCW vector backbone. An shPARG insert flanked by BamHI enzyme restriction sites and an upstream NheI restriction site was purchased from IDT, PCR amplified using designed cloning primers (forward primer: 5′-cagatcgcctggagaattggGGATCCGCTAGCGCCACC-3′ (17), reverse primer: 5′-aaggcgcaaccccaaccccgGGATCCCAAAAAGCGATCTTAGGAAAC-3′) and gel-purified as detailed previously. The insert was then Gibson-cloned into the vector using the NEBuilder HiFi DNA Assembly Cloning Kit (# E5520S, New England BioLabs) at a vector and insert count of 0.05 picomols and 0.12 picomols, respectively. The newly formed plasmid was used to transform NEB 5-alpha–competent cells (# C2988J, New England BioLabs). After a 24-hour incubation at 37°C, individual clones were then selected, cultured, and sequenced for validation using sequencing primers (primer 1: 5′-GGGCTGCCTTGGAAAAG-3′, primer 2: 5′-CAGATCGCCTGGAGAATTG-3′).
Lentivirus production and generation of shPARG cell lines
A commercially available unique 29mer shRNA construct against human PARG in a lentiviral GFP vector (# TL310610, Origene) was packaged in HEK 293T cells using a Lentiviral Packaging Kit (# TR30037, Origene). MIA PaCa-2 and PANC-1 cells were infected with virus particles with the addition of 1 μg/mL Polybrene (# TR-1003-G, Sigma Aldrich). Cells were then puromycin (# P8833, Sigma Aldrich) -selected and validated for PARG mRNA and protein knockdown.
Modified siRNA for transient knockdown
MIA PaCa-2 and PANC cells were transfected with custom-made modified control and PARG siRNA oligonucleotides (Genisphere LLC; PARG sense 5′-UACCAGAGCAGUUUAGUAA-3′). Transfections with unmodified ON-TARGET plus Nontargeting Control siRNA (GE Dharmacon, #D-001810-01-05) and ON-TARGET plus PARG siRNA (GE Dharmacon, # LU-011488-00-0002) were used as controls. All transfections were performed using 30 nmol/L of oligonucleotides and Lipofectamine 2000 (Life Technologies, #11668) according to the manufacturer's instructions.
Drugs and inhibitors
Olaparib (PARPi #S1060) and oxaliplatin (# S1224) were obtained from Selleck Chemicals. 5-Fluorouracil (5-FU) (# F6627) was obtained from Sigma Aldrich. Cell active small-molecule PARG inhibitors PDDX-001/PDD00017273 (hereafter referred to as PDDX-01) and PDDX-02/PDD00017238 (hereafter referred to as PDDX-02) were synthesized as described previously (42, 48) and were utilized because they have been reported to demonstrate robust pharmacology against PARG with great potency and specificity, as compared with other known inhibitors of PARG. PDDX-004/PDD00017272 (hereafter referred to as PDDX-04) was synthesized in collaboration with Dr. Joseph Salvino from The Wistar Institute (Philadelphia, PA) and this compound has the exact quinazolinedione core as PDDX-01 as published before (42). Compounds were either resuspended in DMSO or water to stock concentrations of 10 μmol/L.
Cell survival and combination analyses
A total of 800 cells/well were plated in 96-well plates in 100 μL of cell culture medium, and treated with drugs after 24 hours at indicated concentrations. Percentage cell survival relative to control treatments was assessed after 5 days, by staining cells with Quant-iT Pico Green Reagent (#P7581, Life Technologies) for 1 hour and measuring fluorescence using a fluorescence microplate reader (excitation ∼480 nm, emission ∼520 nm) (49). IC50 values were determined through nonlinear regression analysis.
Cell survival for siDDR experiments was analyzed by MTT assay. A total of 1,000 cells/well were plated in 96-well plates in 100 μL of cell culture medium, and treated with drugs after 24 hours at indicated concentrations. After 5 days, MTT reagent (5 mg/mL in 1×PBS) was added to each well and plates were incubated for 4 hours at 37°C. Following incubation, media was removed and purple formazan crystals were dissolved in DMSO (100 μL). Absorbance was read at 570 nm and results were plotted as percentage relative survival.
Combination analyses were performed by plating cells in 96-well plates at 1,000 cells/well. After 24 hours, cells were treated with PDDX-01 (dose range, 1.5 μmol/L–25 μmol/L) and either olaparib (dose range, 1.5 μmol/L–25 μmol/L), oxaliplatin (dose range, 0.03 μmol/L–1 μmol/L), or 5-FU (dose range, 0.75 μmol/L–25 μmol/L) in a 5 × 5 or 5 × 6 matrix. Each treatment was done in triplicate wells. Cells were treated for 5 days, and percentage cell survival relative to control treatments was assessed by staining cells with Quant-iT Pico Green Reagent. Each experiment was done at least three times. After Pico Green analysis, synergistic/antagonistic/additive combinations were analyzed by using Combenefit software, which provides synergy distribution plots by comparing experimental data to mathematical models (e.g., bliss model) of dose responses for additive/independent combinations (50). In brief, the software first reads each experimental dose response as a matrix of percentage of the control and each single-agent effect is fitted with a dose–response curve. Bliss Independence model then compares experimental combination dose–response surface to the model-generated dose response, resulting in a synergy distribution plot. The Bliss Independence model calculates the expected effect of two drugs as follows: |$Effec{t_{A + B}}\ = \ Effec{t_A} + Effec{t_B} - ( {Effec{t_A} \times Effec{t_B}} )$| where A and B represent two different drugs (51). If the actual effect is greater than that calculated with the Bliss Independence model, this implies synergism = indicates additivity, and less than indicates antagonism. Combenefit determines synergy as the change in efficacy of a combination of drugs as compared with expected. So, for instance, if two drugs through the Bliss model are expected to kill 50% of cells but instead kill 75, this would be a synergy score of 50 as at this combination there is a 50% increase in the expected effectiveness of the combination. Drug matrix heatmap 5 × 5 or 5 × 6 grid illustrating bliss index and percentage inhibition are shown for n = 3.
Colony formation analyses
Long-term colony formation assays were performed as described previously (52) and colonies were counted using Image J software.
qRT-PCR and mRNA expression analyses
Total RNA was extracted using the RNeasy Mini Kit (Qiagen). cDNA was made using 1,000 μg total RNA using Applied Biosystems High Capacity cDNA Reverse Transcriptase Kit (Life Technologies) and qPCR (qRT-PCR) was performed as described previously (49). Relative quantification was performed using the 2-ΔΔCt method.
Immunoblot analyses
Cells were lysed in ice cold RIPA Buffer (#sc-24948A, Santa Cruz Biotechnology Inc.) supplemented with fresh protease inhibitors (#78430, Life Technologies), immunoblotted, and membranes were scanned and quantitated using Odyssey Infrared Imaging System (LI-COR Biosciences) as described previously (49). Primary antibodies used were GAPDH (1:10,000; #2118, Cell Signaling Technology), PARP1 (1:1,000; #sc-365315, Santa Cruz Biotechnology Inc.), PAR (1:1,000; #4335-MC-100, Trevigen), PARG (#NBP2-46320, 1:1,000; Novus), cleaved caspase-3 (1:1,000), caspase-3 (1:1,000), and Histone H3 (#9661, #9662, and #4499, Cell Signaling Technology) followed by LI-COR IRDye secondary antibodies.
Immunofluorescence
A total of 5,000 cells were plated onto coverslips in 24-well tissue culture plates and allowed to adhere for 24 hours. After drug treatment for indicated times and concentrations, coverslips were washed with 1× PBS, fixed (4% paraformaldehyde for 10 minutes at room temperature), permeabilized (0.1% Triton-X), blocked and immunostained with γH2AX 1:500 (#05-636 Millipore Sigma) antibody overnight, 4°C, followed by secondary antibody (Alexa-488 F anti-mouse, #A-10680, Life Technologies) for 1 hour after washing. After the final wash step, coverslips were mounted onto slides using DAPI ProLong Gold Antifade Mounting Medium (#P36931, Life Technologies). Slides were imaged with a Leica DM4B fluorescence microscope and foci were counted using Image J software (28, 53). Percentage of cells expressing >10 γH2AX foci per treatment condition ± SD was calculated and plotted.
PAR ELISA
PARylation was analyzed in total protein lysates using HT Colorimetric PARP/Apoptosis Assay as described previously (28).
Apoptosis assays
Apoptosis was detected by Western blot analysis for cleaved caspase-3 in drug-treated cells at different time points (27).
Chromatin tethering
Treated cells were washed with ice-cold 1×PBS, gently scraped and collected in 1 mL 1×PBS, and pelleted by spinning at 400 × g for 5 minutes. Sequential fractionation was performed with ice-cold 0.1% Triton X-100 in cytoskeletal PIPES buffer as described previously (54). The final pellet (containing chromatin-bound proteins) and total cell pellets were lysed in RIPA buffer. Equal amounts of protein were loaded onto wells and membranes were immunoblotted with primary and corresponding secondary antibodies. Histone H3 was used as a positive control and GAPDH as a negative control for the chromatin-bound fraction.
Xenograft study
Doxycycline-inducible shPARG knockdown model.
Doxycycline-inducible shPARG cell lines were generated for MIA PaCa-2 cells. A total of 3 × 106 cells/flank (MIA.shPARG) were then injected subcutaneously (100 μL) in 5- to 6-week-old Hsd:Athymic Nude-Foxn1nu female mice (Envigo RMS, Inc.). Cells were prepared in 80% DPBS and 20% Matrigel (#356237, Corning Life Sciences). Mice bearing established tumors (50–100 mm3, MIA.shPARG) were randomized into two treatment groups (n = 5 for each group) and one group was administered doxycycline chow (200 mg/kg, Bio-Serv) to downregulate PARG expression.
Statistical considerations.
Mouse weights and tumor sizes were measured three times/week and tumor volumes were calculated using the formula: Volume = (Length × Width2)/2. Relative fold change in tumor volumes was plotted. Relative fold change represents fold change in tumor volume relative to tumor volume on day 1 of randomization of mice into doxycycline/no doxycycline groups. No significant loss in body weight was observed (<5%) for all groups. Animals were euthanized when average tumor volume reached 2,000 mm3.
All mouse protocols were approved by the Thomas Jefferson University Institutional Animal Care and Use Committee.
Statistical analysis
Statistical analyses were performed using the one or two sample t test as indicated in figure legends. GraphPad Prism 7.04 software was used for analysis. Results are expressed as mean ± SEM, if not specifically indicated.
Results
PARG silencing inhibits PDAC tumor growth in vivo and in vitro
In vitro characterization of MIA.shPARG and Lenti.shPARG cell lines confirmed a decrease in mRNA and protein expression of PARG (Supplementary Fig. S1A and S1B). For in vivo–doxycycline-induced shPARG knockdown model, 5- to 6-week-old athymic nude female mice were injected with 100 μL of cell suspension subcutaneously in each flank, and mice were randomized into two groups of: (i) no doxycycline and (ii) doxycycline chow. In the doxycycline chow arm, knockdown of PARG significantly decreased tumor growth in MIA.shPARG (P ≤ 0.05), compared with mice in the no doxycycline group (Fig. 1A). At the conclusion of the study, tumors were harvested (Fig. 1A) and validated for PARG mRNA knockdown. mRNA expression analysis shows PARG levels were significantly decreased in groups fed on doxycycline chow with no change in a negative control, PARP1 mRNA expression (Fig. 1B). Lentivirus shPARG cell lines also show a decrease in growth rate as compared with a shScramble (control) cell line in vitro (Fig. 1C).
Knockdown of PARG with shPARG decreases PDAC tumor growth in vivo and in vitro. A, Left, athymic nude mice bearing MIA.shPARG xenografts were randomized into −doxycycline (DOX) and +doxycycline arms, and relative fold change in tumor volume was plotted. *, P < 0.05 using two-sample t tests. Right, representative images of tumor extracted after completion of in vivo study from MIA.shPARG mice (n = 5/group). B, Relative mRNA expression of PARG and PARP in harvested tumors from MIA.shPARG (−/+doxycycline) in vivo. Mean ± SEM; n = 3; *, P = 0.01 to 0.05. C, In vitro fold change in growth of lentiviral.shPARG PANC-1 cells compared with shScramble. Lentiviral shPARG knockdown enhances sensitivity of cells to oxaliplatin (D) and decreases sensitivity to olaparib in PANC-1 (E; top) and MIA PaCa-2 (bottom) cells in vitro. Representative graphs from n = 3 are shown. *, P < 0.05 using two-sample t tests.
Knockdown of PARG with shPARG decreases PDAC tumor growth in vivo and in vitro. A, Left, athymic nude mice bearing MIA.shPARG xenografts were randomized into −doxycycline (DOX) and +doxycycline arms, and relative fold change in tumor volume was plotted. *, P < 0.05 using two-sample t tests. Right, representative images of tumor extracted after completion of in vivo study from MIA.shPARG mice (n = 5/group). B, Relative mRNA expression of PARG and PARP in harvested tumors from MIA.shPARG (−/+doxycycline) in vivo. Mean ± SEM; n = 3; *, P = 0.01 to 0.05. C, In vitro fold change in growth of lentiviral.shPARG PANC-1 cells compared with shScramble. Lentiviral shPARG knockdown enhances sensitivity of cells to oxaliplatin (D) and decreases sensitivity to olaparib in PANC-1 (E; top) and MIA PaCa-2 (bottom) cells in vitro. Representative graphs from n = 3 are shown. *, P < 0.05 using two-sample t tests.
Silencing of PARG in the doxycycline-inducible MIA.shPARG or lentiviral shPARG models sensitized cells to oxaliplatin, as indicated by a decrease in relative cell survival compared with control arms (Fig. 1D; Supplementary Fig. S1C). Confirming previous reports, silencing of PARG decreased the sensitivity of cells to olaparib (Fig. 1E; ref. 55). Collectively, these findings imply that PARG targeting mediates a reduction in PDAC cell survival and tumor growth in vivo.
PARG inhibitors decrease cell survival of cancer cell lines with genetically diverse backgrounds in vitro
Short-term drug sensitivity assays.
After establishing PARG as a therapeutic target from our in vivo studies, we assessed the efficacy of small-molecule inhibitors of PARG (PARGi; ref. 42) on cell survival of PDAC, PDX, and KPC cell culture models (Fig. 2A–D). Both MIA PaCa-2 and PANC-1 HR-P PDAC cell lines were treated with increasing doses of PARGis and olaparib to determine IC50 concentrations. Hs 766T, PDAC PDX, and KPC BRCA2 WT and BRCA1/2–null cell lines were treated with PDDX-01 and olaparib and cell survival was evaluated using a cell survival, Pico Green assay (49). PARG inhibition was especially effective against cells with an HR deficiency (HR-D); Hs 766T, HR-D PDX cells (SPC_122), and KPC BRCA2--null cells were more sensitive to PARGi compared with their counterpart HR-P cells (Fig. 2A–D). Interestingly, with regard to our PDX lines, the cell line that responded the most to PARGi was SPC_122 (validated as BRCA1 mutant) and those that did not respond were SPC_144 (BRCA WT) and SPC_126 (BRCA2 mutant). Of note, SPC_126 (Supplementary Table S1A) is obtained from a liver biopsy from a patient who became clinically resistant to therapy (i.e., previously exposed to PARPi/platinum therapy and recurred, and is also resistant to PARPi therapy in vitro and in vivo; ref. 45). These data suggest that PARGi may not be the best therapeutic option in PARPi-resistant tumors.
PARG inhibition is synthetic lethal with HR deficiency in PDAC cells. A, Pico Green assays were performed after 5 days of treatment with PDDX-01, PDDX-02, and PDDX-04 to analyze relative cell survival in PDAC cells (MIA PaCa-2, PANC-1). B–D, Relative cell survival of MIA PaCa-2 and Hs 766T treated with PDDX-01 over 3, 5, and 7 days; KPC BRCA WT and KPC BRCA2–null cells as well as pancreatic cancer patient-derived xenograft cell lines (SPC_126, SPC_122, SPC_144). E, Synthetic lethality with PDDX-01 and siRNA DDR genes in MIA PaCa-2 and PANC-1 cells. Representative graphs from n = 2 for A and n = 3 for B–E are shown.
PARG inhibition is synthetic lethal with HR deficiency in PDAC cells. A, Pico Green assays were performed after 5 days of treatment with PDDX-01, PDDX-02, and PDDX-04 to analyze relative cell survival in PDAC cells (MIA PaCa-2, PANC-1). B–D, Relative cell survival of MIA PaCa-2 and Hs 766T treated with PDDX-01 over 3, 5, and 7 days; KPC BRCA WT and KPC BRCA2–null cells as well as pancreatic cancer patient-derived xenograft cell lines (SPC_126, SPC_122, SPC_144). E, Synthetic lethality with PDDX-01 and siRNA DDR genes in MIA PaCa-2 and PANC-1 cells. Representative graphs from n = 2 for A and n = 3 for B–E are shown.
In addition, we found that PARG is differentially expressed in PDAC and PDX models (Supplementary Fig. S2A; Supplementary Table S1B). We observed that high PARG–expressing cell lines responded better to PARGi treatment (SPC_122), as compared with cell lines with medium expression of PARG protein (MIA PaCa-2, PANC-1, SPC_144, and others). The “normal” epithelial HPNE cells with low levels of PARG protein expression (Supplementary Fig. S2A; Supplementary Table S1B) did not respond to PARGi, supporting the possible therapeutic window we described previously (28). We did find a perfect correlation (R2 = 1) in PDX models, SPC_122 (high PARG expression/PARPi naïve; Supplementary Fig. S2B), and SPC_126 (low PARG expression/PARPi resistant; Supplementary Fig. S2B) PDX lines, which corroborates previous published findings (55). However, we did not find a correlation between PARG expression and PARPi response across PDAC lines (Supplementary Fig. S2A and S2C), which suggests that defects in HR genes should still be the gold standard biomarker for PARPi therapy.
To analyze the spectrum of PARGi effects in isogenic models, and the possible therapeutic window between normal and somatic mutations found in a tumor, we also utilized genetically modified DLD1 (DLD1.BRCA2) and RKO (RKO.FANCC) colorectal cancer cells (43, 44). We have previously tested these lines against various DNA damage agents and disruption of BRCA2 or FANCC renders them sensitive to oxaliplatin and PARPi (43). Our results show that both BRCA2 and FANCC null cells required lower doses of PARGi to achieve IC50 when compared with isogenic parental cells (Supplementary Fig. S2D).
To further characterize synthetic lethality with PARG inhibition and defects in genes involved in DNA damage response pathways in an isogenic model, we utilized HR-P PDAC cells, MIA PaCa-2, and PANC-1. Using siRNA oligos against BRCA1, BRCA2, ATM, and BARD1, we compared drug sensitivities of olaparib and PDDX-01 in these knock-down conditions to a scramble siRNA control arm. In our screen, as expected, we found an increase in sensitivity to olaparib with BRCA1/BRAC2 knockdown, while only BRCA2 knockdown increased sensitivity to PDDX-01, demonstrating synthetic lethality of PDDX-01 and BRCA2 in PDAC cells (Fig. 2E; Supplementary Fig. S3A). Knockdown efficiencies of all the genes in both MIA PaCa-2 and PANC-1 cells are represented in Supplementary Fig. S3B.
Drug combination analyses.
To test the efficacy of PARGi in combination with other commonly used DNA damaging agents and PARPi (olaparib) in PDAC cells, we utilized Combenefit software (50) to analyze synergy, additivity or antagonism between the combinations. The bliss index of combination analyses shows that PARGi (PDDX-01) did not synergize with olaparib and the combination is mostly antagonistic as seen in the bliss matrix for olaparib and PARGi in both MIA PaCa-2 and PANC-1 cells (Fig. 3A and B). However, when PARGi was utilized with oxaliplatin or 5-FU (Fig. 3A and B), inhibition of PARG synergized with these compounds, with some drug combinations being more synergistic than others. It is interesting to note that single-agent PARGi, although not as effective in HR-P PDAC cells as HR-D cells, acts as a similar sensitizer to other DNA damage agents (oxaliplatin and 5-FU) in these cell lines. There was no synergy seen in HR-D cells HS 766T with either oxaliplatin/PARGi or 5-FU/PARGi combinations (Supplementary Fig. S4A and S4B).
PARGi synergizes with DNA damage agents and not with PARPi. Drug matrix heatmap 5 × 5 (olaparib and 5-FU) and 6 × 5 (oxaliplatin) grid showing Bliss index and dose response as analyzed by Combenefit analysis for MIA PaCa-2 (A) and PANC-1 cells (B). Combinations that are synergistic appear blue on the heatmap.
PARGi synergizes with DNA damage agents and not with PARPi. Drug matrix heatmap 5 × 5 (olaparib and 5-FU) and 6 × 5 (oxaliplatin) grid showing Bliss index and dose response as analyzed by Combenefit analysis for MIA PaCa-2 (A) and PANC-1 cells (B). Combinations that are synergistic appear blue on the heatmap.
Clonogenicity assays (long term assays).
To understand the long-term effects of PARGi on PDAC cells, we performed clonogenic assays. Both MIA PaCa-2 and PANC-1 cells were treated with PDDX-01 at increasing doses, and colonies were stained and counted after 14 days (Fig. 4). PDDX-001 did not significantly reduce colony formation in either MIA PaCa-2 (Fig. 4A) or PANC-1 cells (Fig. 4B), confirming our results from short-term cell survival assays. From our combination analyses, we further tested the combination of PARGi and oxaliplatin/olaparib in both these cell lines. Interestingly, inhibition of PARG by PDDX-01 sensitized both MIA PaCa-2 and PANC-1 cells to oxaliplatin (Fig. 4A and B), both of which are HR-P PDAC cell lines. Similar to the results from combination analyses, PDDX-01 did not synergize with olaparib in decreasing colony formation of PDAC cells. Together, these data indicate that PARGi has activity and efficacy against PDAC cells and could be utilized to increase the sensitivity of oxaliplatin and other DNA damage agents (like 5-FU) but not PARPi (olaparib) for the treatment of PDAC.
PARGi decreases colony formation in PDAC cells in combination with oxaliplatin. MIA PaCa-2 (A) and PANC-1 (B) cell lines were treated with increasing doses of PDDX-01 (μmol/L) alone and in combination with oxaliplatin (oxali) or olaparib (olap), and resulting colonies were fixed and stained with crystal violet solution. Representative images with graphs of relative colonies are shown. Mean ± SEM; n = 3. *, P = 0.01 to 0.05; **, P = 0.001 to 0.01; ***, P = 0.0001 to 0.001, n.s. nonsignificant using two-sample t tests.
PARGi decreases colony formation in PDAC cells in combination with oxaliplatin. MIA PaCa-2 (A) and PANC-1 (B) cell lines were treated with increasing doses of PDDX-01 (μmol/L) alone and in combination with oxaliplatin (oxali) or olaparib (olap), and resulting colonies were fixed and stained with crystal violet solution. Representative images with graphs of relative colonies are shown. Mean ± SEM; n = 3. *, P = 0.01 to 0.05; **, P = 0.001 to 0.01; ***, P = 0.0001 to 0.001, n.s. nonsignificant using two-sample t tests.
PARGi affects PARylation dynamics of PDAC cells and regulates PARG activity
To further assess the mechanism of PARGi-induced cytotoxicity and whether PARGi causes an increase in PAR accumulation in PDAC cells, we performed chromatin tethering assays (54). We analyzed PARP and PAR protein expression in chromatin-bound fractions after treatment with PARGi alone or in combination with oxaliplatin or olaparib in PDAC cell lines. GAPDH was used as a negative control and histone H3 expression was used as a positive control for chromatin fractions. PDDX-01 increased chromatin bound PAR expression in both MIA PaCa-2 and PANC-1 cells treated alone as shown in Fig. 5A and B. We also observed a synergistic effect of PARGi with oxaliplatin as seen in Figs. 3 and 4, which could not be attributed to synergistic increase in accumulated PAR with the combination (43). Contrary to this affect, and as previously shown (55), olaparib not only decreased relative PARylation as a consequence of inhibiting PARP1 activity alone, but also prevented PARGi-induced increase in PAR (Fig. 5A and B), implying that inhibition of PARGi does not benefit PARPi therapy, in agreement with previous results (55).
PARGi induces PAR accumulation. MIA PaCa-2 (A) and PANC-1 (B) cells were treated with either PDDX-01, olaparib or oxaliplatin, alone or in combination for 24 hours, and chromatin and whole-cell fractions were isolated. Representative Western blot images are shown from n = 3. GAPDH and Histone H3 served as controls for whole-cell fraction and chromatin fractions, respectively. C, PARGi in combination with oxaliplatin causes cleavage of caspase-3. PANC-1 cells were treated with PDDX-01, oxaliplatin, or combination, and apoptosis was assessed by Western blot analysis for cleaved caspase-3 at 24-, 48-, and 72-hour time points. Representative image from n = 3 is shown. D, PARGi induces DNA damage in combination with oxaliplatin. γH2AX foci increases with the combination of PDDX-01 and oxaliplatin at 24 hours in PANC-1 cells. Representative graphs of cells with >10 foci are shown. Mean ± SD; n = 3. *, P < 0.05; ***, P < 0.001.
PARGi induces PAR accumulation. MIA PaCa-2 (A) and PANC-1 (B) cells were treated with either PDDX-01, olaparib or oxaliplatin, alone or in combination for 24 hours, and chromatin and whole-cell fractions were isolated. Representative Western blot images are shown from n = 3. GAPDH and Histone H3 served as controls for whole-cell fraction and chromatin fractions, respectively. C, PARGi in combination with oxaliplatin causes cleavage of caspase-3. PANC-1 cells were treated with PDDX-01, oxaliplatin, or combination, and apoptosis was assessed by Western blot analysis for cleaved caspase-3 at 24-, 48-, and 72-hour time points. Representative image from n = 3 is shown. D, PARGi induces DNA damage in combination with oxaliplatin. γH2AX foci increases with the combination of PDDX-01 and oxaliplatin at 24 hours in PANC-1 cells. Representative graphs of cells with >10 foci are shown. Mean ± SD; n = 3. *, P < 0.05; ***, P < 0.001.
To evaluate whether PARG inhibitors affect relative PARylation, whole-cell PARylation was assessed by ELISA assays (28). All PARGi compounds significantly lead to persistence of whole-cell PARylation, as analyzed by ELISA assays (Supplementary Fig. S5A) in both HR-P and HR-D PDAC cells. PARPi (olaparib) was used as the negative control, which has previously been shown to effectively prevent PAR polymer formation (28).
PARGi affects apoptosis and a marker of DNA damage response
In an effort to find the mechanism of synergy between PARGi and oxaliplatin in PDAC cells, we assessed apoptosis by cleaved caspase-3 expression in both MIA PaCa-2 and PANC-1 cells. Increase in cleaved caspase-3 with oxaliplatin was seen as early as 24 hours for both MIA PaCa-2 cells and PANC-1 cells (Fig. 5C; Supplementary Fig. S6A). This suggests that activation of caspases is a possible mechanism of synergistic cell death between PARGi and oxaliplatin.
In addition, we analyzed γH2AX expression as a marker of DNA damage and found a significant increase in γH2AX foci in combination treatments across both the cell lines, indicating an increase in DNA damage (Fig. 5D; Supplementary Fig. S6B). Taken together, these results demonstrate that targeting PARG with DNA damage agents induces DNA damage and apoptosis in PDAC cells.
Discussion
Current standard-of-care chemotherapeutic regimens show only modest increases in overall survival of patients with PDAC (56, 57). Thus, there remains a desperate need to find targets and therapies, which eventually could be personalized for PDAC (2–7, 10, 58). Recently, a successful phase II study of olaparib (PARPi) showed significant clinical activity in patients with BRCA1/2 (HR-D)–mutant pancreatic cancer previously treated with gemcitabine (19). However, many studies have already started to emerge with PARPi resistance (24–26). Previously, we published that PARPi treatment causes an increase in PARG protein through a novel posttranscriptional mechanism regulated by stress-responsive RNA binding protein, HuR in PDAC (28). On the basis of these data and the fact that PARG is part of the DNA repair process, we explored PARG as a therapeutic target in PDAC by first utilizing shRNA genomic approaches to silence PARG expression and activity, and then validating these findings through the use of established small-molecule PARG inhibitors (38, 42).
The data presented herein also demonstrate that small-molecule inhibitors of PARG decrease cell survival, proliferation, and induce DNA damage in PDAC cells when used as monotherapy or in combination with DNA damage agents in vitro (Figs. 2–5). IC50 values of PARGi values indicate that in various isogenic and nonisogenic PDAC cell models PARG inhibitors are more effective against HR-D cells when compared with HR-P cells, highlighting previous reports that PARG targeting may have beneficial synthetic lethal effects on cells with HR deficiencies (35–38, 59). In sum, these data suggest PARGi should be explored as a novel synthetic lethal approach in PDAC tumors that share molecular features with BRCA-mutant PDAC subtypes.
Although our study shows that PARGi is more effective in HR-D PDAC cells, PARGi caused increased PARylation and a significant decrease in PARG activity in both HR-D and HR-P PDAC cells, without directly affecting protein levels of either PARG or PARP1. The increase in PARylation after PARGi emphasizes the specificity of PARGi and the importance of PARG as a primary PAR catabolizing enzyme (60). This also suggests that PARG may be a more favorable target over other DNA repair proteins like PARP1, which has multiple family members (33, 41). Although these inhibitors currently have limited bioavailability (42), our results support development of these lead compounds for more favorable bioavailable small-molecule PARGis for effective targeting of PARG in vivo. In an ongoing effort, we have started investigating a 3DNA-nanocarrier delivery of modified siPARG to target PARG with a tumor-specific targeting moiety in vivo (Supplementary Fig. S7).
Combination treatments of PARGi with either oxaliplatin or 5-FU (components of FOLFIRINOX) in HR-P PDAC cell lines provide evidence that targeting PARG could synergize with current standard-of-care treatments for PDAC (Fig. 3). Our results highlight that PARGi is antagonistic with olaparib (PARPi), as analyzed through bliss index of combination treatments (Fig. 3). These results are in agreement with a recent study published by Gogola and colleagues that show loss of PARG as a resistance mechanism to PARPi therapy in murine mammary tumor cells and that PARP1 signaling is rescued by PARGi-mediated PAR accumulation (55). To understand the increased sensitivity of oxaliplatin in presence of PARGi, we looked at PAR levels in chromatin fractions. Accumulation of PAR polymers and further loss of NAD+ pools are reported to cause cytotoxicity (61–63). From our results, we did not find a synergistic increase in accumulation of PAR with the combination of oxaliplatin and PARGi (Fig. 5A and B). However, we found an increase in DNA damage and elevated cleaved caspase-3 expression with the combination, both of which could implicate mechanisms behind synergy seen with oxaliplatin and PARGi (Fig. 5C and D; Supplementary Fig. S6A and S6B). Future studies will determine the precise mechanism of increased sensitivity of 5-FU or oxaliplatin with PARGi in an in vivo setting. Overall, these findings favor targeting PARG in combination with DNA-damaging agents (e.g., oxaliplatin) and warrants further in vivo dose optimization and sequencing (i.e., timing) studies.
Taken together, we strongly believe this work provides the first evidence that acute targeting of PARG should be pursued as a treatment strategy for PDAC with both HR-proficient and -deficient genetic backgrounds. More sophisticated, ongoing in vivo studies will evaluate next-generation PARG-targeting strategies (i.e., other small molecules and siPARG nanotherapy) in combination with other DNA-damaging agents for the treatment of all pancreatic cancers.
Disclosure of Potential Conflicts of Interest
G.A. McCarthy reports receiving a commercial research grant from Genisphere, LLC. T. Golan reports receiving a commercial research grant from MSD and AstraZeneca, has received speakers bureau honoraria from Abbvie, and is a consultant/advisory board member for Abbvie and Teva. D.I. James is a consultant for Cancer Research UK. J.R. Brody reports receiving a commercial research grant from Genisphere, LLC. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: A. Jain, L.C. Agostini, S.N. Chand, C.W. Schultz, J.R. Brody
Development of methodology: A. Jain, L.C. Agostini, G.A. McCarthy, S.N. Chand, J. Cozzitorto, D. Ogilvie, D.I. James, A.M. Jordan
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Jain, G.A. McCarthy, S.N. Chand, A. Ramirez, A. Nevler, J. Cozzitorto, C.W. Schultz, C.Y. Lowder, M. Raitses-Gurevich, C. Stossel, Y.G. Gorman, D. Atias, C.J. Yeo, K.P. Olive, T. Golan, D.I. James
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Jain, L.C. Agostini, G.A. McCarthy, S.N. Chand, C.W. Schultz, M.J. Pishvaian, D.I. James, J.R. Brody
Writing, review, and/or revision of the manuscript: A. Jain, L.C. Agostini, G.A. McCarthy, A. Nevler, C.Y. Lowder, I.D. Waddell, C.J. Yeo, J.M. Winter, T. Golan, M.J. Pishvaian, D.I. James, A.M. Jordan, J.R. Brody
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Jain, S.N. Chand, C.J. Yeo, D.I. James, J.R. Brody
Study supervision: A. Jain, S.N. Chand, I.D. Waddell, C.J. Yeo, J.M. Winter, J.R. Brody
Others (synthesis of compound used in the study): K.M. Smith
Others (provision of small molecule tools): A.M. Jordan
Acknowledgments
We would like to thank Genisphere LLC (Dr. Robert Getts, Hatfield, PA) for providing us with the modified siPARG. Mark Levine's contributions to this work were made in memory of Ethel Levine. This work was supported by NIH-NCI R01 CA212600 (to J.R. Brody) and R37CA227865 (to J.M. Winter and J.R. Brody), and was also supported by the NCI of the NIH under Award Number P30CA056036 SKCC Core Grant (Thomas Jefferson University). Additional work is supported by a 2015 Pancreatic Cancer Action Network American Association for Cancer Research Acceleration Network Grant (15-90-25-BROD to J.R. Brody and M.J. Pishvaian). This work is also generously supported by a Mary Halinski Fellowship (to A. Nevler) and the W. Kim Foster Pancreatic Cancer Research Endowment. K.M. Smith, I.D. Waddell, D. Ogilvie, D.I. James, and A.M. Jordan are supported by Cancer Research UK (grants C480/A11411 and C5759/A17098).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.