Targeting DNA Repair with Combined Inhibition of NHEJ and MMEJ Induces Synthetic Lethality in TP53-Mutant Cancers

Genome integrity is critical for cellular survival and is maintained by DNA repair pathways collectively known as the DNA damage response (DDR; ref. 1). Loss of genomic integrity results in permanent changes to the sequence of DNA and is the source of many human diseases, notably cancer (2–7). Double-stranded breaks (DSB) are the most deleterious form of DNA damage and can be repaired through three main DNA repair pathways: homologous recombination (HR), nonhomologous end-joining (NHEJ), and microhomology-mediated end-joining (MMEJ).

NHEJ occurs throughout the cell cycle and is the predominant DSB repair pathway. NHEJ is an error-prone pathway, unlike HR repair, which is homology-guided. The first step of NHEJ is recognition of the DSB by the Ku70-Ku80 heterodimer (8, 9). Depending on the configuration of the DNA ends (blunt ends, 5′ overhangs, and 3′ overhangs), the Ku70-Ku80 heterodimer acts as a scaffold to which other NHEJ proteins are recruited to promote rejoining of the DNA ends (8–12). Next, DNA-dependent protein kinase (DNA-PK) catalytic subunit (DNA-PKcs) binds with high affinity to Ku70-Ku80 heterodimer-DNA ends (9). If the DNA break ends are compatible, either blunt or with complementary overhangs, then repair will likely be error-free with direct ligation by XRCC4-DNA ligase IV. If, however, the DNA break ends are not compatible, requiring additional processing, either resection by nucleases, or addition of nucleotides by polymerases, then repair will be error prone (13).

Because DSBs are the most deleterious form of DNA damage, targeting their repair is a potential therapeutic strategy for the treatment of cancers (1, 14–16). The majority of DSBs are repaired by NHEJ, where DNA-PK is the key driver. Accordingly, inhibitors of DNA-PK activity may provide a therapeutic strategy for sensitizing cancer cells to chemotherapy or ionizing radiation (IR; refs. 17–21). Peposertib is a novel, selective small molecule inhibitor of the serine/threonine kinase activity of DNA-PK (22). Peposertib is a potent and selective pharmacologic inhibitor of DNA-PK, which competes with ATP for binding to DNA-PK with an IC₅₀ of 0.6 nmol/L (22).

In this report, we performed a genome-wide clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9 loss of function genetic screen to identify enhancers and suppressors of cell killing by the DNA-PK inhibitor, peposertib. Single-guide RNAs (sgRNA) directed against all reported members of the MMEJ pathway, including POLQ (encoding Polymerase Theta, POLθ), PARP1, HMCES, FEN1, XRCC1, and LIG3 were synthetically lethal with DNA-PK inhibition (23, 24). We recently identified the ATPase inhibitor, novobiocin (NVB), as a first-in-class inhibitor of POLθ and the MMEJ pathway (25). As expected from the CRISPR-KO screen results, the combination of peposertib plus NVB exhibited synergistic anticancer activity. This drug combination resulted in excessive DSB end-resection and ultimately apoptosis-mediated cell death, providing a molecular mechanism of the observed synthetic lethality. Interestingly, TP53-mutant tumor cells are resistant to peposertib but have elevated sensitivity to the drug combination, resulting from their elevated POLθ expression level. Accordingly, the combination of peposertib plus NVB resulted in synthetic lethality in TP53-deficient tumor cell lines, organoid cultures, and patient-derived xenograft (PDX) models. Thus, a combination of a targeted DNA-PK inhibitor plus a targeted POLθ/MMEJ inhibitor may provide a rational treatment strategy for TP53-mutant solid tumors.

A549 and NCI-H460 stably expressing the Cas9 endonuclease were obtained from the Genetic Perturbation Platform (GPP) at the Broad Institute and were grown at 37˚C in a humidified 5% CO₂ incubator in DMEM (Gibco) supplemented with 10% FBS (Sigma). U2OS and RPE cells were purchased from ATCC and grown in RPMI (Gibco) supplemented with 10% FBS or in F12 (Gibco) supplemented with 10% FBS. MBA-MD-436 isogenic pairs were a gift from Dr. Shapiro and grown in RPMI (Gibco) supplemented with 10% FBS. CAPAN-1 isogenic pairs were a gift from Dr. Christopher Lord and grown in DMEM (Gibco) supplemented with 20% FBS. Mycoplasma testing was performed every 2 weeks using MycoAlert (Lonza). All experiments were performed in low passage number (less than 20) cells.

The primary CRISPR screen was performed in pXPR_311 integrated Cas9 expressing NSCLC cell line A549 and NCI-H460 per the GPP CRISPR screening protocol from the MIT Broad Institute (26, 27). For adequate representation of each gRNA, for example, an average of 300 cells/sgRNA, cells were infected at MOI 0.3 in media containing 8 μg/mL of polybrene. Briefly, cell lines were cultured in RPMI1640 media supplemented with 10% FBS, l-glutamine, and penicillin–streptomycin. Each cell line was infected in 12-well plates for 2 hours under 2,000 rpm at 30°C with the Brunello whole-genome sgRNA pooled lentiviral library consisting of 76,441 sgRNAs targeting 19,114 different genes. Media was replaced immediately following lentiviral infection. Infected cells were then plated 15 cm dishes with 2 μg/mL of puromycin to select 4 days. After selection, 5e6 cells were plated for each biological replicate, and additional 5e6 cells per replicate were collected and frozen for subsequent genomic DNA extraction as time 0. Both cell lines were treated with 1 μmol/L of peposertib (C-1358; Chemgood) or equal volume of DMSO for 17 days. Cells were replated with a density of 5e6 every 3 days, along with fresh DMSO or drug. At the end of screen, the cells were harvested and frozen for gDNA extraction. gDNA was isolated using an QIAGEN QIAamp DNA Blood Maxi Kit, according to manufacturer's protocol. gDNA samples were subjected to the GPP at the Broad institution for barcoding, PCR purification, and sequencing on an illumine HiSeq 2000. Similarly, the secondary CRISPR screen was also performed in these cell lines (A549 and H46) but with a subset of top hit genes from the primary screen.

Sequenced reads per million were log₂-transformed by first adding one to all values, which was necessary to take the log of sgRNAs with zero reads. For analysis, these normalized data were used to determine the log₂-fold change of each sgRNA for the drug treatment group relative to DMSO control group (28). The STARS gene-ranking algorithm was used to determine the top hits (https://portals.broadinstitute.org/gpp/public/software/stars); at least two sgRNAs targeting a gene had to rank within the top 10% of sgRNAs by log₂-fold change in one direction or another (29). The hit genes were scored using the STARS score and the FDR. In addition, the normalized data were also analyzed with the hypergeometric distribution method (https://github.com/mhegde/volcano_plots). The volcano plots of these results were created using R (https://www.r-project.org).

All antibodies and chemicals are shown in Supplementary Table S1.

Cells were lysed in NETN buffer (10 mmol/L TrisCl, pH 7.4, 1 mmol/L EDTA, 0.05% Nonidet P-40, and protease inhibitors) containing 150 mmol/L NaCl (30). Cell lysates were boiled in 2× Laemmli buffer and resolved on Nu-Page Bis-Tris (Invitrogen) polyacrylamide gels using MES buffer, and subsequently transferred to nitrocellulose membranes. Membranes were blocked with 5% milk in TBST and probed with primary and appropriate enhanced chemiluminescence secondary antibodies, then detected with chemiluminescence (Western Lightning, Perkin Elmer) or appropriate LiCor secondary antibodies, and then detected on the LiCor Odyssey.

To generate knockout A549 and NCI-H460 cell lines, the indicated sgRNAs, described in Supplementary Table S1 were cloned into the pLentiCRISPR V2 viral vector.

For siRNA transfection, cells were seeded at about 30% confluence into 6-well plates the day prior to transfection. Specific siRNA duplexes were transfected using Lipofectamine RNAiMAX Reagent Agent (Life Technologies, #13778150) according to the manufacturer's instructions. Sequences of the siRNAs employed in this study are shown in Supplementary Table S1.

Cells of interest were seeded at a concentration of 1,000 to 2,500 cells per well into 6-well plates. After 24 hours, cells were continuously treated with indicated compounds for 10 to 14 days. Colonies were fixed (fixation solution: 5 parts methanol + 1 part acetic acid) at room temperature for 20 minutes and then stained with a solution of 0.5% crystal violet in methanol for 1 to 2 hours. Plates were imaged using the Amersham Imager 600 (GE) and colonies were quantified with ImageJ software (NIH). All clonogenic assays were performed in technical triplicates, with a “n” of at least three.

Cells of interest were seeded at a concentration of 3,000 cells per well into 96-well plates and treated the next day with the indicated compounds. Following 4 days of indicated treatment, the CellTiter-Glo, ATP-based, cell viability assay was performed (Promega). Survival curves were calculated by best-fit analysis of the log of the drug concentration to fold change of treated cells over vehicle-treated cells. Synergy between peposertib and NVB was calculated using Combenefit software as described previously (31). All survival assays were performed in technical triplicates, with a “n” of at least three.

Four different DNA damage reporter assays were employed in this study. NHEJ efficiency was measured using the EJ5-GFP reporter assay. HR efficiency was measured using the DR-GFP assay. MMEJ was measured using either the EJ2 or the P2A MMEJ reporter assays. SSA efficiency was measured using the SSA-GFP assay. U2OS cells with containing the appropriate DNA damage reporter assay were transfected with the indicated siRNA as described above and/or treated with peposertib, NVB, or a combination of both. 24 hours after transfection, cells were infected with an I-Sce-I adenovirus. Forty-eight hours later, reporter assay positivity was determined by flow cytometric quantification. Technical triplicates were performed, and experiments repeated at least three times.

U2OS cells were grown on coverslips in the presence of 10 μmol/L BrdU (Sigma) for 16 hours in the presence of DMSO or the appropriate drug compound. Cells were fixed using 4% paraformaldehyde. Fixed cells were washed twice with PBS and permeabilized with 0.5% Triton X-100 in PBS for 5 minutes at 4°C. Cells were washed with PBST and incubated with the appropriate primary antibody. Following primary incubation, coverslips were washed three times with PBST, then incubated with the appropriate secondary antibody (Alexa Fluor, Life Technologies) for 1 hour. Coverslips were then mounted on glass slides using mounting media containing DAPI (Vectashield) and foci imaged using a florescence microscope and quantified using ImageJ (NIH).

Single molecule analysis of resection tracks was performed as described previously (32). Briefly, cells were treated with bromodeoxyuridine (BrdU; Sigma) for 24 hours. Subsequently, cells were treated with the appropriate drug compound for 24 hours and then trypsinized, counted, and embedded in low melting point agarose plugs for treatment with proteinase K overnight. Then, agarose plugs were washed and digested with Agarase. Agarase-treated samples were then poured into FiberComb wells and combed onto silanized coverslips. Coverslips were probed with rat anti-BrdU antibody (Abcam) and visualized by fluorescence microscopy. Pictures were taken of at least 100 fibers per condition. DNA fibers were measured with Adobe Photoshop CC 2019. Each experiment was performed in triplicate.

U2OS cells stably expressing ER-AsiSI were used to quantify DNA end-resection. Briefly, cells were treated with the appropriate drug compound for 24 hours. During the final 4 hours of treatment, cells were treated either with or without 300 nmol/L 4-OHT for 4 hours to induce DSBs at specific AsiSI sites by allowing transportation of the constitutively expressed AsiSI to the nucleus. Genomic DNA was extracted with the Qiagen DNeasy Blood & Tissue Kit. Genomic DNA was either mock digested or digested overnight with BsrGI-HF (New England Biolabs). Two microliters of digested or mock-digested samples were used as templates in 20 μL of qPCR reaction containing 10 μL of 2× SYBR Green (Thermo Fisher Scientific), and 1 μmol/L of each primer using a QuantStudio 7 Real-Time PCR System (Thermo Fisher Scientific). The percentage of ssDNA (ssDNA%) generated by resection at selected sites was then determined by calculating a △C_t for each sample by subtracting the C_t value of the mock-digested sample from the C_t value of the digested sample.

Alkaline comet assays to detect ssDNA and dsDNA were performed according to the manufacturer's instructions (Trevigen). Cells were trypsinized and suspended in cold PBS then mixed with low melting agarose (Trevigen) at a ratio of 1:10 and plated onto Cometslides (Trevigen). Electrophoresis was performed for 30 minutes at 25 V in an alkaline buffer. Photographs of comets were captured by fluorescence microscopy and analyzed using ImageJ and OpenComet.

Total RNA was isolated using the Qiagen RNAeasy Kit. To quantify gene expression levels, equal amounts of cDNA were synthesized using the SuperScript IV Reverse Transcriptase (Life Technologies) and mixed with PrimeTime Gene Expression Master Mix (IDTDNA) and the appropriate PrimeTime qPCR Primer Assays. Actin was amplified as an internal control. The threshold crossing value was noted for each transcript and normalized to the internal control. The relative quantitation of each mRNA or miRNA was performed using the comparative C_t method.

Cells were seeded into 96-well plates at a cell density of 2,000 cells/well and incubated overnight. The following day, cells were treated with 1 μmol/L peposertib, 100 μmol/L NVB, or the combination, 1 hour prior to ionizing radiation exposure (2 Gy). After four days of incubation, Caspase-Glo 3/7 reagent (Promega) was added. Cells were incubated in the dark for 1 hour at room temperature on a tabletop shaker. Luminescence was measured using a Molecular Devices SpectraMax M5 spectrophotometer and normalized to the DMSO-treated control.

Tumor cells were extracted from PDXs and short-term patient-derived organoid cultures were generated following the protocol described previously (33). Briefly, processed cells were diluted in specialized media for ovarian cancer organoid generation and added into each well of a 96-well plate. Drugs were dispensed in randomized manner according to a prewritten software program using a Tecan D300e digital dispenser. The effect of NVB and peposertib was studied as single agents and combinations by measuring cell death after 5 days of drug treatment using Cell Titer-Glo cell viability assay following the manufacturer's instructions. Luminescence was measured using a Clariostar microplate reader (BMG Labtech). A very high concentration of toxin [puromycin (2 mg/mL)/cycloheximide (50 μmol/L)] was used as a positive control, which killed more than 95% of PXO. For monotherapy, we tested an eight-point drug response and for combination we prepared a concentration grid and calculated synergy.

All animal experiments were approved by the Institutional Animal Care and Use Committee at the Dana-Farber Cancer Institute (protocol 08-036). For the TP53 mutant ovarian tumors (the DF59 PDX model) studies, following written informed consent, tumor ascites was collected from with ovarian cancer at the Brigham and Women's Hospital or at DFCI under Institutional Review Board–approved protocols in accordance with the Declaration of Helsinki and Belmont Report. PDX models were established, luciferized, and propagated as previously described (34). A total of 10 × 10⁶ luciferized PDX cells derived from mouse ascites were injected intraperitoneally into 8-week-old female NRG mice. Tumor burden was measured by weekly BLI imaging using a Xenogen IVIS 200 system. Prior to initiation of treatment, mice were grouped according to BLI signals and subsequently randomized, so that the average tumor BLI value in each group was the same. For treatment, NVB was diluted in saline and administered via intraperitoneally injection at 100 mg per kg twice daily for 16 days and then 75 mg per kg twice daily until day 28. Peposertib (EMD Serono) was formulated in a vehicle of 0.5% Methocel/0.25% Tween20 in 300 mmol/L citrate buffer, pH 2.5, and administered orally at 100 mg per kg daily for 4 weeks.

For the A549 xenograft studies, 2 × 10⁶ A549 cells in a 1:1 mixture with Matrigel (Corning) were subcutaneously injected into the flanks of athymic nude mice. Xenografted mice were randomly divided into the appropriate treatment group. For treatment, NVB was diluted in saline and administered via IP inject at 75 mg per kg twice daily for 29 days. Tumor volume was measured every 3 to 4 days using digital calipers.

Organoids were treated (DMSO, 5 μmol/L peposertib, 100 μmol/L NVB, or combination) for 2 days and then resuspended into HistoGel and fixed overnight using 10% formalin solution. Histogel blocks were paraffin embedded in the specialized histopathology core at the Brigham and Women's Hospital (BWH). Formalin-fixed, paraffin embedded (FFPE) blocks were sliced to thin sections on positively charged slides, and then stained with hematoxylin and eosin or stained with the appropriate antibodies. FFPE sections of irradiated and unirradiated tissue blocks were used as controls. Stained slides were scanned at the BWH specialized histopathology core on an Olympus BX41 microscope equipped with a digital camera with ×40 magnification. Biomarker analysis and scoring for positively stained foci was performed using image analysis software (Leica, Aperio ImageScope).

Quantitative data were analyzed and graphed using GraphPad Prism 9 software. All data are represented as mean ± SEM calculated unless indicated otherwise. Significance was tested using the Student t test unless indicated otherwise. Assessment of synergy was calculated on the basis of the Bliss model of synergy.

Data were generated by the authors and available on request.

To identify synthetic lethal interactors with the highly selective DNA-PK inhibitor (Peposertib), we initially evaluated two TP53 wild-type non-small cell lung cancer cell (NSCLC) lines, A549 and H460, which were engineered to stably express the Cas9 endonuclease (26, 27). Using a clonogenic assay, we analyzed the cytotoxicity resulting from their treatment with peposertib, with or without IR (Supplementary Fig. S1A). Consistent with previous reports, 1 μmol/L peposertib alone had no discernable effect on the clonogenic survival of the cells. When combined with IR (2 Gy), however, there was a significant loss of clonogenic survival (Supplementary Fig. S1A; ref. 35). Using DNA repair reporter assays, peposertib inhibited NHEJ (EJ5) activity but did not inhibit HR (DR-GFP) or MMEJ (EJ2) activity (Supplementary Fig. S1B; ref. 36). In both cell lines, DNA-PK was rapidly auto-phosphorylated on Ser2056 following IR exposure, and autophosphorylation was inhibited by peposertib (Fig. 1A). Drug exposure did not alter the cell-cycle distribution of the tumor cells (Supplementary Fig. S1C).

We next devised a genome-wide CRISPR screen to identify genes, which, when disrupted, would increase sensitivity to the DNA-PK inhibitor, peposertib (Fig. 1B). The screen was conducted in A549 and H460, in the presence or absence of peposertib. To identify sgRNAs that target genes involved in sensitivity to peposertib, we chose a moderate concentration (1 μmol/L peposertib) of drug that does not kill the cells as a monotherapy but slightly increases the doubling time (Fig. 1C; Supplementary Fig. S1D; ref. 29). Cells were collected pre- and post-drug selection, and genomic DNA was extracted. The sgRNA sequences were PCR amplified and Illumina sequenced. Top hits among the gene knockouts that conferred drug sensitivity were identified.

The screen identified several genes from both cell lines that modulated cell killing by peposertib (Fig. 1D; Supplementary Fig. S1E). As expected, sgRNAs against the peposertib target itself (PRKDC) conferred peposertib resistance. Genes that were hits in both cell lines were of special interest (Fig. 1E). Many hits corresponded to specific signaling pathways (Supplementary Fig. S1F), and the top hits were in DNA repair pathways. Analysis of the two screens identified POLQ as a significant synthetic lethal interactor with peposertib (Fig. 1E). POLQ encodes polymerase theta (POLθ), a unique DNA polymerase that promotes MMEJ and is a known synthetic lethal interactor with HR repair deficiency (37, 38). Rescreening the top hits in a secondary screen validated POLQ as a peposertib synthetic lethal interactor (Supplementary Figs. S1G and S1H).

The MMEJ pathway is a stepwise DSB repair pathway (24). The pathway is initiated by 5′ to 3′ end-resection, involving near the DSB, exposing short regions of complementary sequences termed microhomologies. Additional genes, including POLQ, PARP1, HMCES, FEN1, XRCC1, and LIG3, are involved in the downstream events in MMEJ (24). Interestingly, our screens identified all of these known members of the MMEJ pathway as potential synthetic lethal interactors with peposertib-mediated DNA-PK inhibition (Fig. 1E and F).

CRISPR-mediated knockout of individual MMEJ genes in A549 and H460 cells next validated the screening results (Fig. 2A–E; Supplementary Figs. S2A and S2B). Immunoblotting with the corresponding antibody confirmed protein knockdown by CRISPR knockout of POLQ, HMCES, FEN1, XRCC1, LIG3, and PARP1 (Supplementary Figs. S2C–S2H). As predicted, knockout of these individual MMEJ genes increased the sensitivity to peposertib-mediated DNA-PK inhibition (Fig. 2A–E; Supplementary Fig. S2B). The MMEJ pathway, also termed alternative end-joining, is a backup mechanism for repairing DSBs when both HR and NHEJ were unavailable (39, 40). Accordingly, loss of DNA-PK mediated by peposertib, results in greater cellular dependence on MMEJ for repair of DSBs, similar to the observed hyper-dependence of HR-deficient cells on MMEJ (37, 38).

We recently identified NVB as a first-in-class small molecule inhibitor of POLθ’s ATPase activity, which selectively kills HR-deficient cancer cells and overcomes PARP inhibitor resistance (25). NVB specifically inhibits POLθ with little effect on another DNA-repair related ATPase, SMARCA1L. NVB inhibits POLθ with an IC₅₀ of 24 μmol/L (25). Accordingly, NVB-mediated POLθ inhibition might be synthetically lethal with peposertib-mediated DNA-PK inhibition as well, based on our CRISPR screen results. To test this hypothesis, A549 and H460 cell lines were exposed to the combination of peposertib and NVB. As expected, addition of NVB to increasing concentrations of peposertib enhanced the cytotoxicity in both A549 and H460 cell lines (Fig. 3A and B). To eliminate potential off-target effects of peposertib, we performed clonogenic survival assays in A549 and H460, which had been engineered with a CRISPR-knockout of PRKDC (encoding DNA-PK; Supplementary Fig. S3A). Knockout of DNA-PK rendered both cell lines more sensitive to low concentrations of NVB (Fig. 3C and D). Furthermore, to reduce the possibility of off-target effects of NVB, we generated A549 with a CRISPR-knockout of POLQ. As predicted, these A549 POLQ knockout cells were more resistant to NVB treatment than wild-type cells (Supplementary Fig. S3B), further confirming POLθ as the specific target of NVB. We also assessed the synergy of a combination of peposertib and NVB treatment in A549 and H460 cells lines, using the Bliss synergy model (41). The combination of peposertib plus NVB resulted in synergistic cytotoxicity in both cell lines (Fig. 3E and F; Supplementary Figs. S3C and S3D). To extend these findings, we also employed ART558, a recently identified allosteric inhibitor of the POLθ polymerase domain, in combination with peposertib or in A549 and H460 cells (42). Consistently, the combination of ART558 with peposertib or treatment of A549 or H460 with a CRISPR-knockout of POLQ resulted increased cytotoxicity (Supplementary Figs. S3E–S3G).

We reasoned that the peposertib-mediated inhibition of DNA-PK would render the cells more dependent on MMEJ-mediated DNA repair. To test this hypothesis, we first assessed the effect of peposertib treatment on a cell-based MMEJ reporter (43). As predicted, we observed an increase in MMEJ reporter activity upon treatment with peposertib (Fig. 3G; ref. 43). We next assessed the effect of peposertib treatment on POLQ expression, a known predictive and quantitative biomarker of NVB responsiveness (25). Interestingly, peposertib treatment caused a dose-dependent increase in both POLθ protein expression and POLQ mRNA expression in A549 cells (Fig. 3H and I). Similar results were observed in A549 DNA-PK knockout cells (Fig. 3J; Supplementary Fig. S3H). Thus, upon peposertib-mediated inhibition of DNA-PK, cells compensate by increasing POLθ expression and POLθ-mediated MMEJ, rendering them hypersensitive to NVB.

Among the three main DSB repair pathways, NHEJ preferentially repairs unresected DSBs, whereas MMEJ and HR require nucleolytic DNA end-resection of 5′ ends generating a 3′ ssDNA overhangs (24). Because DNA-PK is critical for the NHEJ pathway (Supplementary Fig. S1C), we reasoned that inhibition of DNA-PK with peposertib might increase the cellular dependence on DNA end-resection and activate the other two DSB repair pathways (34). To test this hypothesis, we evaluated the role of DNA-PK inhibition on DNA end-resection using multiple complementary assays. Phosphorylation of the ssDNA binding protein RPA on S33 (p-RPA) is a biomarker of single strand DNA generation at sites of resected DSBs (44). As predicted, treatment of U2OS cells with peposertib induced more fluorescent p-RPA than DMSO vehicle treatment (Fig. 4A and B). To monitor ssDNA formation more directly, we labeled genomic DNA with 5-bromo-2′-deoxyuridine (BrdU). BrdU can be detected by immunofluorescence with an anti-BrdU antibody under native conditions but only if the DNA is single-stranded, thereby allowing quantification of end-resection in cells (45). Again, treatment of U2OS cells with peposertib resulted in more DSB end resection, as visualized by enhanced BrdU fluorescence, consistent with the enhancement of p-RPA foci (Supplementary Figs. S4A and S4B). To further quantify the extent of DNA end-resection after peposertib treatment, we employed a qPCR-based method in which the AsiSI restriction enzyme is fused with a hormone-binding domain in U2OS cells. Upon treatment with tamoxifen, AsiSI translocates to the nucleus and generates DSBs at a sequence-specific site (Supplementary Fig. S4C; ref. 46). Treatment with peposertib increased the level of ssDNA measured at two sites, 355 bp and 1618 bp from the induced DSB (Fig. 4C; Supplementary Fig. S4D). Taken together, peposertib-mediated inhibition of DNA-PK prevents NHEJ, resulting in an increase in DSB DNA end-resection.

Inhibition of POLθ is also known to increase DNA end-resection (25, 42). As predicted, exposure of U2OS cells to NVB caused a dose-dependent increase in p-RPA foci and BrdU foci, markers of ssDNA (Supplementary Figs. S4E–S4H; refs. 44, 45). A quantitative increase in ssDNA was also observed after NVB treatment, as assessed by the ER-AsiSI DSB assay (Supplementary Fig. S4I). Because both DNA-PK inhibition and POLθ inhibition increase DSB end-resection, we hypothesized that the mechanism of synthetic lethality in the cells with dual inhibition of NHEJ and MMEJ might be the generation of toxic levels of DNA end-resected intermediates. Indeed, significantly more γH2AX foci were generated from the combination of peposertib and NVB than from either treatment alone (Fig. 4D and E). This finding is consistent with our hypothesis that combined treatment with peposertib and NVB generates more DNA damage. Next, we assessed the level of resected DNA by measuring the level of p-RPA after treatment with DMSO, peposertib, NVB, or the combination of peposertib and NVB. A significant increase in p-RPA foci was observed following the combination of peposertib and NVB, compared with either treatment alone (Fig. 4F and G). To determine the extent of DNA end-resection, we again employed the AsiSI-induced DSB assay and quantified the production of ssDNA. The combination of peposertib and NVB significantly increased the production of ssDNA, compared with either treatment alone (Fig. 4H; Supplementary Fig. S4J). Finally, we assayed DNA end-resection using the single molecule analysis of resection tracts (SMART) method (32). Treatment of A549 cells with the combination of peposertib and NVB significantly increased the amount of ssDNA generated by DNA end-resection (Fig. 4I). A similar result was also obtained in A549 DNA-PK KO cells treated with NVB (Supplementary Fig. S4K).

In a comet assay, the combination of peposertib plus NVB caused increased tail moments, reflecting an increase in DNA damage (Fig. 4J). The induction of apoptosis was also measured by the production of cleaved caspases. The observed synergy between peposertib and NVB correlated with this induction of apoptosis (Fig. 4K; Supplementary Fig. S4L). Consistent with the role of drug-enhanced extensive end-resection as the mechanism of synthetic lethality, depletion of the DNA end-resection nucleases, EXO1 or BLM-DNA2, reduced DNA end-resection and significantly limited the toxicity of the drug combination (Fig. 4L and M; Supplementary Figs. S4M–S4R). Furthermore, combination of peposertib and NVB, but not monotherapy, resulted in extensive accumulation of RAD51 foci, another measure of enhanced DNA end-resection and HR repair (Supplementary Fig. S4S). Taken together, these results indicate that enhanced DNA end-resection, resulting from the combination of peposertib and NVB, is the key initiating event leading to cell death.

We next identified sgRNAs in our genome-wide CRISPR knockout screen, which confer peposertib resistance (Fig. 1). Interestingly, sgRNAs directed against TP53 significantly increased the growth and survival of both A549 and H460 cells following peposertib exposure (Fig. 1D; Supplementary Fig. S1E). This finding was consistent with previous reports demonstrating that TP53 deficiency can suppress the cytotoxic activity of a DNA-PK inhibitor (17, 35, 47 48). We reasoned that TP53-deficient cells might acquire resistance to DNA-PKi through a compensatory upregulation of POLQ expression and an overall upregulation of MMEJ-mediated DSB repair. Accordingly, inhibition of POLθ with NVB might resensitize TP53-deficient cancer cells to peposertib.

To test this hypothesis, we examined a pair of isogenic retinal pigment epithelium (RPE) cell lines, which differ in their TP53 expression. These RPE versus RPE-TP53-KO cells were treated with increasing concentrations of peposertib. As predicted from the CRISPR screen and consistent with a previous report, the RPE-TP53-KO cells, unlike the parental RPE cells expressing wild-type TP53, exhibited increased resistance to DNA-PK inhibition (Fig. 5A) and increased sensitivity to POLθ inhibition (Fig. 5B; ref. 47). Importantly, the RPE-P53-KO cells were resensitized to peposertib by exposure to NVB (Fig. 5C). Indeed, the combination of peposertib and NVB significantly reduced the IC₅₀ of the peposertib from 2.053 to 0.2948 μmol/L (Supplementary Fig. S5A). Furthermore, the AUC was the least for RPE-P53-KO cells treated with the combination of peposertib plus NVB, compared with monotherapy or RPE cells expressing wild-type TP53, indicative of increased cell killing (Supplementary Fig. S5B). Similarly, in both A549 and H460-cell lines, depletion of TP53 correlated with increased expression of POLQ, increased sensitivity to NVB, and increased sensitivity to the drug combination, as shown by Bliss synergy curves (Supplementary Figs. S5C–S5K; ref. 41). The mechanism of this resensitization is due, at least in part, to increased POLθ expression, which is a consequence of cellular TP53 deficiency as previously reported (47). Pharmacologic inhibition of NHEJ by peposertib led to increased POLθ expression and dependence on MMEJ (Fig. 5D). Taken together, our results demonstrate that the combination of peposertib plus NVB is especially effective in killing TP53-deficient tumor cells. To extend these results, two additional TP53 mutant and HR-deficient cell lines (MDA-MB-436 and CAPAN-1), remained hypersensitive to the combination of peposertib plus NVB, even after their complementation with the wild-type BRCA1 or BRCA2 cDNAs, respectively (Supplementary Figs. S5L and S5M).

To further determine the effect of TP53 expression on POLQ expression, we next generated and analyzed primary human tumor organoid cultures from TP53-deficient high-grade serous ovarian carcinoma (HGSOC). Therapeutic drug responses of human tumor organoids have been shown to correlate, at least in part, with their corresponding patient responses in the clinic (33, 49, 50,). HGSOCs were chosen since 96% of these tumors are TP53 mutated, leading to deficiency in TP53 function and expression, though with the caveat that there may remain some low levels of TP53 function not seen with CRISPR-knockout of TP53 (51). Patient-derived xenograft derived organoids (PXO) were generated from tumor cells from two different TP53 mutant ovarian cancer PDX models, DF59 and DF68 (34, 52). DF59 has no discernible TP53 protein expression by Western blot analysis, whereas DF68 has reduced TP53 protein expression (53).

To determine whether combination of peposertib and NVB would act synergistically in an organoid culture, we treated these two TP53-deficient PXO models with peposertib, NVB, or the drug combination. Both organoid models were sensitive to monotherapy with either peposertib or NVB but the combination treatment, at the monotherapy doses, resulted in a significant increase in lethality as assessed by cell viability using the Bliss synergy model (41). The combination of peposertib and NVB resulted in synergistic cytotoxicity in both PXOs (Fig. 5E and F, top). Indeed, the combination of peposertib and NVB significantly reduced the IC₅₀, thereby improving the therapeutic index of the two drugs (Fig. 5E and F, bottom). We next sought to correlate the tumor suppressive effects with the molecular mechanisms described in Fig. 4. Indeed, both DF59 and DF68 PXOs exhibited a statistically significant increase in γH2AX and p-RPA staining by IHC upon treatment with the combination of peposertib and NVB, consistent with an increase in DSB end resection (Fig. 5G–L). Next, based on the synthetic lethal mechanism of increased DSB end-resection, we performed SMART assays on DF59 PXOs. We observed a marked increase in SMART fiber length in the combination treatment group compared with peposertib-alone treatment or NVB-alone treatment, consistent with enhanced DSB end-resection (Supplementary Figs. S5N–S5O). Thus, we conclude that peposertib and NVB act synergistically in TP53-deficient ovarian cancer organoid models by generating toxic levels of DSB DNA end-resection and cell death.

We next determined whether the combination of DNA-PK and POLθ inhibition would be an effective treatment for a TP53-deficient tumors in vivo (Fig. 6). Mice bearing TP53 mutant ovarian tumors (the DF59 PDX model) were treated with peposertib, NVB, or the drug combination, and tumor growth was monitored by bioluminescence imaging (BLI; Fig. 6A). Peposertib alone did not significantly affect the growth of these TP53-deficient HGSOC tumors, consistent with the inability of a DNA-PK inhibitor to kill TP53-mutant tumor cells in vitro (Fig. 6B). NVB demonstrated significant monotherapy tumor growth inhibition in TP53-deficient HGSOC tumors in vivo, consistent with our previous report (25). Critically, the combination of these two targeted DNA repair inhibitors further enhanced tumor growth inhibition despite the lack of peposertib monotherapy efficacy. Interestingly, the targeted drug combination alone killed the tumor cells in vivo, and no additional cytotoxic therapy, such as IR or an additional chemotherapeutic agent, was required to achieve this synergy. Again, the enhanced antitumor activity of the combination is caused in part by the increased expression of POLQ generated by peposertib exposure (Fig. 6C). This was evident in the TP53-mutant cells, which likely have intrinsic peposertib resistance and high baseline levels of POLQ expression (Fig. 5; Supplementary Fig. S5). Therefore, the addition of NVB reverses the resistance of TP53-mutant cancer cells to peposertib. The enhancement of tumor killing by the drug combination was also demonstrated by the pharmacodynamic increase in DSB end-resection (p-RPA level) and increased DNA damage (γH2AX) observed in extracted tumor cells from the treated mice (Fig. 6D). Importantly, no significant toxicity to normal mouse tissue was observed from this drug combination as evidenced by comparable increase in bodyweights of mice during treatment and hematologic examination of mice following 28 days of treatment (Supplementary Figs. S6A and S6B). Given the lack of DF59 tumor growth inhibition with peposertib monotherapy despite signs of increased DNA end-resection, we treated A549, A549 DNA-PK KO, and A549 DNA-PK P53 DKO xenograft-bearing nude mice with vehicle or NVB. Consistently, tumor growth inhibition after 28 days of treatment was greatest for NVB treated A549 DNA-PK P53 DKO xenografted mice (Supplementary Figs. S6C–S6D).

The DNA damage response maintains genomic integrity by responding to endogenous and exogenous DNA damage and activating specific DNA repair pathways. DNA damage is repaired by one of at least six pathways acting on specific types of DNA damage. Cells become dependent on an alternative pathway if the primary DNA repair pathway is lost due to mutation, deletion, epigenetic silencing, or inhibition with a targeted drug. Indeed, HR-deficient tumor cells are dependent on PARP1-mediated base excision repair or POLθ-mediated MMEJ (25, 37, 38, 54). Here we sought to identify a DNA repair pathway that is upregulated following cellular exposure to peposertib, a targeted and highly selective inhibitor of the kinase DNA-PK in the NHEJ pathway (22, 35). Through a whole-genome CRISPR screen, we demonstrated that loss of POLQ, or loss of other components of the MMEJ pathway, are synergistic with peposertib or with genetic knockdown of NHEJ both in vitro and in vivo. Consistent with this result, POLQ mRNA and POLθ protein expression are upregulated after peposertib exposure. Also, as expected, the combination of peposertib plus POLθ inhibitors, NVB or ART558, killed cancer cells. The synergy of peposertib plus NVB was observed across multiple cancer cell lines, including those with either HR repair proficiency or deficiency.

On the basis of published data and on the results presented here, the mechanism of synthetic lethality between combination NHEJ and MMEJ inhibition is an increase in DNA DSB end resection leading to loss of genetic material, for example by utilization of mutagenic repair processes such as single-strand annealing, and subsequent cellular death through apoptosis (Supplementary Fig. S6E; refs. 25, 42). Cells have several regulatory processes that normally limit the amount of DSB end resection, and these activities function in a stepwise manner (Supplementary Fig. S6F). Under normal circumstances, DSBs can be repaired rapidly by NHEJ and blunt end ligation (1). Because, NHEJ occurs throughout the cell cycle, it is the predominant DSB repair pathway and can quickly join unresected ends, thereby limiting DNA end-resection. Thus, NHEJ is the first barrier to DNA end-resection, and inhibition of DNA-PK with peposertib eliminates this barrier. A second barrier to end resection is the 53BP1–Shieldin complex, which further limits resection and fortifies blunt end ligation by NHEJ (55–60). This complex promotes NHEJ by protecting DNA ends from additional resection. Indeed, depletion of 53BP1 or other SHLD proteins increases DSB end-resection, thereby providing a mechanism of HR restoration in BRCA1-deficient cells as well as a predictive biomarker of POLθ inhibitor response (Supplementary Fig. S6F; refs. 61–63). Finally, if a low level of DNA end-resection occurs, producing short 3′ overhangs and the preferred substrate of POLθ-mediated MMEJ, the POLθ-mediated MMEJ pathway further limits DNA end-resection by performing MMEJ repair. Thus, MMEJ-mediated repair provides a third barrier to DSB end-resection (64). On the basis of this model, disruption of NHEJ with peposertib and disruption of MMEJ with NVB results in a massive level of DSB end-resection and ultimately to apoptosis (Supplementary Fig. S6F).

Prior studies have demonstrated that MMEJ is a backup mechanism for repairing DSBs when NHEJ is not available. When NHEJ is genetically perturbed, there is a compensatory increase in POLθ-mediated MMEJ activity (39, 40). Our findings are consistent with these prior reports and demonstrate that pharmacologic inhibition of NHEJ by peposertib directly increases POLQ mRNA and POLθ protein expression, resulting in increased MMEJ activity and increased dependence on POLθ. We have previously reported that increased POLQ expression is an important predictive biomarker of responsiveness to POLθ inhibition by NVB (25). Indeed, low-dose treatment with peposertib potentiated NVB tumor killing. Thus, our study indicates that upregulation of POLθ-mediated MMEJ activity is a major intrinsic resistance mechanism to the use of peposertib.

DNA-PKi resistance can be overcome when peposertib is combined with a MMEJ inhibitor (NVB). Furthermore, peposertib, among other DNA-PK inhibitors, is currently undergoing early-phase clinical trial investigation as a potentiating agent in combination with radiation or chemotherapy (65). In general, the use of chemotherapy and/or radiotherapy in conjunction with targeted therapy has been limited by toxicity (66). Importantly, our study did not require supplemental radiation or cytotoxic chemotherapy to induce DNA damage. The combination of the targeted NHEJ inhibitor (peposertib) and the MMEJ inhibitor (NVB) alone induced DNA damage and led to tumor killing via apoptosis. These results suggest that a combination of a NHEJ inhibitor (peposertib) and a MMEJ inhibitor (NVB, or ART558) will be both safely tolerated and effective in treating cancer in patients.

Prior work by a number of groups has demonstrated that TP53 deficiency can suppress the toxicity of DNA-PK inhibition (17, 35, 47, 48). Consistently, our whole-genome CRISPR knockout screen identified TP53 loss as a resistance mechanism resulting from peposertib-mediated DNA-PK inhibition. TP53 is the most commonly mutated tumor suppressor, occurring in 50% of newly diagnosed solid tumors (67). TP53 deficiency is associated with resistance to DNA damaging agents and DNA repair pathway inhibitors (68, 69). Furthermore, TP53 deficiency in cancers portends a worse prognosis (68, 70). Kumar and colleagues has demonstrated that POLQ mRNA expression is increased upon TP53 deletion and that there is a correlation between POLQ mRNA overexpression and TP53 mutation status in a wide variety of cancers in The Cancer Genome Atlas dataset (47). Our findings are consistent with this report and extend the results by using the specific POLθ inhibitor, NVB. We hypothesized that TP53-deficient cancer cells acquire resistance to DNA-PK inhibition through a compensatory upregulation of POLQ expression and an overall upregulation of MMEJ-mediated DSB repair. Accordingly, the combination of peposertib plus NVB alone resulted in synthetic lethality in TP53-deficient tumor cell lines, organoid cultures, and PDX models, without the need for additional cytotoxic therapy. Accordingly, we show that inhibition of POLθ with NVB resensitizes TP53-deficient cancer cells to peposertib. Thus, our findings establish a strong rationale for the combination of two targeted DNA damage repair inhibitors (NHEJ-peposertib) and (MMEJ-NVB) in TP53-deficient cancers.

Predictive biomarkers are critical to the successful development of targeted DNA repair in inhibitors (71). These biomarkers aid in the identification of patients with cancer whose tumors are most likely to be sensitive or resistant to the targeted therapy. In our current study, we propose a model in which multiple cellular barriers block DNA end-resection. When these barriers are disrupted, there is the accumulation of toxic DSB end-resected DNA, leading to cell death. This model also predicts responsiveness to NVB-mediated POLθ inhibition. The first barrier to DNA end-resection is NHEJ mediated by DNA-PK. We show that DNA-PK inhibition by peposertib increases DNA end-resection, POLQ expression, POLθ activity, and ultimately sensitivity to NVB. Thus, loss of NHEJ is a predictive biomarker for responsiveness to NVB. Similarly, the 53BP1–Shieldin complex is a second barrier to DNA end-resection and its loss predicts response to POLθ inhibition (25, 42, 72). We previously reported that POLQ expression, and hence MMEJ activity, is the third barrier to DNA end-resection, correlating with response to NVB treatment. Thus, high POLQ expression is a strong predictive biomarker of POLθ inhibitor response. Because loss of these barriers to DNA end-resection results in increased DSB end-resection, assessment of p-RPA accumulation could serve as a pharmacodynamic biomarker for POLθ inhibition. Finally, as noted above, TP53 deficiency or biallelic mutation results in increased POLθ expression and sensitivity to POLθ inhibition and is an important predictive biomarker of response to the DNA-PK and POLθ inhibitor combination. How TP53 normally suppresses POLQ expression remains an important unanswered question. Thus, our study supports the investigation of the combination of targeted NHEJ and MMEJ inhibitors in TP53-mutant cancers.

J. Patterson-Fortin reports grants from Doris Duke Charitable Foundation during the conduct of the study. J.F. Liu reports personal fees from AstraZeneca, Clovis Oncology, Eisai, EpsilaBio, Genentech, Regeneron Therapeutics, and GlaxoSmithKline outside the submitted work. G.I. Shapiro reports grants and personal fees from MERCK KGaA/EMD Serono during the conduct of the study; grants from Eli Lilly, Merck & CO.; personal fees from Bicycle Therapeutics, Cybrexa Therapeutics, Bayer, Boehringer Ingelheim, ImmunoMet, Artios, Concarlo Holdings, Syros, Zentalis, CytomX Therapeutics, Blueprint Medicines, Kymera Therapeutics, Janssen, and Xinthera outside the submitted work; also has a patent for Dosage regimen for sapacitabine and seliciclib issued to Cyclacel Pharmaceuticals and Geoffrey Shapiro and a patent for Compositions and methods for predicting response and resistance to CDK4/6 inhibition pending to Liam Cornell and Geoffrey Shapiro. D. Kozono reports personal fees from Genentech/Roche outside the submitted work. A.D. D'Andrea reports personal fees from Acerand Therapeutics (Hong Kong), Bayer AG, LSV Management, Constellation Pharma, Faze Medicines, Inc., GlaxoSmithKline, LAV Global Management Company Limited, and Patheon Pharmaceuticals; other support from AstraZeneca and Epizyme, Inc.; grants and other support from Bristol Myers Squibb, Lilly Oncology, and EMD Serono; personal fees and other support from Cedilla Therapeutics, Cyteir, Ideaya, Inc., Impact Therapeutics, Oncolinea, and Zentalis Pharmaceuticals/Zeno Management Inc.; grants from Moderna; other support from Pfizer; grants and personal fees from Tango Therapeutics outside the submitted work. No disclosures were reported by the other authors.

J. Patterson-Fortin: Conceptualization, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing. A. Bose: Formal analysis, investigation, methodology, writing–review and editing. W.-C. Tsai: Conceptualization, formal analysis, investigation, writing–review and editing. C. Grochala: Investigation, methodology, writing–review and editing. H. Nguyen: Data curation, formal analysis, investigation, writing–review and editing. J. Zhou: Investigation, writing–review and editing. K. Parmar: Formal analysis, investigation, writing–review and editing. J.-B. Lazaro: Investigation, writing–review and editing. J. Liu: Resources, writing–review and editing. K. McQueen: Investigation, writing–review and editing. G.I. Shapiro: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. D. Kozono: Formal analysis, validation, writing–review and editing. A.D. D'Andrea: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This work was supported by grants from the US NIH (R01HL052725), the Breast Cancer Research Foundation, the Fanconi Anemia Research Fund, the Ludwig Center at Harvard, and the Smith Family Foundation (A.D. D'Andrea), and Grant 2021087 from the Doris Duke Charitable Foundation to J. Patterson-Fortin. This work was also supported by a Sponsored Research Agreement from EMD Serono (to A.D. D'Andrea and G.I. Shapiro).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.