A first-in-class Polymerase Theta Inhibitor selectively targets Homologous-Recombination-Deficient Tumors

A small-molecule screen identifies NVB as a specific POLθ inhibitor

The ATPase activity of POLθ is required for the survival of HR-deficient cells ¹⁴. We took advantage of the strong in vitro ATPase activity of purified POLθ (Extended Data Fig. 1a) and performed a large-scale small-molecule screen to identify POLθ inhibitors based on the ADP-Glo luminescent assay (Fig. 1a,b). We screened a total of 23,513 small-molecules across multiple libraries (enriched with known bioactive compounds, Supplementary Tables 1–2) in duplicate, and identified 72 compounds that significantly (z-score < −4) reduced POLθ ATPase activity (Fig. 1b, Supplementary Table 3). These compounds account for only 0.3% of the total small-molecules screened, arguing for a low rate of false-positive hits. A secondary screen using the 72 initial hits was performed in the presence and absence of ssDNA, in order to confirm the initial screen results and to exclude compounds that directly interact with ssDNA (Extended Data Fig. 1b). Among the 72 small molecules, 10 hits were identified to have inhibited POLθ ATPase more than 50% both with and without ssDNA (≤15% difference in inhibition) (Supplementary Table 4). Aurintricarboxylic acid (ATA), reactive Blue 2 (RB), suramin (SUR), and novobiocin (NVB) were the four most potent POLθ inhibitors.

These top hits were individually verified in a ³²P-based radiometric ATPase assay with POLθ and another known DNA repair-related ATPases SMARCAL1 ²⁹. Only NVB showed high specific inhibition of POLθ with little effect on SMARCAL1, while other compounds inhibited both ATPases (Extended Data Fig. 1c). We thus focused our future studies on novobiocin. Next, we performed a dose response of NVB on POLθ ATPase activity using ADP-Glo assay, along with six other DNA repair related enzymes HSP90AA1, TRIP13, BLM, RAD51, SMARCAL1 and CHD1 (Fig. 1c and Extended Data Fig. 1d). Novobiocin inhibited POLθ with an IC50 of 24 μM, but it did not significantly inhibit other ATPases, supporting the specificity inhibition of novobiocin on POLθ.

To further evaluate the specificity of novobiocin (NVB) for POLθ, we analyzed its binding capacity. NVB-conjugated beads pulled down the purified ATPase domain of POLθ, but not purified SMARCAL1, CHD1, BLM, and RAD51, suggesting that NVB directly and specifically binds to the POLθ ATPase domain in vitro (Extended Data Fig. 1d–f). NVB-conjugated beads also pulled down POLθ-GFP expressed in cells, and this binding was competed by free NVB (Extended Data Fig. 1g–h). Thermal shift assays using cell lysate of GFP-POLθ expressing HEK293T cells showed that NVB stabilized GFP-POLθ (Extended Data Fig. 1i). Thermal shift assays with purified POLθ-ATPase domain showed that NVB increased POLθ stability in a dose dependent manner, further suggesting a direct ligand-protein binding of NVB to POLθ (Extended Data Fig. 1j). As controls, NVB did not increase the thermal stability of BLM or MRE11 (Extended Data Fig. 1k–l). Finally, we performed molecular docking to identify the NVB binding site using the published crystal structure of POLθ ATPase domain ³⁰. The computational docking results revealed a deep tunnel within the POLθ ATPase domain that could accommodate NVB, which is located next to the ATP binding pocket (Fig. 1d). Taken together, these data demonstrate that NVB binds directly and specifically to the POLθ ATPase domain in vitro.

We next assessed whether cellular exposure to NVB has the same consequences as POLθ depletion. Both the ATPase and the polymerase domains contribute to efficient MMEJ repair activity ²⁷, and the PARP-dependent recruitment of POLθ to DSB sites is a crucial step in this function ^14,16. Phenocopying the ATPase mutant of POLθ, inhibition of POLθ by NVB prevented the recruitment of POLθ to laser micro-irradiated DNA damage sites in human cells (Fig. 1e). Consequently, NVB inhibited the MMEJ activity in U2OS cells, as measured by the GFP-based EJ2 repair assay, but it had little effect on HR activity measured by the DR-GFP reporter assay (Fig. 1f). POLθ also functions as an anti-recombinase that antagonizes HR, and depletion of POLθ increases RAD51 ¹⁴ and γH2AX ³¹ foci assembly. Similarly, more RAD51 foci and γH2AX foci were observed after IR in the presence of NVB in U2OS cells (Extended Data Fig. 2a–b), although the effect was not as strong as previously reported with siRNA POLθ knockdown ¹⁴. Taken together, these results establish that NVB-mediated POLθ inhibition impairs POLθ DNA repair function and phenocopies POLθ depletion in human cells.

NVB specifically Kills HR-deficient tumors

NVB is a safe drug and has already been used in anti-cancer trials, although the treated patients were not preselected to have HR-deficient cancers ^32–35. Since depletion of POLθ is synthetically lethal in HR-deficient tumor cells ^14,16, we next directly tested whether NVB can kill HR-deficient tumor cells in animal models. First, we evaluated the efficacy of NVB in tumors derived from the K14-Cre-Brca1^f/f;Trp53^f/f genetically-engineered mouse model (GEMM) of triple-negative breast cancer (TNBC) ^36,37. Tumors from this model were transplanted into immunocompetent FVB/129P2 syngeneic mice that were subsequently treated with NVB. NVB strongly suppressed the growth of these Brca1-deficient GEMM-derived tumors (Fig. 1g and Extended Data Fig. 3a). The NVB-treated tumors were significantly smaller compared to the vehicle-treated after 7 days of treatment and beyond. NVB prolonged the overall survival of the tumor-bearing mice by 3-fold compared to vehicle, with median survival times of 29 and 10 days for NVB- and vehicle-treated mice, respectively (Fig. 1h).

Next, we tested whether the efficacy of NVB in vivo was specific to HR-deficient tumors. FANCF-deficient TOV21G cells are human ovarian cancer cells with reduced HR capacity, which grow well as tumor xenografts in mice ^38,39. We thus performed a mouse xenograft study with the FANCF-deficient (TOV21G + empty vector) and the FANCF-complemented (TOV21G + FANCF cDNA) cell lines (Fig. 2a,b). In vitro, TOV21G + EV cells but not TOV21G + FANCF cDNA cells were sensitive to olaparib and NVB (Extended Data Fig. 3b,c) ³⁹. In vivo, while vehicle treatment had no impact on the growth of FANCF-deficient or FANCF-complemented tumors, NVB specifically impaired the growth of the FANCF-deficient tumors, with no effect on FANCF-complemented tumors (Fig. 2a,b). We also measured RAD51 foci as a pharmacodynamic biomarker and showed that RAD51 foci were strongly induced in the treated tumors (Extended Data Fig. 3d–f). These results demonstrate that NVB strongly and specifically suppresses the growth of HR-deficient tumors in vivo.

To confirm and strengthen the in vivo results, we studied isogenic pairs of HR-deficient and HR-proficient cells in vitro. We generated BRCA1 and BRCA2 knockout RPE1 cells (referred as BRCA1^−/− and BRCA2^−/−) in a TP53^−/− background, which were sensitive to PARPi (Extended Data Fig. 4a,b). Consistent with our in vivo data, clonogenic survival assays showed that NVB significantly reduced the survival of BRCA1^−/− and BRCA2^−/− cells, compared to the isogenic WT cells (Fig. 2c and Extended Data Fig. 4c). We also tested the top initial hits from our screen for their ability to differentially affect the survival of BRCA1^−/− and WT cells. Among those, NVB had the highest efficacy in selectively killing the BRCA1^−/− cells (Extended Data Fig. 4d).

To demonstrate the cellular outcome of NVB exposure, we next showed that NVB exposure induces apoptosis in BRCA1^−/− but not WT cells in a dose dependent manner (Extended Data Fig. 5a). NVB induced chromosomal aberrations and radial chromosomes (a marker of genomic instability) in BRCA1^−/− cells, even in the absence of the crosslinking agent mitomycin C (MMC) (Extended Data Fig. 5b–d), confirming that NVB increases DNA damage. These results further demonstrate that NVB phenocopies POLθ depletion, and they indicate that NVB-mediated POLθ inhibition is synthetically lethal with HR deficiency and specifically induces cell death in HR-deficient tumor cells.

POLθ is the major target of NVB in human cells

Off target effects are major concerns of biological small inhibitors. To further evaluate whether POLθ is the specific target of NVB and to demonstrate the specificity of NVB in cells, we knocked out the POLQ gene in RPE1 cells using CRISPR-Cas9 and asked whether loss of POLQ alleviates the cytotoxicity of NVB. POLQ knockout cells were confirmed by genomic sequencing and their increased sensitivity to irradiation (Fig. 2d). Importantly, RPE1 POLQ^−/− cells showed increased tolerance to NVB compared to the WT cells (Fig. 2e). In addition, POLQ^−/− U2OS cells were also more resistant to NVB than WT cells (Extended Data Fig. 6a–e). These data indicate that POLθ is a major target of NVB in human cells.

NVB is a known coumarin antibiotic that inhibits DNA gyrase in bacteria ⁴⁰. NVB has been reported to inhibit the ATPase of HSP90 and of the DNA gyrase homolog topoisomerase II (TOP2) in eukaryotes ^41,42, suggesting that these enzymes may be off-target enzymes of NVB. However, other reports suggest that NVB does not inhibit TOP2 in eukaryotic cells ⁴³. Additionally, the reported IC₅₀ of these targets are very high (approximately 700 μM for HSP90 and 300 μM for TOP2), compared with the IC₅₀ of POLθ (24 μM by ATPase inhibition assay, Fig. 1c). Nevertheless, we next examined whether the cytotoxic effect of NVB in HR-deficient cells resulted from off-target inhibition of HSP90 or TOP2. Unlike the well-characterized HSP90 inhibitor PU-H71, NVB treatment at a concentration that killed HR-deficient cells (i.e. 100 μM) did not induce the degradation of HSP90 clients such as AKT1, BRCA1, or CDK6, or increase the expression of HSP70 (Extended Data Fig. 7a–c). These results suggest that the cytotoxic effect of NVB is not related to HSP90 inhibition. Furthermore, the combination of NVB and the TOP2 inhibitor etoposide exhibited additive cytotoxicity in TOV21G and CAPAN1 cells (Extended Data Fig. 7d,e), suggesting that NVB and etoposide do not share TOP2 as a target and that NVB cytotoxicity is unrelated to TOP2 inhibition.

NVB potentiates the effect of PARPi in HR-deficient tumors

PARPi are currently being evaluated in different combination therapy settings with cytotoxic chemotherapy, radiation, targeted therapies, and immunotherapies ⁴⁴. We explored whether the combination of NVB and a PARPi may be more effective than PARP inhibition alone in killing HR-deficient tumors. Indeed, NVB further sensitized BRCA1^−/− RPE1 cells to the PARPi rucaparib, but not BRCA-proficient WT cells in clonogenic assays (Fig. 2f). We studied the combination with a second PARPi, olaparib, where synergism of NVB and olaparib was also observed in HR-deficient TOV21G + EV cells, but not in HR-proficient TOV21G + FANCF cells (Extended Data Fig. 7f). Importantly, NVB reduced the IC₅₀ of rucaparib by more than 20-fold in BRCA1^−/− RPE1 cells and the IC₅₀ of olaparib more than 40-fold in TOV21G cells compared to their respective HR-proficient counterparts (Fig. 2f and Extended Data Fig. 7g). One strategy to prevent acquired resistance during treatment is combination therapy. The observed synergy between NVB and PARPi prompted us to evaluate whether NVB could overcome PARPi resistance in HR-deficient tumors.

POLθ inhibition by NVB overcomes acquired PARPi resistance

Multiple PARP inhibitors (PARPi) including olaparib, niraparib, rucaparib, and talazoparib have received approval for the treatment of ovarian and breast tumors with HR deficiency ⁴⁵. Despite the remarkable initial tumor response and improved progression-free survival (PFS) in HR-deficient patients ^5–8, acquired PARPi resistance is emerging as an unmet medical need ⁹.

To evaluate whether POLθ inhibition could represent a novel therapeutic option for PARPi-resistant tumors, we first tested the efficacy in vivo of NVB in the patient-derived xenograft (PDX) model DF83, generated from an HR-deficient (loss of RAD51C expression), human high-grade serous ovarian carcinoma ⁴⁶. NSG mice bearing the HR-deficient DF83 ovarian cancer PDX model were treated with olaparib, NVB, or a combination of the two, and tumor growth was monitored by bioluminescence imaging (BLI) (Fig. 3a). In order to determine the combined effect with NVB, olaparib was delivered at a submaximal dose (50 mg/kg daily), and we observed only a small degree of tumor growth inhibition. Treatment of DF83 with NVB alone (75 mg/kg, twice a day) led to tumor regression, while tumors in vehicle-treated mice showed exponential growth (Fig. 3b). Strikingly, when the DF83 tumor-bearing mice were given NVB and olaparib in combination, we observed complete tumor regression, and few tumor cells were detectable via BLI on day 28 in this group (Fig. 3a,b). These in vivo results suggest NVB may be useful in combination with PARPi to achieve better efficacy and prevent drug resistance in HR-deficient tumors. No significant toxicity to normal mouse tissue was observed from the drug combination.

Next, we evaluated the efficacy of NVB alone and in combination with olaparib in the PARPi-resistant PDX model DF59, which was isolated from a patient with a germline BRCA1 mutation and acquired PARPi resistance. This model harbors biallelic mutations in the TP53BP1 locus, providing a mechanism of the acquired PARPi resistance ⁴⁶. In the DF59 PDX model, there was no response to olaparib monotherapy, but NVB monotherapy substantially reduced tumor growth, indicating a lack of cross-resistance between PARP inhibition and NVB. Importantly, the combination of NVB and olaparib further inhibited DF59 tumor growth, with tumor regression in the first two weeks (Fig. 3c,d). In contrast to DF83 and DF59 models, the BRCA1 WT, HR proficient PDX model DF149 was resistant to both drugs and to the combination (Fig. 3e,f). These data demonstrate that NVB can inhibit the growth of a BRCA1-mutated, PARPi-resistant tumors in vivo and that the combination of NVB and a PARPi is particularly effective in treating at least some PARPi-resistant tumors.

NVB induces excessive DSB end resection and RAD51 foci

We next studied the mechanism of cell killing by NVB. Previous studies demonstrated that the ATPase domain of POLθ counteracts RPA binding at resected DSBs to facilitate their repair by MMEJ ²⁵. We therefore hypothesized that DSBs will be excessively resected when POLθ is inhibited. We measured DNA resection at DSBs, generated by AsiSI, in U2OS cells in the presence or absence of NVB. In this assay, DSBs were induced by 4-hydroxy-tamoxifen (4-OHT), and the percentage of single-stranded DNA (ssDNA) was measured by qPCR ⁴⁷. The amount of ssDNA around the DSB was significantly higher when cells were treated with NVB versus DMSO (Fig. 4a). Consistent with this result, NVB-treated tumor samples from the DF59 PDX model exhibited higher levels of phospho-RPA32 (p-RPA32), a measure of ssDNA accumulation, compared to vehicle-treated samples (Fig. 4b). These data suggest that POLθ inhibition by NVB can promote DSB end resection.

We previously showed that depletion of POLθ in HR-deficient tumors causes RAD51 accumulation ¹⁴, perhaps resulting from excessive DSB end resection. Similarly, 53BP1 and POLθ are synthetic lethal, and 53BP1/POLθ double knockout cells also have increased end resection and elevated RAD51 accumulation ⁴⁸. We reasoned that NVB may induce non-functional RAD51 foci in HR-deficient tumors with PARPi resistance in vivo. To test this hypothesis, we collected tumor cells from the PARPi and/or NVB treated PDX models and evaluated RAD51 foci by immunohistochemistry (Fig. 4c). Since DF83 is an HR-deficient and PARPi sensitive model, RAD51 foci were not observed. The HR-restored and PARPi-resistant DF59 model, with biallelic 53BP1 mutations, exhibited a strong increase in RAD51 staining when treated with NVB alone or in combination with olaparib (Fig. 4c,d). Similarly, TOV21G xenograft tumors exhibited increased RAD51 foci by immunohistochemistry in vivo after NVB treatment, which was further enhanced by olaparib (Extended Data Fig. 3d–f). Next, we evaluated the effect of NVB on a PARPi resistant clone (MDA-MB-436-R) of the BRCA1-mutated breast cancer cell line MDA-MB-436 ⁴⁹. Like the DF59 PDX model, this resistant cell line restored HR by two mechanisms, namely the stabilization of mutant BRCA1 protein and the loss of 53BP1. While highly resistant to olaparib, the MDA-MB-436-R cells were sensitive to NVB similarly to the parental cells (Fig. 4e,f). Importantly, NVB induced RAD51 accumulation in MDA-MB-436-R cells but not in the parental cells (Fig. 4g). Since the parental and MDA-MB-436-R cells are both sensitive to NVB, the elevated RAD51 levels appear to be non-functional for HR, consistent with the increased resection observed in the presence of NVB. Taken together, our data suggest that the elevated DSB end resection, as well as the subsequent accumulation of ssDNA intermediates and non-functional RAD51 loading, are a plausible mechanism of cell killing by NVB in PARPi resistant HR-deficient tumor cells (Fig. 4h).

NVB overcomes multiple PARPi resistance mechanisms

To better understand the mechanism by which NVB overcomes PARPi resistance, we generated several PARPi-resistant clones from BRCA1^−/− RPE1 cells by gradually exposing them to increasing concentrations of PARPi (Fig. 5a). We selected four clones (named R1 to R4), all of which have acquired resistance to olaparib (Extended Data Fig. 8a–c). Multiple PARPi resistance mechanisms were identified in these clones, corresponding to published mechanisms ⁹. Replication fork stabilization was evident in the R1 and R2 clones (Extended Data Fig. 8d). The R1, R3, and R4 clones have restored RAD51 foci, suggesting restored HR activity in these clones (Extended Data Fig. 8e). The restoration of HR repair in these clones may have resulted in part from their downregulation of the Shieldin complex and subsequent downregulation of NHEJ repair. Indeed, the R3 clone had decreased expression of REV7, a component of the Shieldin complex, and the R4 clone had decreased 53BP1 expression (Extended Data Fig. 8f–g). None of the clones had re-expressed BRCA1, thereby excluding somatic reversion of BRCA1 as an underlying PARPi resistance mechanism (Fig. 5b and Extended Data Fig. 8h). Strikingly, all four PARPi-resistant clones remained sensitive to NVB to a similar extent as the parental RPE-BRCA1^−/− cells (Fig. 5c,d), suggesting that NVB may overcome multiple mechanisms of acquired resistance to PARPi. To demonstrate that genetic inhibition of POLθ also causes cell death in the PARPi-resistant BRCA1^−/− clones, we depleted POLθ in R1, R2, and the parental cells and measured their cell survival. Upon POLθ depletion, the parental BRCA1^−/− cells and the PARPi-resistant clones, but not WT RPE1 cells, showed reduced survival, suggesting that POLθ inhibition was the mechanism for NVB cytotoxicity in those cells (Extended Data Fig. 8i).

We also examined more clinically relevant PARPi-resistant, HR-deficient models, using tumor cells derived from patients. MDA-MB-436-R cells, with stabilized mutant BRCA1 protein and loss of 53BP1, were NVB sensitive (Fig. 4e,f). In addition, we studied the PARPi-resistant clone UWB1.289-YSR12 (derived from the BRCA1-null human ovarian cancer cell line UWB1.289), which acquired resistance to PARPi via stabilization of replication forks ⁵⁰. These cells also remained sensitive to NVB, although they are highly resistant to olaparib (Extended Data Fig. 9a–b), again suggesting that acquired resistance to PARPi does not determine cross-resistance to NVB. Taken together, these data show that NVB may overcome multiple mechanisms of acquired resistance to PARPi.

POLθ expression is a predictive biomarker for NVB sensitivity

POLθ mRNA expression is increased in HR-deficient tumors and in several other cancers associated with poor prognosis ^14,51. We asked whether increased POLθ mRNA and protein expression may serve as a predictive biomarker for NVB sensitivity. As expected, CRISPR-mediated knockout of BRCA1 in RPE-1 cells resulted in increased POLθ expression as compare to the parental cells (Fig. 5e and Extended Data Fig. 9c). Interestingly, all selected PARPi-resistant clones (R1-R4) retain high POLθ expression, revealing a direct correlation between POLθ expression and NVB sensitivity (Fig. 5e). This correlation was also present in the BRCA1-deficient MDA-MB-436 cells. MDA-MB-436-R cells with acquired PARPi-resistance exhibited elevated POLθ mRNA expression and retained NVB sensitivity, while complementation of the parental cells with WT BRCA1 cDNA decreased POLθ expression and NVB sensitivity (Fig. 5f). The direct correlation of POLθ expression and NVB sensitivity was further observed in the HR-deficient PDX models, where the elevated POLθ expression in DF83 and DF59 correlated with their high sensitivity to NVB in vivo (Fig. 5g and Fig. 3a–f).

Another known mechanism of PARPi resistance in BRCA1/2^−/− tumors in the clinic is somatic reversion of the BRCA1/2 genes ⁴. We determined whether NVB could overcome PARPi resistance caused by BRCA2 reversion. Unlike other PARPi resistant cells we analyzed, BRCA2-deficient tumor cells with near-perfect somatic reversion or CRISPR-edited reversion of BRCA2 (PEO4 and CAPAN1-CR) ^10,52 were not only resistant to PARPi but also to NVB, when compared to their HR-deficient parental cell lines (PEO1 and CAPAN1) (Fig. 5h–i and Extended Data Fig. 9d–e). These data suggest that NVB can overcome some but not all mechanisms of PARPi resistance. Importantly, the low POLθ expression was a predictive biomarker for NVB resistance. For the PARPi-resistant cell lines with BRCA2 reversion mutations (PEO4 and CAPAN1-CR), their resistance to NVB correlated with their low POLθ protein expression levels by Western blot (Fig. 5j and Extended Data Fig. 9f). The resistance of the DF149 PDX model in vivo correlated with its low POLθ expression (Fig. 5g). Also, the loss of POLθ in RPE1 cells correlated with resistance to NVB (Fig. 2e). Taken together, in all cases we examined, POLθ expression level correlated with cellular sensitivity to NVB, suggesting that POLθ expression in a tumor biopsy could provide a convenient predictive biomarker for patient enrollment in future clinical trials with NVB or other POLθ inhibitors.

Compounds, inhibitors and antibodies

Chemical compounds, including PARP inhibitors Olaparib (#S1060) and Rucaparib (#S1098), Novobiocin (#S2492), etoposide (#S1225), and PU-H71 (#S8039) were purchased from Selleckchem. Chemicals were dissolved in DMSO and kept in small aliquots at −80°C. The key sources of antibodies are: POLθ (Sigma #SAB1402530), γH2AX (Millipore #05–636), RAD51 (Santa Cruz Biotechnology, SCB, #sc-8349), CDK6 (SCB, #sc-7961), AKT1 (SCB #sc-5298), HSP70 (SCB #sc-24), HSP90 (SCB #sc-13119), and FANCF (Everest Biotech, #EB06112).

Protein purification

A POLθ fragment (ΔPol) containing the ATPase domain with a RAD51 binding site (amino acids 1 to 987) was cloned into pFastBac-C-Flag and purified from baculovirus-infected SF9 insect cells. POLθ expressing cells were lysed in lysis buffer (500 mM NaCl, 0.01 % NP-40, 0.2 mM EDTA, 20% Glycerol, 1 mM DTT, 0.2 mM PMSF, 20 mM Tris [pH 7.6]) supplemented with Halt protease inhibitor cocktail (Thermo Fisher) and Calpain I inhibitor (Roche). The cell lysates were incubated with M2 Flag beads (Sigma) for 3 hours at 4 ^oC on a spinning wheel. The beads were washed 3 times lysis buffer and protein was then eluted in lysis buffer supplemented with 0.2 mg/ml of Flag peptide (Sigma). The protein was then concentrated in lysis buffer using 10 kDa centrifugal filters (Amicon). SMARCAL1, CHD1, BLM, RAD51 and TRIP13 were purified in house using similar protocol. BLM-ATPase domain and MRE11-catalytic domains were purified by affinity followed by gel-filtration chromatography. Purified HSP90AA1 protein was purchased from Abcam (ab78425).

High throughput small Molecule screen

The ATPase activity based high throughput small-molecule screen was performed at ICCB Longwood Screen Facility at Harvard Medical School. On the day of screening, Combi machine was used to dispense 10 μl aliquots of a master mix containing 600 nM of 30-mer single-stranded DNA substrate and 10 nM of POLθ protein (ΔPol) into each well of a 384 well plate (Corning 3820). Small molecules were added to wells in the 384 well plates by Seiko Pin transfer robot. Plates were covered with aluminum seals and spun down at 1000 rpm and sat for 1 hour to allow small molecules to bind protein. 3 μL of 433.33 μM ultra-pure ATP was added to every well except no-substrate control wells. Plates were stored overnight (16 hours) in a humidified plastic chamber. Next day, 6.5 μL of ADP-Glo reagent (Promega) was added to every well using the Combi machine. After one-hour incubation, 13 μL of Kinase Detection Reagent (Promega) was added to every well and incubated for another hour. Data was collected using the Envision plate reader by measuring luminescence signal. More information about the screen is available in Supplementary table 1.

Statistical Analysis of the high throughput screen

The inhibition strength of screened compounds was evaluated using Z-score analysis. Since most compounds did not affect POLθ-ATPase activity, data for all compounds was used to calculate plate mean (μ) and standard deviation (σ). Every compound is then assigned a Z-score using the equation z = (x – μ) / σ. Percentage inhibition analysis. Percentage inhibition of compounds was determined by normalizing to the DMSO control. Wells with DMSO were averaged to find the average signal without inhibitor, and a different DMSO average was calculated for each plate. The % activity was calculate using the following formula: % activity = [(signal with compound) / (average signal with DMSO)] x 100%.

ATPase activity assay

The ATPase activity of POLθ was measured using the ADP-Glo kinase assay (Promega). Reactions were done in 50 μl of reaction buffer (40 mM Tris-HCl buffer (pH 7.6), 20 mM MgCl₂, 0.1 mg/ml BSA, and 1 mM DTT) in black 96 well plates. The final concentrations were 600 nM of 30mer single-stranded DNA, 50 nM of purified POLθ-ΔPol protein, and 100 μM of ATP, with DMSO or NVB. The sequence of the 30mer ssDNA used in the assay was: 5’-CCAGTGAATTGTTGCTCGGTACCTGCTAAC-3’. The order for adding the components were: 10 μl 1x reaction buffer, 10 μl of 10x ssDNA, 10 μl of 10x POLθ, 10 μl of 10x NVB or DMSO (incubate at room temp for 15 min), and finally 10 μl of 10x ATP, and all components were prepared in 1x reaction buffer. DMSO wells represented 0% inhibition while no-enzyme wells represented 100% inhibition. Plates were covered with an aluminum seal and incubated at room temperature overnight (16 hours). The following day, 50 μl of ADP-Glo reagent (Promega kit) was added to each reaction well and incubated at room temp for one hour. Next, 100 μl of Kinase detection reagent (Promega kit) was added to the wells, and plates were incubated for another hour. Finally, ATP hydrolysis was quantified by luminescence measured on a plate reader. Measurement of other ATPases were done using the same conditions and protocol. In some cases as indicated, ³²P-based ATPase activity assay was also performed as previously described ⁵⁴.

Thermal Stability Assays (TSA)

POLθ helicase domain was expressed using POLQ-A construct (Addgene#74645) in HF insect cells and purified as described previously ³⁰. Thermal stability of POLθ helicase domain in the presence of NVB was measured using Prometheus with NanoDSF (NanoTemper). Given that the intrinsic fluorescence emission from the protein coincide with the chromophore core in the NVB, we monitored scattering to measure the protein stability with temperature. POLθ helicase domain (0.5 mg/ml in 25 mM HEPES (pH=8) and 200 mM NaCl) was incubated with indicated concentration of NVB for 30 minutes before the sample was loaded into capillaries for the scattering measurement with increasing temperature at 1 °C/minute starting from 25 °C till 90 °C. Shift in the peak max of scattering was estimated from the first derivative of the scattering data. All data points were an average from three replicates and plotted in GraphPad Prism8.

Preparation of Novobiocin–Sepharose 6B Beads

NVB–Sepharose 6B Beads were prepared similarly as previously described ⁵⁵. 0.5 gram of epoxy-activated Sepharose 6B (Sigma-Aldrich E6754) was washed thoroughly and swollen in 50 mL of distilled water for 1 hour at room temperature. The resin was washed once with coupling buffer (0.3 M Sodium Carbonate [pH 9.5]). The washed resin was mixed with 500 mg of NVB in 50 mL of coupling buffer and incubated at 37 °C with rotation overnight. Next day, the resin was washed with coupling buffer three times to remove excess NVB. The remaining epoxy-active groups were blocked with 1 M ethanolamine (in coupling buffer) for 6 hours at 37 °C with gentle rotation. The beads were sequentially washed with coupling buffer, 0.5 M NaCl in coupling buffer, distilled water, 0.5 M NaCl in 0.1 M sodium acetate (pH 4), and twice in distilled water. The beads were re-suspended and stored in 25 mM HEPES (pH 7.6), 200 mM KCl, 10% ethylene glycol, and 1 mM EDTA.

GFP reporter-based DNA repair assays

U2OS cells containing the DR-GFP (for HR) and EJ-2 (for MMEJ) repair substrates were gifts from Dr. Jeremy Stark at Beckman Research Institute of the City of Hope. To measure the repair efficiency, 1×10⁵ cells were plated in each well of a 12-well plate. Indicated compounds or DMSO were added to the medium for 24 hours, and cells were infected with Adenoviruses expressing the I-SceI enzyme. Two days after infection, cells were trypsinized and GFP-positive cells were quantified by flow cytometry (Beckman).

RAD51 and γH2AX focus formation assay

Cells were pre-treated with DMSO or NVB for 24 hours before receiving 5 Gy of irradiation. 4–8 hours after irradiation, cells were extracted and fixed with 1% PFA, 0.5% methanol, and 0.5% TritonX-100 in PBS for 20 min at room temperature on a shaker. Fixed cells were washed with PBS twice and blocked with BTG buffer (1 mg/mL BSA, 0.5% TritonX-100, and 3% of goat serum, 1 mM EDTA) for 1 hour. Cells were then incubated with a rabbit Rad51 antibody (Santa Cruz Technology sc-8349 or Cell Signaling Technology #8875) and mouse γH2AX antibody (Millipore #05–636) in BTG buffer. After wash with PBS three time, cells were incubated with Alexa-488 (Life Technologies A11034) or Alexa-594 (Life Technologies A11005) conjugated secondary antibodies in BTG buffer for 1 hour. The dishes were washed with BTG buffer once and PBS twice, and mounted with mounting solution with DAPI for immunofluorescence microscopy. Images of random fields were taken under a 63x oil lens. Foci were visualized and quantified using ImageJ software. Images were prepared in the Zen software v3.1 (Zeiss, blue edition).

CellTiter-Glo (CTG) cell viability assay

The CellTiter-Glo luminescent cell viability assay kit (Promega #G7573) was used to measure cell viability after drug treatment. The CellTiter-Glo kit determines the number of viable cells in culture based on quantitation of the ATP present, an indicator of metabolically active cells. 500 to 1000 cells per well were seeded in a 96-well plate in 100 μL media volume. 24 hours later, cells were treated with indicated drugs by adding 100 μL medium containing 2x concentrated drug of desired final concentration. Cells were cultured in drug containing medium for 6 days, by then cell viability was determined by measuring the luminescent signal, following the manufacturer’s instructions. Survival fraction of drug-treated cells was normalized to DMSO-treated control. Survival, IC50 and statistics were determined by using the GraphPad Prism software.

Clonogenic survival assay

Cells were seeded, at 2 different concentrations, into each well of a 6-well plate (100 and 200 cells for WT RPE1; 500 and 1000 cells for HR-deficient cells). Next day, cells were treated with the indicated doses of inhibitors. For combination studies, cells were treated first with Novobiocin or DMSO for 24h, washout and then incubated in fresh media containing Novobiocin or DMSO with increasing concentration of PARPi. Cells were allowed to grow in drug-containing medium for 12 to 14 days. Cells were fixed and stained with 0.2% crystal violet in methanol for 30 min (Crystal Violet solution, Sigma HT90132), and rinsed with distilled water three times. The stained dishes were air-dried, and the number of colonies (>50 cells) was counted in each well. Surviving fraction was plotted for each drug concentration by normalizing them to the untreated cells.

Gene knockouts and cell line generation

BRCA1, BRCA2 and POLQ knockouts were generated in RPE-1 cells in which TP53 has been first knocked out (RPE1- P53^−/−). BRCA1 knockout were previously generated in RPE1-TP53^−/− (named here RPE.1a) in Alan d’Andrea’s laboratory (DFCI) ⁵⁶. BRCA2 and POLQ knockouts were generated in Raphael Ceccaldi’s laboratory (Institut Curie) using another clone of RPE1-TP53^−/− (named here RPE.1b). Accordingly, we named the following knockouts: RPE1a BRCA1^−/− and RPE1b POLQ^−/−, RPE1b BRCA2^-/−. Knockouts were generated by co-transfection of Cas9 (pSpCas9(BB)-2A-GFP Addgene #48138) (PX458) and sgRNA vectors (gRNA cloning vector Plasmid Addgene #41824) by Lipofectamine LTX with Plus Reagent (Thermo Fisher). After subcloning into 96 wells plate, single colonies were screened by PCR. Briefly, DNA was extracted by adding 30 μl of a 50 mM NaOH solution per well after which the plate was put at 95°C for 10 minutes. Then, 2.5 μl of 1M Tris-HCL PH 7.5 was added per well and 2 μl of the extracted DNA was used to perform screening PCR (Terra PCR Direct Polymerase, Takara Clontech). Clones were selected for the presence of deletion PCR band and absence of 5’ and 3’ junction PCR band. RPE1b cells were obtained using the TP53 sgRNA sequence: GGCAGCTACGGTTTCCGTC. RPE1b BRCA2^−/− cells and RPE1b POLQ^−/− were obtained using two sgRNA vectors targeting introns flanking exon 2 of the BRCA2 gene: BRCA2–2: GGTAAAACTCAGAAGCGC; BRCA2–3: GCAACACTGTGACGTACT or the POLQ gene: POLQ-2: GGAAGGCTTTTAGGTCAGTA; POLQ-3: GGCAACGGGGGCAGCTCCGC. PCR primer sequences used for genotyping BRCA2 knockout were the following: deletion PCR (seqBRCA2-For: GCTGTATTCCGAAGACATGCTGATGG; seqBRCA2-Rev: TTGTTCCTACTGCTAGTCAAGGG), 5’ Junction PCR (seqBRCA2-For + exon2-Rev: TACCTACGATATTCCTCCAATGCTT), 3’ Junction PCR (seqBRCA2-Rev + exon2-For: AAGCATTGGAGGAATATCGTAGGTA). For genotyping POLQ knockout we used: deletion PCR (seqPOLQ-For: GAAGAGCTACTTCCCTGATCTACC; seqPOLQ-Rev: CCCAACCAATCTCATTACGAATCC), 5’ Junction PCR (seqPOLQ-For + exon2-Rev: GAGCCTGATTCTGAACGCCGCCG), 3’ Junction PCR (seqBRCA2-Rev + exon2-For: CGGCGGCGTTCAGAATCAGGCTC).

POLQ knockout cells were generated as well in U2OS cells. Cas9 was introduced into U2OS cells by lentiviral infection using the lentiCas9-Blast plasmid (Addgene #52962), and stable Cas9-expressing U2OS cells (U2OS-Cas9) were obtained after blasticidin selection. SgRNA oligos designed to target the genomic sequence GATTCGTTCTCGGGAAGCGG of POLQ (exon 1) were annealed and inserted into LentiGuide-Puro plasmid (a gift from Feng Zhang, Addgene #52963). After lentivirus infection, U2OS-Cas9 cells were selected by puromycin for 3 days, and the survived cells were trypsinized and seeded sparsely to form single-cell colonies. Colonies were picked and expanded, and genomic DNA from individual clones was isolated by QIAamp DNA Mini Kit (Qiagen). PCR was performed to amplify the gRNA-targeted region (216 bp), using the following primers: Forward, GGAGGACGCTGGGACTGTGGC; Reverse, CTGCAGCTGCGGCCTTCAGGC. PCR products were cleaned by PCR purification kit (Qiagen) and submitted for Next-Gen Sequencing (Amplicon sequencing provided by Genewiz). The sequencing reads were mapped to the human genome at POLQ locus and the results were visualized by Integrative Genomics Viewer (IGV, Broad Institute). Clones with homologous deletion of POLQ gene were confirmed by Western blot analysis. Western blots were collected using Odyssey Infrared Imagers (Li-Cor), and analyzed using Image Studio Lite.

Cloning, PiggyBac constructions and cell line generation

Wild type (WT) full length POLQ cDNA was assembled by PCR with a FLAG-P2A-BLAST cassette and subsequently cloned by Gibson assembly (New England BioLabs, #E2611L) in the PiggyBac PBCAG-eGFP vector (Addgene #40973) using BsrGI and SbfI sites to obtain the PiggyBac-eGFP-POLQ-FLAG-P2A-BLAST vector (referred here as GFP-Polθ WT). Point mutations were introduced using PCR mutagenesis and Gibson assembly in active sites as follow: Helicase (K121A, DE-216/7-AA), polymerase (DE-2540/1-AA), helicase/polymerase (K121A, DE-216/7-AA, DE-2540/1-AA). RPE1b POLQ^−/− cells were complemented with GFP-Polθ WT and mutant constructs. Cells were plated in a 6-well plate, transfected the day after with 1 μg of PiggyBac vector and 0.5 μg of PiggyBac transposase (System Biosciences, #PB210PA-1) using Lipofectamine LTX with Plus Reagent (Thermo Fisher) and selected with Blasticidin (21 μg/ml) for 10 days. Vectors expression was confirmed by Western Blot.

Laser micro-irradiation

RPE1b POLQ^−/− cells expressing eGFP-POLθ constructs were plated in a 35 mm μ-Dish with glass bottom (Ibidi, #81158). The day after, DMSO (0.1%) or Novobiocin (NVB, 100 μM) were added to the cells for 16 hours. Media was replaced by fresh one (with DMSO or NVB) 2 hours prior the irradiation. Micro-irradiations were performed with a two photons laser (800 nm, 20% power) on an inverted Laser Scanning Confocal Microscope equipped with Spectral Detection and Multi-photon Laser (LSM880NLO/Mai Tai Laser - Zeiss/Spectra Physics) with Airyscan module. Images were analyzed using “Plot Z-Axis Profile” macro on Fiji (version 2.1.0/1.53c). RPE1b POLQ^−/− cells expressing PBCAG-eGFP vector were used as a control.

Mouse Xenograft and PDX studies

All animal experiments were conducted in accordance with Institutional Animal Care and Use Committee-approved protocols at Beth Israel Deaconess Medical Center and Dana-Farber Cancer Institute. For PDX studies, tumor ascites were collected from patients with suspected or established ovarian cancer at the Brigham and Women’s Hospital or the Dana-Farber Cancer Institute (DFCI) under IRB-approved protocols conducted in accordance with the Declaration of Helsinki and the Belmont Report. Written informed consent was obtained from patients when required.

Quantitative RT-PCR

Total RNA was extracted using the RNeasy Mini kit (Qiagen), and cDNA was generated using the SuperScript™ III First-Strand Synthesis kit (Thermo Fisher). The PowerUp SYBR Green Master Mix (Thermo Fisher) reagent was used to run quantitative PCR on a QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). Two set of POLQ primers were used which gave same POLQ mRNA measurements. POLQ primer set 1 (forward: 5-TATCTGCTGGAACTTTTGCTGA-3; reverse: 5-CTCACACCATTTCTTTGATGGA-3); POLQ primer set 2 (forward: 5-CTACAAGTGAAGGGAGATGAGG-3; reverse: 5-TCAGAGGGTTTCACCAATCC-3). Internal control primers are beta-Actin (forward: 5-CACCATTGGCAATGAGCGGTTC-3; reverse: 5-AGGTCTTTGCGGATGTCCACGT-3) or GAPDH (forward: 5-GTCTCCTCTGACTTCAACAGCG-3; reverse: 5-ACCACCCTGTTGCTGTAGCCAA-3).

Immunohistochemistry (IHC)

NSG mice bearing DF-59 and DF-83 tumors were dosed as in the efficacy study, with 75 mg/kg NVB twice a day and 50 mg/kg Olaparib daily. 4 hours after last dose (52 hours after first dose), tumor cells were isolated from the peritoneum of the mice, cleaned, washed with PBS and fixed in 10% normal buffered formalin for 10 min at RT. Next, the tumor cells were washed with PBS 3X and embedded in Histogel (Richard Alen Scientific). Histogel cores were processed using standard histology methods and embedded in paraffin. RAD51 staining was performed on paraffin sections (4 μm) using a previously standardized protocol (https://doi.org/10.1158/1538–7445.AM2017–2796). Images of stained slides were acquired on an Olympus BX41 microscope equipped with a digital camera at 40X magnification. The percentage of RAD51-foci positive cells fields was estimated.

Molecular Docking

We used the AMP-PNP-Mg²⁺ bound POLθ-HD (PDB ID-5AGA) structure for our docking studies as it has the best resolution (2.9 Å) among the reported structures in the protein data bank (PDB) ³⁰. For the docking studies, the protein structure is prepared using Protein Preparation Wizard from the Schrödinger suite. The AMP-PNP-Mg²⁺ was stripped from the structure as well. Using the minimized structure, a 25 Å grid created around the Asn451 residue given its close proximity to the ADP/AMP-PNP binding site in the helicase domain for the search of a potential binding site for Novobiocin. The LigPrep Wizard from the Schrödinger suite is used to prepare novobiocin ligand in different protonated states for the docking. All minimizations in the Protein Prep and LigPrep wizards are performed with OPLS3 force field ⁵⁸. Glide Extra Precision docking module of the Schrödinger workflow is used for producing novobiocin and POLθ-HD complex structure with the best scoring poses ⁵⁹. For all the generated poses, binding free energy MM-GBSA calculations were performed using the Prime module and the best model selected based on the resulting protein-ligand complex with a lowest MM-GBSA binding free energy value ⁶⁰.

Statistics and Reproducibility

No statistical methods were used to predetermine sample sizes. The experiments were not randomized except for mouse studies, where mice were randomized based on their initial tumor sizes to achieve an approximately equal mean. No data were excluded from the analyses. The person who performed the mouse xenograft studies was blinded to the identities of the tumor cells. The person who performed the PDX studies was blinded to the genotypes of the tumors being studied. For other studies, no specific blinding method was used. The statistical analyses performed were specified in the figure legends. Statistical analyses were performed in Prism version 8 (GraphPad). Error bars represent ±SD or ±SEM as described in figure legends, and p values are indicated by *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, unless values are given.

Data Availability

All related data have been presented in this manuscript. Source data for each Figure and Extended Data Figure have been provided as Source Data files. Complete results of small molecule screen and lists of top hits have been provided in Supplementary Tables 1–4.

Code Availability

No code was generated or used for this study.