Molecular disruption of DNA polymerase β for platinum sensitisation and synthetic lethality in epithelial ovarian cancers

Targeting PARP1 [Poly(ADP-Ribose) Polymerase 1] for synthetic lethality is a new strategy for BRCA germ-line mutated or platinum sensitive ovarian cancers. However, not all patients respond due to intrinsic or acquired resistance to PARP1 inhibitor. Development of alternative synthetic lethality approaches is a high priority. DNA polymerase β (Polβ), a critical player in base excision repair (BER), interacts with PARP1 during DNA repair. Here we show that polβ deficiency is a predictor of platinum sensitivity in human ovarian tumours. Polβ depletion not only increased platinum sensitivity but also reduced invasion, migration and impaired EMT (epithelial to mesenchymal transition) of ovarian cancer cells. Polβ small molecular inhibitors (Pamoic acid and NSC666719) were selectively toxic to BRCA2 deficient cells and associated with double-strand breaks (DSB) accumulation, cell cycle arrest and increased apoptosis. Interestingly, PARG [Poly(ADP-Ribose) Glycohydrolase] inhibitor (PDD00017273) [but not PARP1 inhibitor (Olaparib)] was synthetically lethal in polβ deficient cells. Selective toxicity to PDD00017273 was associated with poly (ADP-ribose) accumulation, reduced nicotinamide adenine dinucleotide (NAD+) level, DSB accumulation, cell cycle arrest and increased apoptosis. In human tumours, polβ-PARG co-expression adversely impacted survival in patients. Our data provide evidence that polβ targeting is a novel strategy and warrants further pharmaceutical development in epithelial ovarian cancers.


Introduction
In BRCA germ-line deficient and platinum sensitive sporadic epithelial ovarian cancers, PARP inhibitor (Niraparib, Olaparib, Rucaparib) maintenance therapy improves progression-free survival (PFS) [1][2][3]. However, not all patients respond [4] either due to intrinsic resistance or acquired resistance to PARP inhibitors. Therefore, the development of alternative DNA repair targets and synthetic lethality approaches is required.
Base excision repair (BER) is essential for the removal of bases damaged by alkylation, oxidation or deamination [5,6]. BER is performed by at least two major sub-pathways: the short-patch pathway (SP-BER) and long-patch pathway (LP-BER) [7]. Both pathways are initiated by a damage-specific DNA glycosylase, which removes the damaged base creating an abasic site. APE1 then cleaves the phosphodiester bond 5′ to the AP site thereby generating a nick with 5′-sugar phosphate (dRP) and 3′hydroxyl group. DNA polymerase β (polβ) adds the first nucleotide to the 3′-end of the incised AP site. Normally, the reaction continues through the short-patch repair pathway where polβ removes the 5′-sugar phosphate residue by the process of β-elimination [8] and DNA ligase III-XRCC1 heterodimer (or DNA ligase I) then completes the repair [9,10]. Polβ-mediated lyase activity is the ratelimiting step in SP-BER. The processing of oxidised AP sites generates a 5′ residue that is resistant to β-elimination (mediated by polβ) and therefore requires additional DNA synthesis via LP-BER. A role for the polymerase activity of Pol β under the coordination of the Rad9-Rad1-Hus1 sliding clamp complex (9-1-1 complex) in LP-BER has also been described [10].
Polβ is a key player in BER [11]. Polβ interacts with several components of the BER machinery including XRCC1, ligase III and PARP1 to accomplish its biochemical functions [11]. In the current study, we hypothesised that Polβ could promising target in ovarian cancers.

Polβ expression links to aggressive epithelial ovarian cancers
A total of 379 tumours were suitable for analysis of nuclear expression of polβ. 217/379 (57.3%) tumours were low for polβ and 132/379 (42.7%) of tumours were high in polβ expression (Fig. 1A). Low polβ expression was more common in mucinous cystadenocarcinoma, endometrioid and clear cell carcinomas (p = 0.001). Whereas, high polβ expression was seen in sub-optimally debulked (p = 0.023) (Supplementary Table 1). On the other hand, FIGO stage I disease were more common in tumours with low polβ expression (p = 0.003). Platinum resistance was more common in tumours with high polβ expression although it did not reach significance (p = 0.076). High polβ expression was associated with poor PFS (p = 0.020) (Fig. 1B) and poor overall survival (OS) (p = 0.029) ( Supplementary Fig. S1A). At the transcriptional level, similarly, high polβ mRNA expression was associated with poor PFS and OS in both the test and validation cohort 1 (all p values < 0.05) ( Supplementary  Fig. S1). In the cancer genome atlas (TCGA) cohort (Validation cohort 2), high polβ mRNA expression was associated with poor PFS (Fig. 1C). Clinical data suggest that high polβ is a marker of adverse phenotype. We proceeded to pre-clinical studies.
Polβ localises to the nucleus after platinum treatment in ovarian cancer cells We chose A2780 (platinum sensitive) and A2780cis (platinum resistant) ovarian cancer cell lines for initial preclinical studies [12] (Fig. 1D). To evaluate for alterations in sub-cellular localisation of polβ upon cisplatin treatment, we generated nuclear and cytoplasmic extracts at baseline and following 48 h cisplatin therapy. We observed a significant accumulation of polβ protein in the nucleus in A2780cis cells compared A2780 cells (Fig. 1E) and (Supplementary Fig. S2A). There was no alteration in cytoplasmic level of polβ protein in A2780 or A2780cis cells (Fig. 1E) and (Supplementary Fig. S2A). For further validation, we monitored polβ sub-cellular localisation upon cisplatin treatment using immunofluorescence assay at 24 and 48 h. As shown in Supplementary Fig. S2B, C, we observed substantial nuclear accumulation of polβ after 48 h of cisplatin treatment in A2780cis cells compared to A2780 cells. The data suggest that alterations in sub-cellular localisation of polβ after platinum therapy may influence sensitivity.

Bioinformatics analyses of the polβ interactome
As polβ mutants [13] have been reported previously to influence biochemical function and platinum sensitivity, we performed next generation exome sequencing and used the variant effect predictor tool [14] to identify genetic variants in platinum sensitive (A2780, PEO1) and platinum resistant (A2780cis, PEO4) cells (manuscript in preparation). We did not observe any polβ variants in A2780, A2780cis, PEO1 and PEO4 cells. While variants affecting polβ functionally associated genes were not statistically enriched in platinum-resistant cell lines, given the crucial role of polβ as a mediator of platinum resistance, we next examined loci encoding polβ-interacting proteins, as defined in the BioGRID database [15] (Supplementary Fig.  S3A). Among these polβ interacting proteins, we identified coding variants affecting LIG3, XRCC3, WRN; synonymous variants affecting EP300, UNG, XPC and noncoding variants affecting the APEX1, APTX, BCKDHA, BCKDHB, HUS1, KTN1, NEIL1, RAD9A, SRPK2, SRPK2, STUB1, TDP1, TLE1, TPP2, XPC, CRBN, TAF1D loci in platinum resistant A2780cis and PEO4 cells compared to platinum sensitive A2780 and PEO1 cells (Supplemental Table 2). These include novel variants in EP300, HUS1, KTN1, UNG, WRN, XRCC3 and XPC. Taken together, the data suggest that polβ and variants affecting the polβ functional interactome may contribute to platinum resistance either directly or indirectly through interactions with other factors involved in processing platinum induced DNA damage.

Polβ depletion increases platinum sensitivity
BER may operate predominantly during G1 phase of the cell cycle [10]. Moreover, a BER complex (consisting of polβ, APE1, UNG2 and XRCC1) was shown to physically associate with MCM7 suggesting that BER may also operate at sites of base damage and single-strand breaks occurring at replication forks [16]. In addition, polβ activity may also be prominent during the S-phase of the cell cycle [17]. We therefore generated transient knockdowns (KD) of polβ in A2780cis cells using siRNAs (Fig. 1F, Supplementary Fig. S4A) and evaluated cell cycle progression in control and polβ _KD cells. Compared to control cells, we observed significant S-phase arrest in polβ_KD cells (Fig. 1F) which was associated with accumulation of ATR and pChk1 ser 345 (Fig. 1G, H) and ( Supplementary Fig. S4B, C), a feature that would be consistent with replication stress in A2780cis polβ_KD cells. In clonogenic assays, Polβ_KD_A2780cis cells (transient KD generated using siRNA) were strikingly sensitive to platinum therapy (Fig. 1I). We also confirmed this observation using another siRNA construct which also showed robust Polβ_KD and also lead to platinum sensitisation ( Supplementary Fig.  S4D, E). Polβ_KD was associated with double-strand break increased platinum sensitivity. We confirmed this result using two siRNA constructs which showed robust polβ knock down and similarly lead to platinum sensitisation ( Supplementary Fig. S4H, I). Polβ depletion increased γH2AX nuclear foci accumulation (Fig. 2K, L) and ( Supplementary Fig. S5A), 53BP1 nuclear foci accumulation (Fig. 2M), S-phase arrest (Fig. 2N) and apoptotic cells (Fig. 2O).
We have tested another Isogeneic ovarian cancer cell line model. PEO1, which is BRCA2 deficient, is a platinum sensitive ovarian cancer cell line. We have recently We then generated A2780cis polβ-knockout (KO) cells using CRISPR-Cas-9 methodology (Fig. 3A) and (Supplementary Fig. S5C). As shown in Fig. 3B, polβ_KO strikingly increased sensitivity to platinum which was associated increased nuclear γH2AX nuclear foci accumulation  Fig. S5D), 53BP1 foci accumulation (Fig. 3E), S-phase arrest (Fig. 3F) and accumulation of apoptotic cells (Fig. 3G). As platinum treatment can induce the generation of free-radicals in cells and cause oxidative DNA base damage, we also explored if the sensitivity to cisplatin could be altered upon antioxidant treatment. To address this hypothesis, we pre-treated cells with curcumin (antioxidant) for 24 h followed by cisplatin treatment. Polβ deficient cells remained sensitive to platinum therapy ( Supplementary Fig. S6A). We then tested another DNA cross-linker Mitomycin C. We again observed that polβ KO cells were also sensitive to mitomycin C treatment (Supplementary Fig. S6B). Together, the data provide evidence that polβ is a predictor of platinum sensitivity. A role for Polβ in the regulation of invasion, migration and EMT There is also emerging evidence for the role of epithelial to mesenchymal transition (EMT) in ovarian cancer chemotherapy resistance [18]. To evaluate if polβ can also influence the aggressive behaviour of ovarian cancer we proceeded to investigate invasion, migration and EMT in control and polβ_KO ovarian cancer cell lines. Polβ_KO in A2780 and in A2780cis cells significantly reduced invasion ( Supplementary Fig. S7A). Given the well-defined roles of E-cadherin [19], N-cadherin [20], TGFβ [21] and MMP-9 [22] in invasion and EMT regulation, we evaluated if polβ could influence the expression of these key EMT markers in ovarian cancer cells. As shown in Supplementary To provide additional evidence we performed real-time PCR using RT 2 Profiler EMT PCR Array for 86 EMT genes. Full list of EMT genes included in the PCR array is summarised in Supplementary Table S3. Compared to control cells, downregulation of several genes with defined roles in EMT was evident in polβ_KO cells ( Supplementary Fig. S8A, B). Interestingly, we also observed upregulation of some genes with roles in EMT ( Supplementary Fig. S8A, B). Together the data would imply a complex role for polβ is influencing the expression of genes involved in EMT.
Polβ blockade is synthetically lethal in BRCA2 deficient cells BRCA2 is a key player in homologous recombination (HR). Patients with germ-line mutations in BRCA2 are predisposed to ovarian cancer development with a cumulative life time risk of about 20-30% [23]. PARP inhibitor maintenance therapy improves PFS in BRCA2 germ-line deficient ovarian cancers [1][2][3]. However, not all patients respond either due to intrinsic resistance or acquired resistance to PARP inhibitors [4]. We hypothesised that polβ could be a promising alternative synthetic lethality target in BRCA2 deficient ovarian cancers. Polβ deficiency impairs BER and leads to accumulation of single-strand breaks, which if unrepaired, result in generation of DSBs during replication. In cells deficient in HR repair (HRR), DSBs would persist and lead to synthetic lethality. We therefore tested synthetic lethality in BRCA2-deficient (PEO1) and BRCA2-proficient (PEO4) ovarian cancer cells. PEO1 and PEO4 cells have robust Polβ expression (Fig. 3H) and ( Supplementary Fig. S9A). Cell viability, as investigated by clonogenic assay, was significantly impaired when polβ was depleted in PEO1 cells, but not in PEO4 cells (Fig. 3I). Polβ depletion in PEO1 cells resulted in increased γH2AX foci accumulation (Fig. 3J-L) 53BP1 foci accumulation (Fig. 3M), S-phase cell cycle arrest (Fig. 3N) and induction of apoptosis (Fig. 3O). We then tested cytotoxicity of polβ inhibitors (polβi) in BRCA2-proficient and deficient cell lines. Pamoic acid is a well described polβi [24]. The cytotoxicity of Pamoic acid was first tested in control and polβ_KO cells. We did not observe any significant cytotoxicity in polβ_KO cells compared to control cells (Supplementary Fig. S9B). The data suggest that Pamoic acid may have specific activity against polβ. Compared to PEO4 cells, PEO1 cells were sensitive to Pamoic acid treatment (Fig. 4A) which was associated with DSB accumulation (Fig. 4B), S-phase arrest (Fig. 4C) and apoptosis (Fig. 4D). We then validated Pamoic acid activity in BRCA2 deficient and control HeLa cells (Fig. 4E). As shown in Fig. 4F, Pamoic acid was selectively toxic in HeLa BRCA2-deficient cells compared to control HeLa cells. Increased sensitivity was associated with DSB accumulation (Fig. 4G), S-phase arrest (Fig. 4H) and increased apoptotic cells (Fig. 4I).

PARG inhibitor (PARGi) is selectively toxic in polβ deficient cells
At sites of DNA damage, PARP1 is recruited where it induces the synthesis of poly(ADP-ribose) (PAR). PARylation of PARP1 and other DNA repair factors is essential for coordination of DNA repair [28]. PARylation is transient and reversible process. PAR glycohydrolase (PARG) is a key factor in the PAR degradation pathway [29][30][31] A coordinated activity of PARP and PARG is also essential for efficient DNA repair. Recently, PARG has also emerged as a promising drug target in cancer [32,33]. Importantly, a recent study by Pillai et al. demonstrated that ovarian cancer cells respond differently to PARGi (PDD00017273) compared to PARPi (Olaparib) [34]. DNA replication vulnerabilities was shown to particularly render ovarian cancer cells sensitive to PDD00017273 treatment leading to persistent replication fork stalling and replication catastrophe in that study [34]. As polβ-deficient ovarian cancer cells are not only sensitive to platinum treatment but also show features of replication stress as evidenced by accumulation of ATR and pChk1 ser 345 , we tested a synthetic lethality application for either PARPi or PARGi. We first evaluated cellular activity of Olaparib and PDD00017273 in polβ-deficient and -proficient ovarian cancer cells. Olaparib did not induce selective cytotoxicity in polβ_KO or control cells (Supplementary Fig. S9F, G). However, as shown in Fig. 5A, B, PDD00017273 treatment was selectively toxic in polβ_KO A2780 or polβ_KO A2780cis cells compared to controls. To also validate this observation in PEO4 ovarian cancer cell line, we generated transient knockdown of polβ and again observed increased sensitivity to PDD00017273 treatment in PEO4 cells ( Supplementary  Fig. S9H). We proceeded to functional studies to understand potential mechanisms of PDD00017273 toxicity.

Mechanistic studies of PARGi in polβ-deficient cells
At baseline, there were no changes in PARP1 levels in both control and polβ_KO cells (Fig. 5C, Supplementary  Figs. S9I, S10A). However, in polβ_KO cells there was a reduction in PAR level compared to control cells (Fig. 5D) and ( Supplementary Fig. S11B, C). Using PARG ELISA assay we show that polβ_KO cells have increased PARG activity compared to control cells (Fig. 5E). We then evaluated PAR level in polβ_KO and controls treated with either Olaparib or PDD00017273 (Fig. 5F) and (Supplementary Fig. S10D, E). In Olaparib-treated cells, as expected, there was a reduction in PAR levels in both control and polβ_KO cells which also confirmed in a PAR ELISA assay (Supplementary Fig. S10G). PDD00017273 treatment increased PAR level in control and polβ_KO cells (Fig. 5F) and (Supplementary Fig. S10D-F). To further validate, we monitored PAR level by confocal microscopy at 1, 6, and 24 h following PDD00017273 treatment in A2780_ polβ_KO and control cells (Fig. 5G-I). In A2780 control cells, following PDD00017273 treatment, there was transient increase in PAR accumulation at 6 h which returned to baseline levels at 24 h compared to untreated cells. However, in A2780_ polβ_KO cells, there was accumulation of PAR at 24 h compared to untreated cells (Fig. 5G-I). PARP1 catalyses the polymerisation of ADP-ribose units (PAR) from NAD+. On the other hand, PARG catalyses the hydrolysis of PAR producing free mono-and oligo(ADP-ribose) [28]. We evaluated NAD+ levels in untreated as well as PDD00017273-treated control and polβ_KO cells (Fig. 5J). In untreated A2780_ polβ_KO cells, there was a significant increase in NAD+ levels compared to control cells. Upon PDD00017273 treatment there was a significant reduction in NAD+ levels in A2780_ polβ_KO cells compared to controls. Interestingly in untreated A2780cis_ polβ_KO cells, there was no increase in NAD+ levels compared to controls. However following PDD00017273 treatment NAD+ levels significantly reduced compared to controls (Fig. 5J). Together, the data suggest a complex cell line dependent network operating between polβ, PAR and NAD+. Previous studies have shown that PAR accumulation can not only impair efficient DNA repair but can also be directly toxic to cells [35,36]. Depletion of NAD+ is known to alter NAD+/SIRT1 axis which also leads to impaired DNA repair and cell survival [37,38]. PDD00017273 treatment in A2780_polβ_KO cells promoted γH2AX nuclear foci accumulation (Fig. 5K, L) and ( Supplementary Fig. S11G), an indirect biomarker for DSBs and replication stress-induced defects. As a further validation of DSBs, we also observed 53BP1 nuclear foci accumulation in PDD00017273-treated A2780_polβ_KO cells (Fig. 5K, M). Consistent with PDD00017273 induced replication arrest, CHK1 was phosphorylated on serine 345 in PARGi-treated polβ deficient cells (Fig. 5N, Supplementary Fig. S10H), which was then associated with S-phase arrest (Fig. 6A) and induction of apoptosis (Fig. 6B). Similarly, in A2780cis_ polβ_KO cells compared to controls, PDD00017273 treatment was associated with γH2AX nuclear foci accumulation (Fig. 6C, D), 53BP1 nuclear foci accumulation (Fig. 6E), increased phosphorylated CHK1 (Fig. 5M, Supplementary Fig.  S10H), S-phase arrest (Fig. 6F) and apoptosis (Fig. 6G). In 3-D spheroid models, PDD00017273 treatment reduced spheroid size and promoted cell death in polβ_KO spheroids compared to control spheroids (Fig. 6H-J). Olaparib did not influence cell viability in 3-D spheroid assays (Fig. 6H-J). Together the data suggest that PARGi in polβ deficient induces selective cytotoxicity through replication stress, DSB accumulation and cell death. Talazoparib, has at least 100 times more potency to trap PARP at replication forks compared to Olaparib [28]. As expected, A2780_polβ_KO and A2780cis_ polβ_KO cells were sensitive to Talazoparib therapy compared to controls ( Supplementary Fig. S11A, B). The data would suggest that replication stress induction contributes to the observed cytotoxicity.

Discussion
Polβ, a member of the X-family of DNA polymerases is a key player in BER [10]. Pre-clinically we demonstrated that polβ depletion increased platinum sensitivity in platinum resistant A2780cis, PEO4 and PEO1R ovarian cancer cells. In platinum sensitive A2780 cells, although there were some variations due to different experimental conditions, Polβ depletion using multiple siRNA constructs consistently increased platinum sensitivity. Interestingly, in cisplatin untreated A2780cis cells polβ deficiency caused S-phase arrest and a replication arrest phenotype. However, we did not observe this phenomenon in A2780 or PEO4 cells, implying a cell line dependent role for polβ during replication in ovarian cancer cells.
Higher levels of polβ have been described in colonic and prostate compared with normal tissue [39,40]. Mutations in polβ may influence aggressive solid tumour phenotype and response to chemotherapy [13]. In ovarian cancers, mutations in polβ have been recognised [41] although clinical significance remains unclear [41]. In clinical cohorts of ovarian cancers, we show that low Polβ expression is associated with better PFS. In the current study, we did not observe any cytoplasmic staining for polβ protein in ovarian tumours. For nuclear expression, we have observed some peri-nuclear or annular staining for polβ. Whether this represents altered polβ expression or variable polβ expression due to DNA damage is unknown but will be an interesting area for further exploration in the future. Importantly, we provide evidence that low polβ expression at the protein as well as at the mRNA levels is linked with better PFS.
Polβ deficiency in mice is embryonically lethal [42] and embryonic fibroblasts derived from such mice are hypersensitive to alkylating agents [43]. Depletion of polβ delayed the repair of oxaliplatin-induced DNA damage and increased sensitivity in colorectal cancer cell lines [44]. On the other hand, polβ overexpression increased resistance to DNA-damaging agents [45]. Our data concur with a previous study in colorectal cell lines showing platinum sensitivity with polβ depletion [46]. Chronic platinum exposure can alter function of microtubules and microfilaments and can reduce migration [47,48], a feature also seen in A2780cis cells. We show that wild-type cells have expression of N-Cadherin, TGFβ. and MMP9. Polβ depletion reduced N-Cadherin, TGFβ, and MMP9 levels along with induction of E-cadherin. Cadherin switch is the hallmark of EMT [49]. TGFβ and MMP-9 also have a role migration/invasion [49]. Whilst downregulation of several EMT genes was also evident in polβ-depleted cells, we speculate an indirect role for polβ in EMT which require future mechanistic study confirmation. Interestingly, previous studies show PARP1 depletion [50] and Histone H2AX depletion [51] in promoting EMT through transcriptional regulation suggesting a complex role for DNA repair in EMT. In the current study we not only observed increased nuclear localisation of polβ following cisplatin treatment but basal levels of polβ protein also appears to be higher in platinum resistant cells. Although we did not identify any activating mutations of polβ, we speculate that post transcriptional or translational mechanisms could result in overexpression of polβ. Another interesting observation was that despite the lack of polβ variants, bioinformatics analyses revealed several key polβ-interacting proteins harboured novel non-synonymous variants, affecting the LIG3, WRN and XRCC3 loci. However, a limitation to the current study is that further detailed functional and mechanistic studies would be required to fully evaluate if polβ functional interactome may contribute to platinum resistance in ovarian cancer cells.
We show that polβi can induce synthetic lethality in BRCA2-deficient cells. Although promising, future in vivo xenograft studies are required to confirm our findings. PARG blockade is also selectively toxic in polβ-deficient cells. Polβ depletion or inhibition increase SSBs. PARG is essential for efficiency of SSB repair [52]. In polβ deficient cells we observed increased PARG activity. PARG inhibition or depletion can also increase reversed replication forks and post-replicative single-strand breaks [32,33]. In polβ deficient cells with impaired BER and increased SSB, PARG inhibition leads to profound accumulation of SSBs which get converted to DSB leading to synthetic lethality. Accordingly, we observed DSB accumulation, S-phase arrest and apoptosis in PDP00017273-treated polβ-deficient cells compared to -proficient cells. Additionally, PARG inhibition was associated with accumulation of PAR polymers in polβ-deficient cells. Accumulation of PAR was accompanied by NAD+ depletion via two potential mechanisms; (1) PARP uses NAD+ as ADP-ribose donor and catalyse PARylation of target proteins including itself. 2) PARG mediates rapid turnover of PAR to mono-ADP-ribose units which is recycled as the ATP precursor, an important substrate to generate NAD+. PARGi will therefore increase NAD+ [31]. PARG mediated reversal of auto-modification of DNA bound PARP1 leads to poly-ubiquitination of PARP1 by the E3 ligase CHFR. Removal and degradation of PARP1 contribute further to the restoration of NAD+ levels. In PARGi-treated cells, therefore, reduced level of NAD+ may persists [53]. There are at least two mechanisms for NAD+ depletion induced cell death; (1) caspaseindependent programmed necrosis, where excess PAR activates apoptosis-inducing factor, which triggers the apoptotic cascade [35,36]. (2) NAD+ depletion can alter Sirtuin level and impair BER [37,38] contributing further to SSB accumulation. Moreover, PAR accumulation is also known to be toxic to cells and can also impair DNA repair [35]. However, it should be noted that persistent accumulation of PAR polymers in A2780 Polβ_KO cells but transient formation of PAR polymers in the control A2780 was observed in the current study. The exact molecular mechanisms of this phenomenon are not clear although an increased level of PAR formation is expected with inhibition of PARG (which hydrolyses PAR polymers). As described, PAR accumulation and NAD depletion can contribute directly and indirectly to the impairment of DNA machinery, which could be more lethal in the context of polβ deficiency [54,55]. A recent study proposed that loss of PARG activity is a possible mechanism for PARPi resistance in BRCA2 deficient tumours. PARG operates in the same direction as PARPi by preventing PAR accumulation. Hence, decreased PARG activity can allow tumour cells to escape the PARPi mediated synthetic lethality [29]. These findings concur with our observations in polβ deficient ovarian cancer cells. Another study by Pillay et al. has revealed that ovarian cancer cells respond differently to PARGi (PDD00017273) compared to PARP inhibitor (Olaparib) and may be dependent on DNA replication vulnerabilities in cells [34]. Therefore endogenous PARG levels could predict Olaparib response in HR deficient as well as in polβ deficient ovarian cancer cells. PARGi is also synthetically lethal in BRCA1 [32], BRCA2, PALB2, FAM175A and BARD1 deficient breast cancer cells [33].
In conclusion, polβ blockade is a novel approach warrants for development in ovarian cancers.

Materials and methods
Full details are available in Supplementary material and methods.

Clinical study
Polβ immunohistochemistry was completed in 525 epithelial ovarian cancers. Polβ mRNA expression in human epithelial ovarian was investigated in three ovarian tumour gene expression data sets [test set, validation cohort 1 and validation cohort 2 (TCGA). See Supplementary methods for full details. Ethical approval which was obtained from the Nottingham Research Ethics Committee (REC Approval Number 06/Q240/153). All patients provided informed consent.
Methodology for transient knockdown of Polβ and generation of Polβ knockouts are described in Supplementary methods.