Introduction

5–15% of all epithelial ovarian cancers (EOC) are Ovarian Clear Cell Carcinomas (OCCC) [1, 2]. OCCC cases tend to present at an earlier age and are more likely to present with early-stage disease [1, 3]. However, in the advanced setting, OCCC is associated with resistance to standard cytotoxic chemotherapy: response rates to second line chemotherapy in OCCC are as low as 0–8% [4, 5]. Advanced OCCC also has a poorer prognosis compared to stage-matched high grade serous ovarian cancer (HGSOC) [1, 6]. This suggests that the identification of targeted approaches to treating OCCC is of vital importance. Unlike HGSOC, OCCC usually have wild-type (WT) TP53 and a lower frequency of BRCA1 or BRCA2 mutations [7], and presumably a lower frequency of homologous recombination defects, suggesting treatment with platinum salts or PARP inhibitors might not be as widely effective as in HGSOC. Truncating mutations in the SWI/SNF tumour suppressor ARID1A are the most observed genetic aberration in OCCC, seen in 40–57% of cases [8,9,10]. Mutations in PIK3CA (33%), PPP2R1A (7–10%), TERT promoter (16%), SYNE1 (20%) and KRAS (5%) are also seen in OCCC, along with ERBB2 (14%) and AKT2 (14%) amplification events [8, 9, 11,12,13,14].

Previously, we identified a synthetic lethal interaction between ARID1A loss-of-function and ATR inhibition in preclinical models of OCCC [15]. ATR is a serine-threonine kinase that plays an essential role in the DNA damage response [16]. Our original work, and subsequent studies, indicated that this is a relatively generalizable effect, being caused by ARID1A mutation or other SWI/SNF component defects causes [15, 17,18,19,20]. This pre-clinical work led to a phase II clinical trial, ATARI (NCT04065269), to evaluate the activity of the ATRi ceralasertib (AZD6738, AstraZeneca) in advanced ARID1A-defective gynaecological cancers, including OCCC and endometrial clear cell carcinoma [21]. Other clinical trials involving ATR inhibitors are also underway [22]. In parallel, we adopted a functional genomics approach to identify genetic modifiers of ARID1A/ATRi synthetic lethality. As described below, we found that loss-of-function driver mutations in the protein phosphatase 2A coding gene, PPP2R1A, a recurrent event in OCCC, also cause ATRi sensitivity, even in OCCC tumour cells with pre-existing ARID1A defects.

Results

Genome-wide CRISPR-Cas9 screen identifies protein phosphatase 2A subunits as genetic determinants of ATRi sensitivity in OCCC

To better understand what, in addition to ARID1A, controls ATRi responses in OCCC, we carried out a genome-wide CRISPR-Cas9 mutagenesis ATRi chemosensitivity screen in ATRi-sensitive, ARID1A mutant TOV21G OCCC cells (ARID1A p.549fs/p.756fs). We generated Cas9-positive cells and confirmed their sensitivity to two different clinical ATRi (Fig. 1A, B). Cas9+ TOV21G cells were then infected with a short guide RNA (sgRNA) library designed to target 18,010 genes (Fig. 1C). Transduced cells were exposed to AZD6738 or drug vehicle (DMSO) for two weeks, after which sgRNA frequency in the surviving cell populations was determined by deep sequencing. Screen quality was assessed by determining the performance of sgRNA designed to target “core essential genes” – i.e. those genes that are essential in most cell lineages [23]. We observed significantly lower Norm Z scores in the core essential genes when compared to the non-essential genes (p < 0.0001, non-parametric unpaired Wilcoxon rank sum test, Supplementary Fig. 1). By comparing sgRNA frequencies in DMSO and ATRi-exposed cultures, we determined the identity of genes that controlled ATRi sensitivity/resistance. Consistent with prior observations [24, 25], sgRNA designed to target either POLE3 or POLE4 (DNA polymerase ε accessory subunit encoding genes) and genes encoding components of the ATR signalling cascade (RAD9A, CHEK1) caused AZD6738 sensitivity whereas sgRNA designed to target CDC25B and CDK2 caused resistance (Fig. 1D, Supplementary Table S1). We also noted that sgRNA targeting the protein phosphatase 2A (PP2A) subunits PPP2CA and PPP2R2A caused ATRi sensitivity (Fig. 1D, E), an effect analogous to the effect of PPP2R2A shRNA in non-small cell lung cancer tumour cell lines [26]. The PP2A scaffolding subunit gene, PPP2R1A did not score as a “hit” in this screen. Since PP2A subunits are recurrently altered in OCCC, we assessed whether this might represent an exploitable vulnerability in OCCC.

Fig. 1: Genome wide CRISPR-Cas9 screens identify PP2A subunits as genetic determinants of ATR inhibitor sensitivity.
figure 1

A, B Dose response curves illustrating the ATRi sensitivity of OCCC TOV21G cells. Cells were plated in 384 well plates and exposed to either AZD6738 (A) or VX970 (B) for five continuous days, at which point cell viability was assessed using CellTiter-Glo®. ARID1A+/+ and ARID1A–/– HCT116 cells are included as controls. Error bars represent standard error of the mean (SEM) from 5 replicates. C Screen schema for genome wide CRISPR-Cas9 mutagenesis screen performed in TOV21G cells. TOV21G cells expressing a doxycycline-inducible SpCas9 transgene were used. AZD6738 SF50 was 0.1 µM. D Waterfall plot showing the ranked gene-level NormZ scores from the genome wide CRISPR-Cas9 screen. PPP2R2A and PPP2CA were identified as sensitising ”hits” along with several other genes previously known to cause both sensitivity to ATRi such as POLE3, POLE4, CHEK1 and RAD9A; known resistance-causing effects such as CDK2 and CDC25B were also identified. E Guide-level Z scores from the genome wide screen for sgRNAs designed to target PPP2R2A or PPP2CA. F Experimental scheme for additional CRISPR mutagenesis (CRISPRn) (1, 2) or CRISPR interference (3,4) (CRISPRi) screens for genetic determinants of ATRi sensitivity. Non-tumour epithelial MCF10A p53mutant cells, with a deleterious P53 mutations, expressing a doxycycline-inducible Cas9 transgene or constitutive dCas9-KRAB transgene were used as shown. AZD6738 SF80 in MCF10A p53mutant cells was 1 µM. VX970 SF80 was 0.1 µM. G Violin plots showing the quantile normalized NormZ scores for either 18,009 genes (CRISPRn screens) or 18,905 genes (CRISPRi screens) from the screens described in and labelled (1–4) as in F. NormZ scores for PPP2R1A (green), PPP2R2A (red) and PPP2CA (blue) and ATM (magenta; positive control) are highlighted. Dotted line represents a Z –2 threshold used to define “hits”. H Scatter plot showing the gene-level Norm Z score from the CRISPRi screens performed with AZD6738 and VX970 described in E and F. Data points for PPP2R1A (green), PPP2R2A (red) and PPP2CA (blue) and ATM (magenta; positive control) are highlighted.

Full size image

To assess whether PP2A-CRISPR ATRi sensitivity was private to TOV21G cells, or a more general effect, we carried out four additional genome-wide CRISPR screens in a non-tumour epithelial line with an engineered p53 mutation, MCF10A p53mutant. In this case, a p53 mutant cell line was used as loss of p53 often allows non-tumour cell lines to tolerate tumour suppressor gene loss, thus potentially maximizing the ability to detect relationships between defects in tumour suppressors and ATRi sensitivity. We carried out both CRISPR mutagenesis (CRISPRn) and CRISPR interference (CRISPRi) screens in MCF10A p53mutant cells, using two different ATR inhibitors, AZD6738 and VX970 (Fig. 1F). All four screens identified PPP2CA (encoding the catalytic subunit), PPP2R1A (scaffolding subunit) and PPP2R2A (regulatory subunit) as determinants of ATRi sensitivity (Fig. 1G, H, Supplementary Fig. 2). Taken together with the TOV21G screen data, this suggested that defects in protein phosphatase 2 A cause a penetrant [27] drug sensitivity effect, operating in multiple, molecularly distinct settings. As we saw enhanced ATRi sensitivity caused by PP2A defects in both ARID1A defective, p53 wild type TOV21G as well as in ARID1A wild type, p53 mutant MCF10A p53mutant cells, we surmised that the effect of PP2A defects on ATRi sensitivity was not wholly dependent upon either ARID1A or p53 status.

PPP2R1A p.R183W and p.R183P mutations cause ATRi-induced S phase stress, premature mitotic entry, genomic instability and ATRi sensitivity in OCCC tumour cells

In gynaecological cancers, deletions in PPP2R2A, which encodes a PP2A regulatory subunit, are recurrent in serous ovarian cancer (Fig. 2A, B), whereas dominant negative missense mutations in the scaffolding subunit, encoded by PPP2R1A, are more frequent in non-serous gynaecological malignancies such as OCCC [28]. For example, in a study of 42 OCCCs (of which the majority were primary tumours) 7% exhibited a heterozygous PPP2R1A mutation; the clustering of PPP2R1A mutations around residue p.R183 in this study suggested that PPP2R1A might act as an oncogene in OCCC [8]. To independently assess the frequency of PPP2R1A mutations, we genotyped PPP2R1A in archival tissue from a cohort of 23 OCCC cases which were previously characterized with regards to 59 genes but were uncharacterized regarding their PPP2R1A mutation status (patient details in Supplementary Table S2): 52% (12/23) exhibited a heterozygous missense mutation at residue p.R183 (Fig. 2A, C, Supplementary Fig. 3). This led us to explore the possibility that PPP2R1A mutations in OCCC could be a clinically relevant biomarker of ATRi synthetic lethality in this disease. We also noted that in the ARID1A mutant cases in our OCCC cohort, 33% (4/12) had a co-occurring PPP2R1A mutation (Fig. 2C). As ARID1A mutations in OCC have previously been linked to ATRi synthetic lethality [15, 17, 18] we also tested whether PPP2R1A mutations could enhance ATRi sensitivity in OCCC cells with a pre-existing ARID1A defect.

Fig. 2: PPP2R1A mutations in OCCC cause ATRi synthetic lethality.
figure 2

A Model of the PP2A holoenzyme (PDB 2NPP) with scaffolding subunit shown in blue, regulatory subunit shown in green and catalytic subunit shown in yellow. Somatic missense mutations in PPP2R1A (scaffolding subunit), highlighted in magenta, are conserved across multiple cancer types and are involved in direct interactions with regulatory B-subunits of the PP2A holoenzyme. Oncoprints illustrating PPP2R2A deletions in serous ovarian cancer (B) or PPP2R1A mutations (52% of cases) in ovarian clear cell carcinoma (OCCC – shown in C). Oncoprint in B generated from data in (ICGC/TCGA, 2020). Oncoprint in C generated by genotyping of 23 cases of OCCC from the Royal Marsden Hospital. D Heterozygous PPP2R1A missense mutations introduced to TOV21G cell line using CRISPR-prime gene editing. Cells were transfected with plasmids encoding prime editing guide (PEG), prime editor 2 (PE2) and nicking sgRNA before being single cell sorted, expanded and genotyped. Two resultant clones were generated with heterozygous PPP2R1A p.R183P or p.R183W missense mutations. DNA sequence traces from PPP2R1A mutant clones are shown. Each mutation is heterozygous, present in 5/10 TOPO-cloned sequences. E Western blot illustrating that PPP2R1A p.R183P mutation causes reduced levels of PPP2R2A and PPP2CA. Western blot performed from whole cell extracts from TOV21G PPP2R1A WT, p.R183P and p.R183W cells. F Volcano plot illustrating results from phosphoproteomic profiling of PPP2R1A mutant cells, indicating a significant enrichment (Log2 fold change >0, unpaired t-test <0.05) of known PP2A phosphorylation sites in SPRY1 (two sites) HDAC5 and IL6ST (highlighted). G Heterozygous PPP2R1A p.R183P or p.R183W mutation causes increased sensitivity to the ATR inhibitor AZD6738. Dose response curve for cells from D exposed to AZD6738 for two weeks. Error bars represent SEM from 4 replicates. Significance determined using two-way ANOVA.

Full size image

Most prior functional studies have used gene silencing to model the effect of PPP2R1A dysfunction [29] or used ectopic expression of mutant PPP2R1A cDNA to model cancer associated PPP2R1A mutations [30,31,32,33,34]. With the advent of CRISPR-Cas9 mutagenesis, introducing precise mutations into endogenous genes is now more achievable. We therefore used CRISPR-prime editing [35] to introduce p.R183 missense mutations into PPP2R1A wild type, ARID1A mutant TOV21G OCCC cells, generating daughter clones with either heterozygous PPP2R1A p.R183P or p.R183W mutations, similar to those seen in OCCC (Fig. 2D). In characterizing daughter clones with p.R183P or p.R183W mutations, we found that total levels of PPP2R2A and PPP2CA were reduced (Fig. 2E, Supplementary Figure 4), consistent with PPP2R1A mutation destabilizing the PP2A complex [30]. In addition, when we carried out mass spectrometry based phosphoproteomic profiling of PPP2R1A p.R183P and wild type isogenic cells, we found phosphosites previously shown to be dephosphorylated by PP2A (e.g. those on HDAC5 [36], IL6ST [37] and SPRY1 [38]) were enriched in p.R183P mutant cells (Fig. 2F, Supplementary Table S3), consistent with a PP2A defect in these cells. When we assessed ATRi sensitivity in PPP2R1A mutant cells, we found that TOV21G clones with a heterozygous p.R183P or p.R183W mutation were more sensitive to ATRi than wild type cells (Fig. 2G, Supplementary Fig. 5). We presume the prior failure to see PPP2R1A as a “hit” in the prior TOV21G genome-wide screen (although other PP2A subunits were hits) was due to a false negative effect, common to such high-throughput screens.

ATRi sensitivity in tumour cells is often associated with a decrease in the replicating S phase fraction of cells and premature mitotic entry [39, 40]. To understand the phenotypes associated with ATRi sensitivity in PPP2R1A mutant OCCC cells, we used EdU labelling to label cells in replicating S phase and used phosphorylation of histone H3 (pH3) as a marker of cells in mitosis. We found that ATRi exposure in PPP2R1A p.R183 mutant cells caused a significant reduction in the fraction of cells in replicating S phase (Fig. 3A–C). In addition, PPP2R1A p.R183P mutation caused a modest increase in the fraction of pH3-positive cells, an effect that was exacerbated by ATRi exposure (Fig. 3D, E); we also noted an increase in the proportion of pH3-positive cells with a DNA content of <4 N, consistent with premature mitotic entry (Fig. 3F). Chromosomal lesions, such as those that are generated prior to mitosis are often sequestered in nuclear compartments marked by p53-binding protein 1 (53BP1) [41]. We found that ATRi exposure in PPP2R1A p.R183P or p.R183W mutant cells caused a significant increase in the proportion of cells with more than one 53BP1 body but did not do so in PPP2R1A wild type cells (Fig. 3G, H). Consistent with previous reports [41, 42], 53BP1 bodies were observed in cells which stained negative for cyclin A (Fig. 3G), in keeping with their accumulation in the G1 phase of the cell cycle. Given the observed S phase defects and premature mitotic entry, we proceeded to perform time-lapse microscopy in order to examine mitotic kinetics in PPP2R1A mutant cells. ATRi exposure increased the metaphase duration in PPP2R1A mutant cells (Fig. 3I). Indeed, in a proportion of PPP2R1A mutant cells, ATRi exposure led to cells failing to progress beyond metaphase and ultimately undergoing apoptosis (Supplementary Fig. 6). Commensurate with ATRi having a more deleterious effect in PPP2R1A p.R183P and p.R183W mutant cells than in wild type cells, we found that ATRi exposure in PPP2R1A mutant cells caused a 12-fold (p.R183P) or 10-fold (p.R183W) increase in micronuclei, a common form of genomic instability caused by ATRi (Fig. 3J, Supplementary Fig. 7). Taken together, these observations suggested that in PPP2R1A mutant OCCC cells, ATRi elicits increased S phase stress, premature mitotic entry, genomic instability and synthetic lethality.

Fig. 3: PPP2R1A p.R183W and p.R183P mutations cause ATRi-induced S phase stress, premature mitotic entry and genomic instability (previous page).
figure 3

A FACS plots illustrating that PPP2R1A R183P mutation and ATRi exposure causes a significant increase in the non-replicating S phase fraction at the expense of the active (replicating) S phase fraction. Quantification of changes in fraction of cells in replicating S phase (B) or non-replicating S phase (C) from the experiment described in A. Error bars represent mean and SEM from three replicates. Pairwise significance determined by two-way ANOVA with Sidak correction for multiple comparisons. D FACS plots illustrating that PPP2R21A mutant cells more readily enter mitosis following AZD6738 exposure and are most likely to do so with sub-4n DNA content. Quantification of phospho-histone H3 positive cells (E) and sub 4n fraction of cells (F) from experiment shown in D. Error bars represent mean and SEM from three replicates. Pairwise significance determined by two-way ANOVA with Sidak correction for multiple comparisons. G Heterozygous PPP2R1A p.R183P mutation causes the formation of 53BP1 bodies. TOV21G PPP2R1A p.R183P isogenic cells were exposed to either DMSO or AZD6738 (500 nM) for 24 h before being fixed and immunostained for 53BP1, cyclin A and with DAPI. Representative confocal microscopy images are shown. Scale bar represents 10 µM. H Quantification of 53BP1 bodies from experiment described in G. Error bars represent mean and SEM from three replicates. Pairwise significance determined via Two-way ANOVA with Sidak’s correction for multiple comparisons. I Exposure to ATRi leads to a significant increase in the duration of metaphase in cells with a PPP2R1A mutation. Quantification of the duration mitotic cells spend in metaphase. Viable TOV21G PPP2R1A WT, p.R183P or pR183W were labelled with live cell stains for DNA, tubulin and actin before being exposed to AZD6738 (500 nM) or DMSO for 24 h with an image being captured every 5 min. Error bars represent SEM for 50 mitotic cells. Pairwise significance determined via Two-way ANOVA with Sidak’s correction for multiple comparisons (J) Exposure to ATRi leads to a significant increase in the proportion of cells with micronuclei in cells with a PPP2R1A mutation. TOV21G PPP2R1A isogenic cells were exposed to either DMSO or AZD6738 (500 nM) for 24 h before being fixed and stained with DAPI. Error bars represent SEM from three triplicate experiments. Pairwise significance determined via Two-way ANOVA with Sidak’s correction for multiple comparisons.

Full size image

PPP2R1A/ATRi synthetic lethality operates in vivo

In order to assess whether PPP2R1A mutations conferred ATRi sensitivity in vivo, we xenotransplanted isogenic TOV21G cells and the daughter clone with a heterozygous PPP2R1A p.R183W mutation into recipient mice. To model local metastasis in OCCC, we introduced these into the peritoneal cavity of mice, monitoring tumour growth by IVIS-based luminescence, as previously described [15]. After tumours had established, we treated tumour bearing animals with AZD6738 or the drug vehicle DMSO and monitored tumour volume for 17 days (Fig. 4A). As TOV21G OCCC cells have an ARID1A loss-of-function mutation and are somewhat sensitive to ATRi, both in vitro and in vivo [15], we used a relatively low dose of AZD6738 in these experiments (25 mg/kg, using cycles of five days on treatment followed by two days break) to see if PPP2R1A mutation enhanced ATRi sensitivity in the context of a pre-existing ARID1A/ATRi effect. We found that when compared to vehicle treatment, this low dose ATR inhibitor treatment had minimal effects on tumour growth in the absence of a PPP2R1A mutation (Fig. 4B, D, Supplementary Fig. 8) but suppressed tumour growth in the presence of a PPP2R1A mutation (Fig. 4C, D, Supplementary Fig. 8). As expected, this low dose ATRi treatment had minimal effects on animal body weight or condition (Fig. 4E, F).

Fig. 4: PPP2R1A/ATRi synthetic lethality operates in vivo.
figure 4

A In vivo study schema. Luciferase tagged PPP2R1A WT or p.R183W TOV21G cells were introduced into recipient mice by intraperitoneal injection. Treatment with either AZD6738 or vehicle commenced after tumours had established (after day six post IP injection) using on a five days on, two days off schedule. A low dose of ATRi (25 mg/kg) was used in these experiments to account for the pre-existing ARDI1A/ATR inhibitor synthetic lethality [15]. Total flux luminescence was measured twice per week. B Treatment with AZD6738 does not impair the growth of PPP2R1A WT tumours. Line plots showing fold change in luminescence compared to pre-treatment level for mice xenografted with TOV21G PPP2R1A WT cells. Error bars represent mean and SEM. Significance determined by two-way ANOVA. C Treatment with AZD6738 impairs the growth of PPP2R1A mutant tumours. Line plots showing fold change in luminescence compared to pre-treatment level for mice xenografted with TOV21G PPP2R1A p.R183W cells. Error bars represent mean and SEM from 15 mice. Significance determined by two-way ANOVA. D Representative images of vehicle and AZD6738 treated mice taken 14 days post initiation of treatment. E, F Low dose ATRi treatment had minimal effects on animal body weight. Weight of each animal expressed as a percentage relative to weight at time of tumour implantation.

Full size image

Discussion

We found that the introduction of oncogenic hotspot mutations in residue p.R183 of PPP2R1A enhanced ATRi sensitivity in pre-clinical models of OCCC, even in the presence of ARID1A mutations, a known cause of ATRi sensitivity [15]. The effect of PPP2R1A mutation on ATRi sensitivity was observed both in vitro and in vivo and is characterized by decrease in cells in replicating S phase of the cell cycle, an increase in cells entering mitosis with sub-4n genomic content and the accumulation of 53BP1 bodies, a feature of residual DNA damage acquired in S phase and persisting in mitosis [41]. The reoccurrence of PPP2R1A mutations in OCCC and our data presented here suggests that in addition to ARID1A, PPP2R1A status should also be assessed retrospectively as a biomarker of ATRi response in trials such as NCT04065269 (ATARI) and NCT03682289 or even assessed prospectively in new clinical trials. The potential for sub-stratifying relatively rare cancer subtypes, such as OCCC, into additional biomarker-defined subsets associated with a therapy (e.g. subclassifying PPP2R1A mutant patients to receive ATRi), does however, present some challenges, both economic and logistic. For example, the ability to translate our findings dictates that PPP2R1A status in OCCC can be established. In this regard, we note that PPP2R1A mutations can be detected using now commonly used DNA sequence capture panels [43], suggesting that patients with defects in this gene could be identified and, if necessary, stratified to receive treatment including an ATRi.

We do note several potential limitations of our study and additional areas for future work. Firstly, although our work establishes the idea that PPP2R1A mutations are associated with ATRi sensitivity, only future clinical trials where patients are pre-stratified according to the mutational status of these two genes will formally establish this. Ideally, such trials would be appropriately scaled so that the impact of tumoural mutations in either ARID1A or PPP2R1A on the anti-tumour efficacy of ATR inhibition could be assessed, alongside an assessment of how the co-occurrence of both ARID1A and PPP2R1A mutations influences ATRi responses. In ARID1A wild type MCF10A cells (Fig. 1) we show that PP2A defects in the absence of an ARID1A defect cause ATRi sensitivity and in Figs. 1–4 we show that mutagenizing PPP2R1A in an ARID1A mutant OCCC tumour cell also causes ATRi sensitivity. Taken together this suggests that PPP2R1A modifies ATRi sensitivity via a mechanism independent to the ARID1A/ATR synthetic lethal effect. This, however, remains to be formally tested and could only be achieved by generating a new series of isogenic OCCC tumour cell lines from progenitor cells which have neither a SWI/SNF nor a PP2A defect, upon which different constellations of ARID1A and/or PPP2R1A mutations are imposed.

Secondly, most of our modelling of PPP2R1A mutations and their effect on ATR inhibitor sensitivity was based on experiments in PPP2R1A wild type TOV21G cells where we mutagenized the PPP2R1A gene. Other OCCC tumour cell lines are commonly used but these already possess oncogenic PPP2R1A mutations (e.g. OVTOKO and OVISE) and are already known to be ATR inhibitor sensitive [15], and so could not be used to establish, using an isogenic approach, the impact of imposing a PPP2R1A mutation upon a OCCC cell. We also used PPP2R1A mutagenized TOV21G cells to model the in vivo effect of PPP2R1A mutations on ATRi response in ARID1A defective tumour cells. Whilst this approach allowed us to assess the effect of PPP2R1A mutation in an isogenic setting and in an orthotopic (peritoneal) site, it is possible that modelling this scenario using patient derived xenografts (PDX) could be of use. However, we do note that in order to carry out such an experiment, appropriate OCCC PDX do need to be available, something that will only come with increased efforts to biopsy and establish PDX from this relatively rare gynaecological cancer subtype. In addition, we also note that our ability to generate isogenic OCCC tumour cell lines with/without PPP2R1A mutations generated by prime editing highlights that new models of the disease might come from genetic manipulation of existing OCCC models. This in turn, might allow not only the identification of new candidate targets in the disease, but could also lead to a greater degree of treatment stratification, allowing targets to be identified for not only the PPP2R1A and ARID1A mutant patients, but also those OCCC cases where only one or neither of these driver genes are mutated.

We also note that the key PP2A substrates that explain ATRi sensitivity remain to be identified, mainly as PP2A dephosphorylates multiple substrates; as such ascribing the effect of PPP2R1A mutations to one such substrate remains challenging. Nevertheless, substrates of the PP2A holoenzyme have previously been implicated in modulating proteins such as Myc that control the extent of replication fork stress, a candidate driver of ATRi sensitivity [24, 44,45,46,47]; specifically PP2A dephosphorylates Myc on residue Ser62 and PPP2R2A (not PPP2R1A as described here) silencing has been shown to cause an increase in Myc protein levels [26]. In addition, PP2A dephosphorylates WEE1 on residues S53 and S153 and PP2A defects cause decreased WEE1 activity [48]. In principle, changes in either WEE1 and/or Myc could explain the ATRi sensitivity of PPP2R1A mutant OCCC cells. However, our preliminary investigations suggest that when assessed in isogenic systems, neither p.R183P or p.R183W mutations result in substantially elevated cMyc or decreased WEE1 (Stewart et al., in preparation), suggesting that neither of these provide trivial explanations for the ATRi sensitivity seen. To investigate this further, our future work will focus on using whole cell proteomics and phosphoprotomics to gain a more holistic view of how the proteome rewires in response to precise PPP2R1A mutations.

Methods

Cell lines

TOV21G cell line was obtained from American Type Culture Collection (ATCC). HCT116 ARID1A isogenic pair was obtained from Horizon Discovery. OVISE and OVTOKO cell lines were obtained from Dr Hiroaki Itamochi (Tottori University School of Medicine, Yonago, Japan). All cell lines were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium (Gibco) and supplemented with 10% foetal bovine serum (FBS, Gibco) and penicillin-streptomycin 100 units/mL (Invtirogen). Cell line identity and Mycoplasma infection was tested prior to and during the study by using short tandem repeat typing StemElite Kit (Promega) and MycoAlert Mycoplasma Detection Kit (Lonza), respectively

Cell viability assays

For short term viability assays, cells were seeded into 96-well or 384-well plate at a density of 500 cells per well in 200 µL or 50 µL, respectively. 24 h after seeding, drug was added at indicated concentrations using the Echo liquid handler (Labcyte). In the case of 96-well plates this entailed the removal of culture medium, addition of drug before 200 µL fresh medium added to bring drug to intended concentration. This process was repeated after 3 days. For 384-well plates, drug was added directly to the plate (i.e. medium was not removed). After 7 (96-well) or 5 (384-well) days cell viability was measured using CellTiter-Glo® (CTG, Promega). Medium was removed from each well before 50 µL (96-well) or 25 µL (384-well) CTG (diluted 1:4 in PBS) was added to the plate which was subsequently incubated at room temperature for 20 min under constant agitation. Luminescence was measured using the Victor X5 Multilabel plate reader (Perkin Elmer). From these luminescence readings a surviving fraction was calculated by normalising the luminescence reading for each well to that of the vehicle treated well. Dose response curves were then generated using GraphPad Prism and significance determined using two-way ANOVA. For long term viability assays, Cells were seeded into 24 well plate at density of 500 cells per well in 1 mL culture medium. 24 h after seeding medium was removed and replenished with medium containing drug at indicated concentration. This process was repeated every 3 days until vehicle treated well became confluent (typically 14–21 days). Cell viability was then assessed using CTG.

Genome-wide CRISRPn/i screens

Doxycycline inducible SpCas9-expressing (Horizon, #CAS11229) or constitutively expressing catalytically-inactive, “dead”, dCas9-KRAB (Addgene, #50917) cells were generated by transduction (Dharmacon) and selected in Blasticidin 7 µg/mL. Cas9 expression was confirmed via Western blot.

The genome-wide CRISPRn screen was performed as previously described [24]. Inducible SpCas9 expressing cells were seeded for 1000x representation per sgRNA, and infected at a multiplicity of infection (MOI) of 0.3, to avoid multiple sgRNA infections per cell, using the previously published and validated genome-wide human lentiviral library (Kosuke Yusa Human GW CRISPR guide RNA library V1) [49]. After puromycin selection (1 µg/mL), cells were exposed to AZD6738/VX970 or DMSO and harvested at day 0 (T0) and on day 14 (T1). Media containing AZD6738 or DMSO was replenished twice per week. Genomic DNA was extracted from the T = 0 and T = 1 samples using DNeasy Blood and Tissue Kit (Qiagen) as per manufacturers instructions. sgRNA sequence were amplified with Q5 polymerase (NEB) with the following primers Forward 5’- ACACTCTTTCCCTACACGACGCTCTTCCGATCTCTTGTGGAAAGGACGAAACA and Reverse 5’- TCGGCATTCCTGCTGAACCGCTCTTCCGATCTCTAAAGCGCATGCTCCAGAC. The PCR products sequenced on HiSeq2500 and the data was analysed as previously described [50]. The genome wide CRISPRi screen was performed using the same experimental design without the need for dCas9-KRAB induction.

Derivation of models using CRISPR prime gene editing

CRISPR prime gene editing was performed using PE2 system as described in [35]. Briefly, TOV21G cells were reverse transfected with CRISPR prime gene editing machinery. 1125 ng PE2 plasmid, 375 ng pegRNA plasmid, 124.5 ng nicking sgRNA plasmid and 2 µL P3000 (Invitrogen) mixed in 250 µL Optimem and incubated at room temperature for 10 min. 5 µL lipofectamine 3000 mixed with 250 µL optimum and incubated at room temperature for 10 min. The CRISPR prime editing mix and transfection mix were then combined in a 6 well plate and incubated at room temperature for 10 min. 300,000 TOV21G cells suspended in serum free medium and added to combined transfection mix. Five days post transfection, cells were single cell sorted into 96 well plates and expanded. Once 90% confluency was reached, cells were trypsinised and replicated into two replicate plates. Genomic DNA was extracted from one plate, whilst the duplicate plate was viably frozen in FBC with 10% DMSO and stored at –80 °C. PCR based genotyping of clones was performed. PCR products were purified using QIAquick PCR Purification Kit (Qiagen) eluted in 50 µL nuclease free water as per the manufacturer’s instructions. For Sanger sequencing, 15 µL purified PCR product mixed with 2 µL PPP2R1A forward primer (10 µM) described in section 2.2.8. Sanger sequencing was performed by Eurofins Genomics and results were analysed using SnapGene software. The pegRNA sequences used to generate the PPP2R1A p.R183P model was 5’-AGGCTGCGGCGGGCCGCACCATGGGGGT and the sequence for the PPP2R1A p.R183W model was 5’-AGGCTGCGGCCCACCGCACCATGGGGGT. The nicking sgRNA sequence was 5’-GTTGTCCAGCTCCAGCACCT.

siRNA gene silencing experiments

Reverse transfections using the siRNA SMARTpool and siCON2 negative controls (Dharmacon) were carried out in 384-well plates or 6-well plates. Cells were reverse transfected with 20 nM siRNA with 5% RNAiMAX (Thermo Fisher Scientific) in 50 µL (384-well plate) or 500 µL (6-well plates). Lysates were retrieved or viability experiments were performed after 2 to 3 days following transfection.

Western blot analysis

Whole cell lysates were extracted using Radioimmunoprecipitation assay (RIPA) buffer (Thermo scientific) supplemented with Complete mini proteinase inhibitor (Roche). Western blots were carried out with precast Bis-Tris gels (Invitrogen, Paisley, UK). The following primary antibodies were used in this study PPP2R1A (1:1000, Cell Signalling, 2041S), PPP2R2A (1:1000, Cell Signalling, 5689S), PPP2CA (1:1000, Abcam, Ab106262) and vinculin (1:1000, Santa Cruz, SC-73614).

Cell cycle analysis

150,000 cells were seeded to 6 well plates. 24 h later, medium was removed and replenished with medium containing drug or DMSO and the cells were exposed for an additional 24 h. For the final 2 h of drug exposure 20 µM 5-ethynyl-2’-deoxyuridine (EdU, ThermoFisher) was added to the culture medium. Cells were harvested, washed in ice cold PBS before fixation in 70% ethanol at 4°C overnight. Staining of cells was carried out using Click-iT™ Plus EdU Alexa Fluor™ 647 Flow Cytometry Assay Kit (Invitrogen) as per manufacturer’s instructions. Briefly, cells were permeabilised in 1× saponin diluted in PBS before being stained with Alexa Fluor™ 647. Cells were washed and then incubated with Alexa Fluor™ 488-conjugated anti-phosphohistone H3 (S10) antibody (Invitrogen) at 1:50 concentration for 1 h at room temperature, protected from light. Cells were then digested and stained with PI for DNA content with FxCycle™ PI/RNase Staining Solution (ThermoFisher). Detection of EdU staining, phospho-histone 3 staining, or PI was performed on a BD LSR II flow cytometer (BD biosciences) and the selected population was analysed regarding its cell cycle distribution using FlowJo (BD) FACS analysis software.

Immunofluorescence

Cells were seeded to poly-L-lysine coated cover slips (Corning). Once cells reach 60% confluence medium was removed and replenished with medium containing drug or DMSO. After 24 h cells were fixed in 4% PFA for 15 min at room temperature. Fixed cells were washed in PBS 3 times and then stored in PBS at 4 degrees. For staining of intracellular proteins and DNA, cells were first permeabilised in 0.25% Triton-X in PBS for 15 min before being washed 3 times and blocked in immunofluorescence buffer (IFF, 5% FBS + 3% bovine serum albumin [BSA, Sigma] in PBS) for 1 h at room temperature. Cells were then incubated in primary antibody overnight at 4 °C. Cells were washed again in PBS 3 times before being incubated with Goat anti-Mouse IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 555 (Thermo Fisher Scientific) and Goat anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 488 (ThermoFisher) secondary antibody (1:1000 dilution) for 1 h at room temperature. After a further 3 washes in PBS, cells were mounted on glass cover slips using ProLong™ Gold Antifade Mountant with DAPI (Invitrogen). Mounted coverslips were stored in the dark at 4 °C until imaging. Z stack images were acquired using Marianas advanced spinning disc microscopy (3i) with the generation of maximum projections performed using Slidebook6 (3i) software. Quantification of 53BP1 bodies was performed using CellProfiler imaging analysis software. Micronuclei were measured manually and expressed as proportion of cells in visual field (minimum 100 cells per condition).

Time lapse microscopy

Cells were seeded in flat bottom 96-well imaging plates (Perkin Elmer) in phenol-free RPMI medium (Gibco). Once cells reached approximately 60% confluency medium was removed and replenished with medium containing SPY505-DNA nuclear stain (Tebubio), SPY650-tubulin Probe (Universal biologicals) and SPY555-actin Probe (Universal biologicals) all at 1:1000 dilution. After culturing for four hours either DMSO or AZD6738 to a final concentration of 500 nM was added to each well. Cells were imaged every 5 min for 24 h with the Marianas spinning-disk microscope system (Intelligent Imaging Innovations, 3i) with the generation of maximum projections performed using Slidebook6 (3i) software. Throughout imaging, cells were maintained at 37°C in 5% CO2. Duration of metaphase for 50 mitotic cells per condition was measured manually.

In vivo assessment of ATRi sensitivity

1 ×106 luciferase expressing TOV21G cells were diluted in 100 µL sterile PBS and injected directly into the intraperitoneal cavity of 30 BALB/c nude mice per PPP2R1A genotype. Cohort size for this experiment was based on prior experiments described in [15]. Six days post inoculation, animals were randomised to two cohorts: (i) those subsequently treated with vehicle; or (ii) those treated with a low dose AZD6738 25 mg/kg on a 5 days on, 2 days off regime. A low dose of ATRi was used in these experiments to account for the pre-existing ARDI1A/ATR inhibitor synthetic lethality [15]. In vivo imaging using IVIS® Spectrum In Vivo Imaging System was performed three times per week until 17 days post treatment initiation. Ten minutes before imaging luciferin 150 mg/kg was injected into the peritoneum. IVIS imaging was discontinued after 17 days as the development of abdominal ascites resulted in unreliable luminescence readings. Animals continued to be treated and monitored for survival assessment up until 30 days. Home office end points granted on the project licence were applied for decisions regarding culling. These included signs linked to tumours such high IVIS signal, abdominal distention due to ascites, low body condition score and change in health status related to abdominal tumours. Throughout the study, body weight was measured prior to the administration of vehicle/AZD6738, with animals being culled if more than a 20% relative weight loss was observed. Analysis of the data (e.g. tumour burden by IVIS and date of animal death) was carried out by investigators blinded to the allocation of mice into the two cohorts. All in vivo experiments were carried out according to ARRIVE guideline and via protocols described in a UK Home Office Licence held by CJL and according to an approval by ICR’s Animal Welfare and Ethical Review Body.

Sample preparation for proteomics

Cell pellets were lysed in 150 μL lysis buffer of 100 mM triethylammonium bicarbonate (TEAB), 1% sodium deoxycholate (SDC), 10% isopropanol, 50 mM NaCl and Halt protease and phosphatase inhibitor cocktail (100X) (Thermo, #78442) on ice, with 15 sec of pulsed probe sonication followed by heating at 90 °C for 5 min and another round of sonication for 5 s. Protein concentration was measured with the Quick Start Bradford protein assay (Bio-Rad) according to manufacturer’s instructions. Protein aliquots of 60 μg were reduced with 5 mM tris-2-carboxyethyl phosphine (TCEP) for 1 h at 60 °C and alkylated with 10 mM iodoacetamide (IAA) for 30 min in the dark. Proteins were digested overnight with trypsin at final concentration 75 ng/μL (Pierce). Peptides were labelled with the TMTpro reagents (Thermo) according to manufacturer’s instructions. The pooled sample was acidified at 1% formic acid, the precipitated SDC was removed by centrifugation and the supernatant was SpeedVac dried.

LC-MS analysis

LC-MS analysis was performed on a Dionex UltiMate 3000 UHPLC system coupled with the Orbitrap Lumos Mass Spectrometer (Thermo Scientific). Peptides were loaded onto the Acclaim PepMap 100, 100 μm × 2 cm C18, 5 μm, trapping column at flow rate 10 μL/min and analysed with an Acclaim PepMap (75 μm × 50 cm, 2 μm, 100 Å) C18 capillary column connected to a stainless-steel emitter. Mobile phase A was 0.1% formic acid and mobile phase B was 80% acetonitrile, 0.1% formic acid. For the phosphopeptide analysis, the separation method was as follows: for 60 min gradient 5–38% B, for 10 min up to 95% B, for 5 min isocratic at 95% B, re-equilibration to 5% B in 5 min, for 10 min isocratic at 5% B at flow rate 300 nL/min. For the flow-through analysis a 90 min gradient 5–38% B was used. MS scans were acquired in the range of 375–1500 m/z with mass resolution of 120k, AGC 4 ×105 and max IT 50 ms. Precursors were selected with the top speed mode in 3 sec cycles and isolated for HCD fragmentation with quadrupole isolation width 0.7 Th. Collision energy was 36% with AGC 5 ×104 and max IT 86 ms at 50k resolution. Targeted precursors were dynamically excluded for further fragmentation for 30 s (or 45 s for flow-throughs) with 7 ppm mass tolerance.

Mass spectra were analysed in Proteome Discoverer 2.4 (Thermo Scientific) with the SequestHT search engine for peptide identification and quantification. The precursor and fragment ion mass tolerances were 20 ppm and 0.02 Da respectively. Spectra were searched for fully tryptic peptides with maximum 2 missed-cleavages. TMTpro at N-terminus/K and Carbamidomethyl at C were selected as static modifications. Oxidation of M, Deamidation of N/Q and Phosphorylation of S/T/Y were selected as dynamic modifications. Spectra were searched against reviewed UniProt Homo Sapiens protein entries, peptide confidence was estimated with the Percolator node and peptides were filtered at q-value < 0.01 based on target-decoy database search. The reporter ion quantifier node included a TMTpro quantification method with an integration window tolerance of 15 ppm. Only peptides with average reporter signal-to-noise >3 were used and phosphorylation localization probabilities were estimated with the IMP-ptmRS node. Statistical analysis of the proteomics data was further performed with the Perseus platform [51]. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD040422.

Sequencing of patient derived samples

Tumour samples included in this analysis were collected as part of the Royal Marsden Hospital (RMH) NHS Foundation Trust study: CCR3705 “Analysis of tumour specimens for biomarkers in gynaecological cancers” and were used in the validation of the ARID1A immunohistochemistry assay utilised for patient stratification in the ATARI clinical trial [52]. ARID1A mutations were identified using a targeted capture panel as described previously in [52], with raw targeted sequencing data deposited into the NCBI Sequence Read Archive under the accession PRJNA432413 and PRJNA432343 [52]. In this dataset, the median per-sample sequencing depth for ARID1A was 718 [52]. PPP2R1A genotyping was carried out by Sanger Sequencing. PPP2R1A PCR amplicons were generated using 100 ng genomic DNA in 50 μL reactions, using Q5 high-fidelity polymerase kit (New England Biolabs), according to manufacturer’s protocol. Primers used were 5’ACTGTTACTATCAGCTCCGTTTC 3’ (forward) and 5’ CTCATCTACCTCTGTGAACTTGTC 3’ (reverse). PCR was carried out on a thermocycler as follows: 98 °C for 30 s, followed by 25 cycles of: (i) 98 °C for 10 s (melting); (ii) 65 °C for 30 s (annealing); and (iii) 70 °C for 20 s. This was followed by 72 °C for 2 min (final extension). PCR products were analysed by agarose gel electrophoresis after mixing with 6× loading dye (New England Biolabs). Agarose gels were made by dissolving 1gram ultra-pure agarose (Life Technologies) in 100 mL 1× TAE buffer with1:10,000 GelRed nucleic acid stain (Biotium). Hyperladder 1 (Bioline) was used to estimate size of PCR product. PCR product was then purified using QIAquik PCR purification kit (Qiagen) and eluted in 50 μL nuclease free water (Ambion). For Sanger sequencing, 15 μL purified PCR product mixed with 2 μL PPP2R1A forward primer (10 μM). Sanger sequencing was performed by Eurofins Genomics and results were analysed using SnapGene software.