Modeling Therapy Resistance in BRCA1/2-Mutant Cancers

This article is featured in Highlights of This Issue, p. 1727

Heterozygous germline mutations in the BRCA1 or BRCA2 tumor suppressor genes strongly predispose to cancers of the breast, ovary, pancreas, and prostate (1, 2). BRCA1 and BRCA2 are involved in homologous recombination (HR), a process used to repair DNA double-strand breaks (DSB), and other DNA lesions that impair replication forks (3–6). Extensive preclinical and clinical data has established that the loss of BRCA1 or BRCA2 function is associated with sensitivity to small-molecule PARP inhibitors (7). Recently, the PARP inhibitor (PARPi) Lynparza (olaparib/AZD2281 – KuDOS/AstraZeneca) was approved for the treatment of platinum-responsive, BRCA1- or BRCA2-mutant high-grade serous ovarian carcinomas (HGSOC; ref. 8).

Despite a number of profound and sustained antitumor responses in patients treated with PARP inhibitors, drug resistance limits the overall effectiveness of these agents (9–12). A number of mechanisms of PARP inhibitor resistance have been identified, including upregulation of PgP drug transporters, loss of 53BP1 or REV7 function, or secondary “revertant” mutations within the BRCA1 or BRCA2 genes themselves (13, 14). These secondary BRCA gene mutations restore BRCA1 or -2 open reading frames and encode proteins that have partial function (13–16). In some BRCA2-mutant patients, initial clinical responses to PARPi are seen, followed by the emergence of profound PARPi-resistant lesions (13). The gradual emergence of PARPi resistance during treatment has led to the hypothesis that PARPi treatment might provide a Darwinian-selective pressure effect, where a secondary mutant clone has a selective advantage over nonsecondary mutant tumor clones, once PARPi treatment is applied (7, 17). Although this hypothesis has not as yet been tested, if such a Darwinian process did exist, a secondary mutant clone might be expected to gradually dominate a tumor cell population over the course of PARPi therapy. To date, only one approach for targeting tumor cell clones with secondary BRCA1/2 mutations has been proposed, namely the use of thiopurines (18). The wide application of thiopurines in the treatment of cancer has been limited by safety concerns (18), suggesting that additional therapeutic approaches for targeting secondary BRCA1/2-mutant tumor cells might also be required.

We set out to assess, both in vitro and in vivo, whether tumor cells with secondary BRCA1 or -2 gene mutations are selected for by PARPi treatment in a Darwinian fashion. To do this, we used CRISPR-Cas9–mediated gene targeting in BRCA1- or BRCA2-mutant tumor cells to generate daughter clones with secondary mutations. By mixing these secondary mutant clones with parental tumor cells in in vitro cocultures or in vivo tumor xenografts, we established that PARPi treatment can select for secondary mutant clones in a Darwinian fashion. Using these same systems, we also found that exposure to a clinical WEE1 kinase inhibitor (ref. 19; AZD-1775) minimized the selection of secondary mutant tumor cells, targeting parental and secondary mutant cells to a similar extent, while having minimal effects on nontumor cells.

CAPAN1 and SUM149 cells were obtained from ATCC. DLD1-BRCA2^WT/WT and DLD1-BRCA2^−/− cells were purchased from Horizon Discovery Inc. All cells were cultured according to the manufacturer's instructions. All cells were STR typed to confirm identity and verified to be mycoplasma-free (Lonza MycoAlert) prior to the study.

Olaparib, talazoparib, and AZD-1775 were obtained from Selleck Chemicals. Inhibitor stock solutions were prepared in DMSO and stored in aliquots at −20°C. Inhibitors were added to cell cultures so that final DMSO concentrations were constant at 1% (v/v).

CAPAN1.B2.S* and SUM149.B1.S* were generated from CAPAN1 and SUM149 parental tumor cell lines. Parental lines were transiently transfected with 5 μg of gRNA (described below) and 5 μg of a Cas9 pMA-T expression vector (GE Healthcare) using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Twenty-four hours later, cells were replated into 15-cm dishes at 2,000 cells/dish and exposed to 100 nmol/L talazoparib for two weeks after which clones were cultured in talazoparib-free media until visible. Clones were manually isolated, and replated into 96-well plates for expansion. DNA from clones was isolated using DNeasy Blood and Tissue Kit (Qiagen) and PCR amplified using BRCA1 or BRCA2 primers described below. PCR products were subcloned using TOPO TA Cloning Kit, with pCR2.1-TOPO (Invitrogen). Sanger sequencing confirmed secondary BRCA1 or BRCA2 mutations from 20 subcloned E.coli colony sequences per cell line.

gRNA (in pMA-T vector): BRCA2, 5′-GAGCAAGGGAAAATCTGTCC-3′ and BRCA1, 5′-CCAAAGATCTCATGTTAAG-3′. The BRCA2 gRNA contains the c.6174delT mutation specific to the CAPAN1 cell line.

Primers for PCR amplification were BRCA1, 5′-TGCTTTCAAAACGAAAGCTG-3′, 5′-ACCCAGAGTGGGCAGAGAA-3′; BRCA2, 5′-CTGTCAGTTCATCATCTTCC-3′, 5′-ATGCAGCCATTAAATTGTCC-3′.

Cells on glass coverslips were exposed to 5 Gy IR using an X-ray machine and then fixed and stained 5 hours later as described previously (20). Nuclei were stained using DAPI diluted in PBS (1:10,000 v/v), RAD51 using the Abcam (ab137323) antibody and γH2Ax using the Millipore (05-636) antibody. At least 100 cells were assessed per coverslip. Cells scoring positive had >5 foci per nucleus.

Immunoprecipitation (IP) and Western blotting were performed as described previously (15, 21). Antibodies targeting phospho-CDC (Y15; #4539), PARP1 (#9542), γH2AX (#9718; all Cell Signaling Technology), and tubulin (#T6074, Sigma) were employed in Western blots.

Short-term drug exposure and clonogenic assays were performed as described previously (22). In brief, cells were seeded either into 384-well or 6-well plates at a concentration of 500 to 2,000 cells per well. After 24 hours, cells were exposed to olaparib, talazoparib, or AZD-1775. For short-term drug exposure, cell viability was assessed after five days of drug exposure using CellTiter-Glo Luminescent Cell Viability Assay (Promega) as per the manufacturer's instructions. For clonogenic assays, drug was replenished every three days for up to 14 days, at which point colonies were fixed with TCA and stained with sulforhodamine B. Colonies were counted and surviving fractions calculated by normalizing colony counts to colony numbers in vehicle-treated wells. Survival curves were plotted using a four-parameter logistic regression curve fit as described in ref. 23.

Cells were plated in a fixed starting ratio of secondary mutant:parental cells in either 24 well or 6 well plates, or T75 flasks and exposed to either olaparib, talazoparib, or AZD-1775 for 14 or 21 days. In “constant drug exposure” experiments, media containing drug were replenished every three days. In “intermittent drug exposure” experiments, cells were exposed to media containing drug for 24 hours after which, media were removed, cells were washed using PBS, and then cells were recultured in media without drug for 48 hours, at which point cells were “refed” with media containing drugs as before.

DLD1-BRCA2^WT-GFP and DLD1-BRCA2^−/−-RFP cells were plated at one of the following ratios: 1:1, 1:10, 1:100, or 1:1,000. Twenty-four hours later, cells were exposed to DMSO, olaparib, or talazoparib. Aliquots of cell populations were analyzed by flow cytometry, every 3–4 days, (LSR II, Beckman-Coulter) for GFP and RFP cell populations. Drug was replenished every three days.

Cells were pelleted and genomic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen) following the manufacturer's instructions. DNA concentration was measured using Qubit broad range detection kit (Invitrogen). Restriction digestion was performed with EcoRI (BD Biosciences) and final working dilutions were made at 5 μg/uL per sample. DNA reaction mixtures were performed as described previously (24).

Primers and probes were as follows:

• CAPAN1:

  • ddPCR Probe: VIC-CTGGACAGATTTTC,

  • Forward primer: 5′- TCTCATCTGCAAATACTTGTGGGATT-3′

  • Reverse primer: 5′- TTGTCTTGCGTTTTGTAATGAAGCA-3′

• CAPAN1.B2.S*

  • ddPCR Probe: 6FAM- CTGATACCTGATTTTC

  • Forward primer: 5′- TCTCATCTGCAAATACTTGTGGGATT -3′

  • Reverse primer: 5′- TTGTCTTGCGTTTTGTAATGAAGCA -3′

• SUM149

  • ddPCR Probe: VIC- TTTGTCAACCTAGCCTTCCA

  • Forward primer: 5′- TGACAGCGATACTTTCCCAGA -3′

  • Reverse primer: 5′- GAGATCTTTGGGGTCTTCAGC -3′

• SUM149.B1.S*

  • ddPCR Probe: 6FAM-ACCAGGTGCATTTGTTAACTTCA

  • Forward primer: 5′- TGACAGCGATACTTTCCCAGA -3′

  • Reverse primer: 5′- GCAAAACCCTTTCTCCACTTACT -3′

A total of 146 cancer cell lines were profiled as described previously (22, 25). In brief, cells were plated at a density of 250 or 500 cells per well. Twenty-four hours later, media containing WEE1 inhibitor was added to adherent cells. After five days of drug exposure, cell viability was measured using CellTiter-Glo (Promega). Luminescence data was log2 transformed and centered according to the plate median value. Surviving fractions were calculated relative to the DMSO-exposed control wells to generate AUC data.

Female BALB/c nude mice aged 4–6 weeks and 15–22 g in weight (Charles River Laboratories) were inoculated subcutaneously with 5 × 10⁶ tumor cells into the right flank. When tumors reached 100 mm³, 6 animals from each cohort were sacrificed as sentinels to enable estimation of parental and secondary mutant tumor cell frequencies prior to treatment. Remaining animals were randomized into different treatment arms (n = 6) as described in the main text. Mice were weighed once weekly; tumors were measured twice weekly. When tumors reached 1,500 mm³, tumors were harvested and half of the tissue was formalin fixed, the other half was snap frozen in liquid nitrogen for DNA isolation.

Asynchronously growing CAPAN1 and CAPAN1.B2.S* cells were plated in 10-cm dishes (5 × 10⁵ cells/plate) and treated with 1 μmol/L AZD-1775 or DMSO for 72 hours. The cells were pulse labeled with 30 μmol/L bromodeoxyuridine (BrdUrd; Sigma, B5002) for 1 hour prior to collection. Cells were harvested using trypsin, washed with PBS, fixed in cold 70% ethanol and stored at −20°C overnight. Samples were washed with 2 mol/L NaCl/0.5% Triton X-100 and incubated for 30 minutes at room temperature. Cell pellets were resuspended in 0.1 mol/L sodium tetraborate for 2 minutes and subsequent cell pellets were incubated at room temperature for 1 hour in anti-BrdUrd antibody (BD Biosciences, 347580) diluted in 0.5% TWEEN-20/1%BSA/PBS (1 μg antibody per 1 × 10⁶ cells). Sample pellets were washed in PBS/1% BSA. Cells were then incubated for 30 minutes at room temperature with goat anti-mouse IgG FITC antibody (Sigma, F0257) diluted in 0.5% TWEEN-20/1%BSA/PBS (1 μg antibody per 1 × 10⁶ cells). Cells were pelleted and resuspended in PBS containing 10 μg/mL RNaseA (Sigma, R4642) and 20 μg/mL propidium iodide (Sigma, P-4170) and incubated at room temperature for 30 minutes. Cell-cycle analysis was performed on a FACS LSR II and analyzed using FlowJo software (FlowJo).

We used CRISPR-Cas9–mediated gene targeting to generate novel tumor cell models with secondary mutations in either BRCA1 or BRCA2 and then used these in in vitro and in vivo coculture systems to assess the clonal evolution of tumor cell populations in response to therapy (Fig. 1A). To generate these models, we used two PARPi-sensitive tumor cell lines, the pancreatic ductal adenocarcinoma tumor cell line, CAPAN1 (BRCA2 mutation c.6174delT, p.S1982fs*22; refs. 13, 15, 26), and the breast tumor cell line SUM149 (BRCA1-mutant c.2288delT, p.N723fsX13; ref. 27). We designed specific CRISPR guide RNA (gRNA) expression constructs targeting PAM (protospacer adjacent motifs) sequences close to either the BRCA2 c.6174delT mutation in CAPAN1 cells or the BRCA1 c.2288delT mutation in SUM149 cells, and transiently transfected these into cells, alongside a Cas9 expression construct. We reasoned that the error-prone repair of DSBs in these BRCA gene–defective tumor cell lines would in some cells cause frameshift secondary mutations in either BRCA1 or BRCA2 that restored the open reading frame. In a CAPAN1-derived daughter cell clone, CAPAN1.B2.S*, we identified a five base pair (bp) BRCA2 secondary mutations, c.[6174delT;6182del5], in addition to the parental c.6174delT mutation (c.6174delT: c.[6174delT;6182del5] allele ratio of 3:2; Fig. 1B; Table 1). The BRCA2 c.[6174delT;6182del5] secondary mutation in CAPAN1.B2.S* was predicted to restore the open reading frame of the gene and to encode a 3612 amino acid (aa) BRCA2 protein (Fig. 1C), confirmed by Western blotting (Supplementary Fig. S1A). The secondary mutation in CAPAN1.B2.S* was associated with olaparib and also talazoparib resistance (Fig. 1D; Supplementary Fig. S1B, ANOVA P < 0.0001) and the ability to form nuclear RAD51 foci in response to ionizing radiation (Student t test P < 0.001; Fig. 1E and F; Supplementary Fig. S1C), a biomarker of functional DNA repair by BRCA2. The CAPAN1.B2.S* clone also exhibited a similar doubling time to the parental CAPAN1 clone, of 2.5 days (Supplementary Fig. S1D). Using the same approach, we also identified other CAPAN1 daughter clones with secondary mutations, including CAPAN1.B2.S*2, which had three different BRCA2 alleles (BRCA2 c.[6174delT;6185del5], BRCA2 c.[6174delT;6183delG], and BRCA2 c.[6174delT;6184delTC]) and CAPAN1.B2.S*3, which also had three different BRCA2 alleles (BRCA2 c.[6174delT;6183del6], BRCA2 c.[6174delT;6183del5], and BRCA2 c.[6174delT;6185del3]; Supplementary Table S1).

Using a similar approach in SUM149 cells, we identified SUM149.B1.S*, a daughter clone which possessed a secondary mutation (an 80-bp BRCA1 deletion, c.[2288delT;2293del80]), as well as the parental c.2288delT mutation (c.2288delT: c.[2288delT;2293del80] alleles in a 1:2 ratio; Fig. 1G; Table 2). This secondary mutation was predicted to restore the open reading frame in the parental BRCA1 c.2288delT allele and to encode an 1836 aa BRCA1 protein (Fig. 1H; Supplementary Fig. S2A). SUM149.B1.S* exhibited both olaparib and talazoparib resistance (Fig. 1I; Supplementary Fig. S2B, ANOVA P < 0.0001), and restoration of RAD51 nuclear localization in response to DNA damage (Fig. 1J; Supplementary Fig. S2C; Supplementary Fig. S2D, P < 0.01, Student t test). SUM149 and SUM149.B1.S* cells exhibited similar proliferation rates (Supplementary Fig. S2E).

To assess whether a Darwinian-selective process might operate in the case of PARPi resistance, we mixed parental and secondary mutant tumor cells in in vitro cocultures (i.e., CAPAN1 parental with CAPAN1.B2.S* secondary mutant cells, or SUM149 with SUM149.B1.S* cells) and then exposed the cocultures to two different BRCA gene–selective drugs, olaparib or the platinum salt cisplatin (Fig. 2A). We then monitored the relative frequency of each clone in response to drug exposure using droplet digital PCR (ddPCR; ref. 28). To do this, we used duplex PCR reactions that included fluorophore-labeled digital PCR probes that were complementary to either parental or secondary mutant alleles (Supplementary Table S2, see Materials and Methods). In pilot experiments, where we mixed CAPAN1 and CAPAN1B2.S* cells in 1:1, 1:10, 1:100 ratios, we were able to accurately detect these different ratios using the ddPCR assay (Supplementary Fig. S3). Using this ddPCR approach, we assessed whether olaparib or cisplatin exposure would preferentially select for the secondary mutant clones in either CAPAN1 plus CAPAN1B2.S* cocultures or SUM149 plus SUM149.B1.S* cocultures. In these experiments, we exposed cocultures to either: (i) DMSO (the drug vehicle), (ii) constant exposure to olaparib (drug replenished every three days), (iii) intermittent exposure to olaparib (where drug was applied for 24 hours then washed out and replenished with media not containing drug for 48 hours), (iv) constant exposure to cisplatin (drug replenished every three days), or (v) intermittent 24-hour pulses of cisplatin (where drug was applied for 24 hours then washed out and replenished with media not containing drug for 48 hours; Fig. 2A). As expected, in both CAPAN1 and SUM149 cocultures, constant exposure to either olaparib or cisplatin caused a greater reduction in the total tumor cell population size than intermittent drug exposure (Fig. 2B). For example, in the CAPAN1 coculture, 37% of the cell population survived after 14 days when exposed to constant olaparib, compared with 55% when intermittent drug exposure was used (Fig. 2B). Despite these reductions in population size, both olaparib and cisplatin exposure caused an increase in the relative frequency of the secondary mutant clones compared with the parental clone (Fig. 2C), effects replicated when mixed cultures were exposed to a chemically distinct PARP inhibitor, talazoparib (Supplementary Fig. S4A). We found that the enrichment in the secondary mutant clones compared with the parental clones was most profound when cultures were constantly exposed to either olaparib or cisplatin, compared with intermittent drug exposures (Fig. 2C). In cultures exposed to the drug vehicle, the proportion of CAPAN1.B2.S* and SUM149.B1.S* in DMSO-exposed cultures remained the same throughout the experiment (1:20 secondary mutant:parental clone, Supplementary Fig. S4B). We therefore concluded that while constant drug exposure elicited a more profound reduction in the size of the tumor cell population, it did enrich for secondary mutant clones.

We then assessed whether the initial frequency of a secondary mutant clone in a tumor cell population might influence the time taken for this clone to dominate the population when it was exposed to the selective pressure of PARPi therapy. To do this, we generated CAPAN1.B2.S*: CAPAN1 mixed in vitro cultures with 1:1, 1:20, and 1:40 clone ratios and then exposed these to olaparib. We then estimated the temporal evolution of the culture in response to drug treatment by using ddPCR to measure CAPAN1.B2.S*: CAPAN1 ratios over time (Fig. 2D). We found that in each culture, olaparib exposure caused an increase in the frequency of the secondary mutant clone compared with DMSO-exposed cultures, with the fraction of CAPAN1.B2.S*cells in each PARPi exposed culture gradually increasing over time (Fig. 2D). We also found that the initial frequency of the secondary mutant clone prior to drug treatment influenced the ability of the secondary mutant clone to eventually dominate the population (i.e., >75% of the cell population) once cells were exposed to PARP inhibitor, as might be expected of a Darwinian process (Fig. 2D, compare 1:1, 1:20, and 1:40 ratio cultures).

We also used a different model system, isogenic DLD1 tumor cell lines with or without targeted mutations in BRCA2 (DLD1.BRCA2^WT/WT and DLD1.BRCA2^−/− (29, 30), to validate these observations. We labelled DLD1.BRCA2^WT/WT cells with a GFP and DLD1.BRCA2^−/− cells with a red fluorescent protein (RFP) marker to enable detection and monitoring of coculture populations via FACS (Supplementary Fig. S5A). We found that in the absence of drug exposure, the DLD1.BRCA2^WT/WT cells exhibited a selective advantage over DLD1.BRCA2^−/−cells, as previously shown (ref. 29; Supplementary Fig. S5B), and that these cells exhibit more than a 10-fold difference in olaparib sensitivity (Fig. 3A). We then mixed DLD1.BRCA2^WT/WT cells into DLD1.BRCA2^−/− cells in vitro at starting ratios of 1:1, 1:10, 1:100, and 1:1,000, exposed these cocultures to either olaparib or talazoparib, and monitored the temporal evolution of the population in response to PARPi (Supplementary Fig. S5A). Similar to the CAPAN1 and SUM149 isogenic models, we observed that olaparib and talazoparib both selected for DLD1.BRCA2^WT/WT cells over DLD1.BRCA2^−/− cells in a Darwinian fashion (Fig. 3B). For example, both olaparib and talazoparib exposure resulted in a 3-fold increase in DLD1.BRCA2^WT/WT cells compared with the DMSO-exposed cell population after 13 days of drug exposure (Fig. 3B). In addition, we noticed that the time taken for the DLD1.BRCA2^WT/WT clone to reach clonal dominance was less in cell populations that had higher starting proportion of DLD1.BRCA2^WT/WT cells (Fig. 3C–E), as observed in the CAPAN1 coculture model.

We also assessed whether a Darwinian process influenced the in vivo response to PARPi treatment. To do this, we generated cohorts of mice bearing subcutaneous xenografts consisting of a mixture of CAPAN1 parental and CAPAN1.B2.S* secondary mutant tumor cells (Fig. 4A). We found that inoculating 5 × 10⁶ tumor cells at a 1:1 CAPAN1:CAPAN1.B2.S* ratio reproducibly generated 100 mm³ xenografts 10 days after innoculation, where each clone was present in equal proportion (Fig. 4B). When tumors reached 100 mm³, tumor-bearing mice were randomized into the following treatment cohorts to assess the selective pressure of PARPi treatment in vivo: (i) olaparib (50 mg/kg) administered once daily, (ii) olaparib (50 mg/kg) administered every other day, (iii) olaparib (50 mg/kg) administered twice a week on days 1 and 4, (iv) drug vehicle administered daily. In addition, sentinel mice were sacrificed prior to treatment so that the CAPAN1:CAPAN1.B2.S* ratio in tumors prior to therapy could be confirmed (Fig. 4A; Supplementary Fig. S6A). We found that 50 mg/kg olaparib treatment, administered daily, every other day, or twice weekly, though well-tolerated, did not decrease tumor growth compared with the vehicle (P > 0.05 ANOVA for tumor volume in each olaparib treatment cohort versus vehicle; Supplementary Fig. S6B and S6C). We hypothesized that the absence of overall antitumor efficacy in this particular case might be due to failure to inhibit the PARPi secondary mutant clone in xenografts. To test this, we isolated tumor DNA from olaparib-treated mice (after 28-day treatment) and assessed the relative ratio of parental versus secondary mutant clones by ddPCR. In mice that received drug vehicle alone, the ratio of parental versus secondary mutant clones remained approximately 50% (Supplementary Fig. S6A). However, in mice that received olaparib treatment, the relative frequency of CAPAN1.B2.S* cells increased in response to therapy (Fig. 4C). This increase in CAPAN1.B2.S* frequency, in preference to the parental clone, was dependent upon the periodicity of PARPi administration, for example, daily administration of olaparib caused the greatest increase in CAPAN1.B2.S* enrichment, followed by every other day treatment and then biweekly administration (Fig. 4C and D). This suggested that PARPi administration also selected for secondary mutant tumor cell clones in vivo and that the degree of secondary mutant clone selection was related to the extent of selective pressure applied.

The coculture model systems described above allowed us to establish that PARPi resistance, when driven by secondary mutations in BRCA1 or BRCA2, can operate along Darwinian principles. We also assessed whether we could identify therapeutic vulnerabilities that would allow targeting of both parental and secondary mutant tumor cell clones as a means to minimize the impact of secondary mutation. We assessed whether small-molecule WEE1 cell-cycle checkpoint kinase inhibitors (WEE1i; ref. 19) might have utility in this regard. We focused on WEE1 inhibitors for a number of reasons. WEE1 prevents premature mitotic entry by phosphorylating and inhibiting cyclin-dependent kinases such as cyclin dependent kinase 1 (CDK1; refs. 31, 32). This activity is particularly critical in tumor cells with p53 pathway defects; p53 defects often cooccur with BRCA mutations, and although secondary mutations in BRCA1/2 drive PARPi resistance, resistant tumors and cell lines remain p53 mutant (13). CAPAN1.B2.S* and SUM149.B1.S* clones retained the p53 mutations present in CAPAN1 and SUM149 parental tumor cell clones (Supplementary Figs. S7 and S8). We also found that in an analysis of in vitro sensitivity to the clinical WEE1 kinase inhibitor, AZD-1775, in a panel of tumor cell lines, CAPAN1.B2.S* and SUM149.B1.S* were among the most sensitive of 146 lines profiled (Fig. 5A). We confirmed this AZD-1775 sensitivity in subsequent clonogenic survival experiments and found that, when compared with nontumor breast epithelial cell lines (MCF10A and MCF12A), both CAPAN1 and SUM149-derived secondary mutant tumor cell clones retained profound sensitivity to AZD-1775 seen in parental tumor cells (average 22-fold difference in AUC, P < 0.0001 vs. MCF10A or MCF12A, ANOVA; Fig. 5B). We confirmed these observations using coculture systems and found that at SF₅₀ concentrations (concentration required to inhibit 50% of cells) of either olaparib or AZD-1775, olaparib exposure increased the relative frequency of the secondary mutant clones, but AZD-1775 did not (Fig. 5C). This observation was confirmed when we used ddPCR to monitor the frequency of the secondary mutant clone over time in cocultures exposed to AZD-1775 (Fig. 5D). We also observed that parental and secondary mutant SUM149 and CAPAN1 clones were sensitive to additional small-molecule cell-cycle checkpoint inhibitors including PF-477736, a CHK1 inhibitor (33, 34), and VX-970, an ATR inhibitor (35) when compared with nontumor epithelial cells (Supplementary Fig. S9A and S9B). This suggested that even when partial BRCA1 or BRCA2 protein function was restored by secondary mutation, vulnerability to small-molecule inhibitors that target cell-cycle checkpoints still existed. These effects did not appear to represent a relatively nonspecific sensitivity to cytotoxic agents in the parental and secondary mutant tumor cells, as these did not display an overtly distinct level of sensitivity to paclitaxel, capecitabine, or gemcitabine when compared with MCF10 or MCF12A cells (Supplementary Fig. S9C–S9F).

Previous studies have shown that WEE1 inhibitors cause tumor cell cytotoxicity by reducing the extent of CDC2 phosphorylation at Y15 (36). We found that in both CAPAN1 and CAPAN1.B2.S* cells, AZD-1775 exposure caused a decrease in CDC2 Y15 phosphorylation, an effect that was enhanced with prolonged drug exposure (Fig. 5E). We noted that AZD-1775 exposure caused an increase in H2AX phosphorylation (γH2AX), a biomarker of DNA damage, in both CAPAN1 and secondary mutant CAPAN1.B2.S* cells (Fig. 5E). This increase in γH2AX was commensurate with an increase in PARP cleavage, a measure of apoptosis (Fig. 5E). Using FACS profiling, we found that AZD-1775 exposure had a very similar effect on cell-cycle fractions in both CAPAN1 and CAPAN1.B2.S* cells, both of which demonstrated a profound reduction in the fraction of cells in active S-phase, with a commensurate increase in the proportion of cells in nonreplicating S phase (Supplementary Fig. S10). In CAPAN1 cells, AZD-1775 exposure caused a reduction in the active S-phase fraction from 25.9% to 3.4% (with a 3.9%–52.1% increase in nonreplicating S-phase), while CAPAN1.B2.S* cells showed a reduction in active S-phase from 27.8% to 4.2% (with a 3.3%–51.4% increase in nonreplicating S-phase). These observations were reminiscent of those seen in H3K36me3-deficient cells, where WEE1 inhibition also caused a severe reduction in the active S-phase fraction (37). This suggested that WEE1 inhibition targeted CAPAN1 cells in S-phase, regardless of whether BRCA2 was dysfunctional (as in CAPAN1) or somewhat reconstituted by the presence of a secondary BRCA2 mutation (as in CAPAN1.B2.S*).

To investigate whether WEE1 inhibitor sensitivity in PARPi-sensitive and resistant clones also operated in vivo, we assessed the effect of AZD-1775 treatment on mice bearing mixed CAPAN1/CAPAN1.B2.S* xenografts (each clone present at a 1:1 ratio, Fig. 5F). Mice with established tumors were treated with either AZD-1775, olaparib, or drug vehicle. Sentinel mice sacrificed prior to treatment showed the CAPAN1:CAPAN1.B2.S* ratio in tumors prior to therapy was 1:1 (Fig. 5G). We used the time taken for tumors to reach 1,500 mm³ as a surrogate measure of survival (Fig. 5H) and found that while olaparib treatment had minimal benefit (P = 0.86, log-rank Mantel–Cox test compared with vehicle), AZD-1775 treatment led to a significant survival benefit (P = 0.011, log-rank Mantel–Cox test compared with olaparib; Fig. 5I). Consistent with these observations, ddPCR analysis of tumors at the end of treatment showed that olaparib therapy caused a relative enrichment in the frequency of the secondary mutant clone (P = 0.058 compared with vehicle, Student t test) while AZD-1775 did not (P = 0.43, compared with vehicle, Student t test; Fig. 5J).

In this study, we used CRISPR-generated BRCA1 or BRCA2 secondary mutant daughter clones alongside isogenic parental cell lines to demonstrate that PARPi exposure selects for secondary mutant clones in a Darwinian manner, both in vitro and in vivo. We found that the extent of selection for secondary mutant clones was influenced by the frequency of drug administration. In mice bearing tumors comprised of an equal proportion of BRCA2-mutant and BRCA2 secondary mutant tumor cells, olaparib had minimal effects on tumor growth, but did preferentially select for the secondary mutant daughter clone over the parental tumor cell. It would be reasonable to infer that high frequencies of secondary mutant cells hinder the therapeutic effectiveness of PARP inhibitors. We also found that a WEE1 inhibitor, AZD-1775, had a greater therapeutic effect on mixed parental/secondary mutant tumors than olaparib. This example suggests that therapeutic vulnerabilities might still exist in tumors that have a high frequency of secondary mutant clones. Our data also suggest that secondary mutant and parental tumor cells also show sensitivity to other cell cycle/DNA damage repair inhibitors, including CHK1 and ATR inhibitors (Supplementary Fig. S9). It seems possible that while secondary BRCA1 or BRCA2 gene mutations restore some HR function, these are unlikely to reverse the complex set of genomic rearrangements, aneuploidy, and p53 mutations found in BRCA1 or BRCA2 mutant tumors prior to treatment (38). We hypothesize that these latter characteristics sensitize tumor cells to drugs such as WEE1 inhibitors, perhaps explaining why AZD-1775 targets both parental and secondary mutant clones. This hypothesis remains to be tested, but the observation that secondary mutant tumor cells are sensitive to AZD-1775 raises the possibility that therapeutic vulnerabilities still exist in PARPi-resistant tumors.

In clinical studies, the MTD for single-agent AZD-1775 was identified as 225 mg twice per day orally over 2.5 days per week for 2 weeks per 21-day cycle, a dosing regimen sufficient to elicit a number of antitumor responses (39). In our in vivo studies (Fig. 5) we used 30 mg/kg AZD-1775 twice-daily treatments for the entire duration of the study (150 days). This treatment approach was well tolerated in mice and based on prior mouse-based experiments using this WEE1 inhibitor (40). Nevertheless, it is possible that a similar constant dosing approach may not be well-tolerated in humans. Subsequent work might assess the potential of using intermittent WEE1 inhibitor dosing schedules to assess whether these also elicit a survival benefit in experiments similar to those shown in Fig. 5.

One implication of this work is that the detection of secondary BRCA1 or BRCA2 mutations in patients could be important in influencing the choice of therapy. At present, secondary mutations in BRCA1 or BRCA2 can be detected by Sanger DNA sequencing (14–16) or by targeted DNA capture and deep sequencing (13). Circulating tumor DNA and circulating tumor cells might also display some of the secondary BRCA1 or BRCA2 mutations found in solid tumors. Detecting secondary mutations in such liquid biopsies might allow the early emergence of secondary mutations to be identified as a biomarker predicting the eventual clinical manifestation of PARPi resistance. One avenue we will now explore is to utilize the in vivo system we have described here to assess this possibility. A key quality of the model systems described here is that they allow the construction of cocultures and xenografts where the frequency and identity of secondary mutants is known. This will hopefully facilitate experiments that aim to examine further principles that govern clonal evolution and influence drug resistance in BRCA1- or BRCA2-mutant cancers. Alongside these models, we also note that the first PDX with PARPi resistance-causing mutations have been recently described (41). These provide another system in which to assess how the clonal structure of tumors evolve in response to therapy. The combined use of engineered systems, such as that described here, alongside PDX systems will be critical in establishing what factors determine the response to treatment, and importantly, what therapeutic approaches could be taken to minimize the impact of secondary BRCA1/2 gene mutations.

N.C. Turner reports receiving commercial research support from AstraZeneca and is a consultant/advisory board member for AstraZeneca. A. Ashworth provided expert testimony for AstraZeneca. C.J. Lord has received speakers bureau honoraria from AstraZeneca and Vertex and is a consultant/advisory board member for Vertex and Sun Pharma. A. Ashworth and C.J. Lord are named inventors on patents describing the use of PARP inhibitors and stand to gain from their use as part of the ICR “Rewards to Inventors” scheme. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Dréan, C.T. Williamson, A. Ashworth, C.J. Lord

Development of methodology: A. Dréan, C.T. Williamson, I. Garcia-Murillas, C.J. Lord

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Dréan, R. Brough, I. Brandsma, M. Menon, A. Konde, H.N. Pemberton, J. Frankum, N. Badham, C.J. Lord

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Dréan, C.T. Williamson, A. Konde, I. Garcia-Murillas, H.N. Pemberton, J. Campbell, A. Gulati, N.C. Turner, C.J. Lord

Writing, review, and/or revision of the manuscript: A. Dréan, C.T. Williamson, N.C. Turner, S.J. Pettitt, A. Ashworth, C.J. Lord

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Menon, A. Konde, R. Rafiq, C.J. Lord

Study supervision: C.T. Williamson, A. Ashworth, C.J. Lord

This work was funded by Breast Cancer Now and Cancer Research UK as part of Programme Grants (to C.J. Lord). We thank the Tumour Profiling Unit (TPU) at the Institute of Cancer Research for carrying out DNA sequence analysis. We acknowledge NIHR funding to the Royal Marsden Biomedical Research Centre.

This work was supported by Breast Cancer Now (CTR-Q4-Y2; to C.J. Lord) and Cancer Research UK (CRUK/A14276; to C.J. Lord).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.