Acquired RAD51C Promoter Methylation Loss Causes PARP Inhibitor Resistance in High-Grade Serous Ovarian Carcinoma

Women diagnosed with high-grade serous ovarian carcinoma (HGSC) are generally responsive to standard-of-care platinum-based chemotherapy initially, and in those with inherited mutations in BRCA1/2, the addition of PARP inhibitors (PARPi) can significantly improve clinical outcomes (1, 2). However, the development of resistance is common and limits long-term effectiveness of these therapies (3). Although the impacts of deleterious homologous recombination repair (HRR) gene mutations (e.g., in BRCA1, BRCA2, and RAD51C) on platinum and PARPi responses in HGSC are well recognized, consequences of HRR gene promoter methylation in these contexts have remained controversial (4–8).

HRR gene methylation has been shown to correlate with reduced gene expression for the BRCA1 and RAD51C genes, and is usually found to be mutually exclusive of deleterious BRCA1/2 mutations (7–12). Preclinical studies have provided some evidence that silencing of the BRCA1 and RAD51C genes by methylation results in HRR deficiency (HRD; refs. 13–15). Importantly, genomic scarring related to defective HRR and BRCA1/2 mutations has also been detected in breast and ovarian cancers with BRCA1 methylation (meBRCA1) and RAD51C methylation (meRAD51C; refs. 11, 16, 17). However, on the basis of mixed outcomes in several PARPi clinical studies that included screening for HRR gene methylation, it has remained unclear whether meBRCA1 and meRAD51C predicts PARPi response (7–9, 11, 12, 18).

Our group recently determined, using a cohort of HGSC patient-derived xenograft (PDX) models, that the zygosity of meBRCA1 is a key factor for determining PARPi and platinum response in HGSC (19). This discovery was made possible by the fact that only human tumor cells are propagated as PDX, excluding normal stroma and other nontumor human cells (20), thus enabling accurate determination of zygosity in the tumors. Limitations of previous studies had included the use of nonquantitative methods, such as methylation-specific PCR (MSP; refs. 7, 8), as well as the presence of contaminating nontumor tissue, which made it challenging to interpret the zygosity of meBRCA1 (6). Our study demonstrated that in a tumor, where multiple copies of BRCA1 may be present, all copies must be methylated for PARPi response, and that losing methylation of a single BRCA1 copy was sufficient to restore HRR DNA repair and cause PARPi resistance (19).

Although we are now beginning to better understand meBRCA1 stability under platinum or PARPi treatment pressure, little is known about meRAD51C in HGSC. The RAD51C protein is critical for HRR, playing a pivotal role in the resolution of Holliday junctions during recombination. RAD51C is a known ovarian cancer susceptibility gene (21–23), and recently it was shown that deleterious mutations in this gene cause PARPi sensitivity, whereas secondary reversion mutations that restore function of at least a single RAD51C copy cause PARPi resistance (24).

MeRAD51C is found in approximately 2% of HGSC (6, 18), and its effects on PARPi response remain unclear (8). This uncertainty is likely due to the same limitations encountered when studying meBRCA1, such as the ability of a single demethylated copy to cause therapy resistance (19) and the difficulties in accurately assessing methylation zygosity (i.e., whether all copies of the gene are methylated) in the presence of contaminating stroma using qualitative or semiquantitative assays (19).

We hypothesized that, as for meBRCA1 HGSC (19), meRAD51C zygosity could influence PARPi response in HGSC and that homozygous meRAD51C could become heterozygous under PARPi or platinum treatment pressure, thus leading to therapeutic resistance. Here we use several HGSC PDX models to characterize the impacts of meRAD51C on PARPi treatment and then to align these outcomes with those of a cohort of patients with meRAD51C HGSC.

List of reagents available in Supplementary Materials and Methods.

All experiments involving animals were performed according to the animal ethics guidelines and were approved by the Walter and Eliza Hall Institute of Medical Research (WEHI) Animal Ethics Committee. Ovarian carcinoma PDX were generated from patients enrolled in the Australian Ovarian Cancer Study or Stafford Fox Rare Cancer program. Informed written consent was obtained from all patients, and all experiments were performed according to the human ethics guidelines. Additional ethics approval was obtained from the Human Research Ethics Committees at the Royal Women's Hospital, the WEHI and QIMR Berghofer (P3456 and P2095).

HGSC PDX models #183, #1240, and PH039 were established by transplanting fresh fragments (#183, #1240) or viably frozen minced tumor (PH039) subcutaneously into NOD/SCID IL2Rγnull recipient mice (T1, passage 1) as described previously (25). Patient details are available in the Supplementary Materials and Methods.

Recipient mice bearing T2 to T6 (passage 2 to passage 6) tumors were randomly assigned to treatments when tumor volume reached 180 to 300 mm³ (PDX #183 and #1240) or 180 to 250 mm³ (faster growing PDX PH039). In vivo cisplatin treatments were performed as described previously (25). The regimen for rucaparib treatment was oral gavage once daily (Monday to Friday) for 3 weeks at 150, 300, or 450 mg/kg. The regimen for niraparib treatment was oral gavage once daily (Monday to Friday) for 3 weeks at 100 mg/kg. Tumors were harvested once tumor volume reached 600 to 700 mm³ or when mice reached ethical endpoint. Rucaparib retreatment experiments in PDX #183 involved standard 300 mg/kg treatment of rucaparib followed by an additional 2 weeks of 300 mg/kg rucaparib treatment (oral gavage once daily, Monday to Friday) once tumor volume reached 500 mm³. Tumors were then transplanted into a second recipient mouse where the treatment was repeated for a total of four treatment cycles (two per mouse). Rucaparib retreatment experiments in PDX PH039 involved standard 450 mg/kg treatment of rucaparib followed by an additional 2 weeks of 450 mg/kg rucaparib treatment (oral gavage once daily, Monday to Friday) once tumor volume reached 300 mm³. Tumors were then transplanted into a second recipient mouse where the treatment was repeated for a total of four treatment cycles (two per mouse). Nadir, time to progression (TTP or PD), time to harvest (TTH), and treatment responses are as defined previously (19).

Automated IHC staining was performed on the DAKO Omnis (Agilent Pathology Solutions) to confirm HGSC characteristics of PDX.

PDX tumor protein lysates were prepared using two-dimensional lysis buffer containing protease inhibitor PMSF (7M urea, 2M Thiourea, 4% CHAPS in 0.1M Tris buffer with 4.6% 100 mmol/L PMSF). Equal protein loads of 8 μg were resolved on precast 8% to 16% Bis-Tris gels (Bio-Rad) under reducing conditions. Protein was transferred to PVDF membranes using Bio-Rad Criterion wet transfer method (Bio-Rad), then probed with anti-RAD51C (Santa Cruz Biotechnology, sc-56214) followed by peroxidase-labeled secondary antibody (SouthernBiotech, 1010-05). Membranes were then stripped and probed with anti-β-actin (Abcam), followed by peroxidase-labeled secondary antibody (Santa Cruz Biotechnology, sc-2020) for loading control. Blots were visualized by enhanced chemiluminescence (Amersham ECL Prime, GE Healthcare Life Sciences) according to manufacturer's instructions.

DNA was extracted from tumor samples using the QIAamp DNA Mini Kit (Qiagen) according to manufacturer's instructions. Two hundred fifty nanograms of DNA was bisulfite converted (EZ Methylation Direct Kit; Zymo Research) and evaluated with MSP for RAD51C as described previously (11). Briefly, the methylated reaction produces a product of 85 nucleotides using forward primer 5′-TGTAAGGTTCGGAGTTTCGTGC-3′ and reverse primer 5′-TCGCTAAAACGTACGACGTAACG-3′. The unmethylated reaction product is 103 nucleotides using forward primer 5′-GTGTAAAGTTGTAAGGTTTGGAGTTTTGTGTG-3′ and reverse primer 5′-CACACACCCTCACTAAAACATACAACATAACA-3′.

Patterns of meRAD51C were assessed by MS-HRM as previously described (26) on the CFX (Bio-Rad) or MIC (Bio Molecular Systems) thermocycler platforms, using primers targeting the RAD51C promoter across genomic region chr 17:56,769,849–56,769,990 (hg19; ref. 26).

The primers used in RAD51C-targeted MS-HRM were modified by the addition of Illumina sequencing adaptors to create an NGS-based assay that would allow interrogation of individual CpG sites and amplicons (alleles) within meRAD51C PDX samples, shown in Supplementary Materials and Methods Table S1. Sequence data have been deposited at the European Genome-phenome Archive (EGA), under accession number EGAS00001005395. Details available in Supplementary Materials and Methods.

BROCA-Cancer Risk panel sequencing was performed and analyzed as previously described (27) using version BROCA-HR v7 of the panel (http://tests.labmed.washington.edu/BROCA).

Gene expression was analyzed using the TaqMan RNA-to-C_T 1-Step Kit (Applied Biosystems) per manufacturer's instructions. Details available in Supplementary Materials and Methods.

Dissociated PDX tumors were stained for DNA repair foci and analyzed using methods outlined in the Supplementary Materials and Methods.

Methylation data from publicly available 450K array data (GEO accession no. GSE65820) for AOCS patient samples were analyzed using the Miss Methyl package in R (28). The CpG sites used to assess meRAD51C were based on the RAD51C promoter region covered by the MS-HRM assay: (cg05214530 is NM_058216.2:c.-114/chr17:g.56769891 and cg27221688 is NM_058216.2:c.-101/chr17:g.56769904).

Four samples from the AOCS cohort were identified to have meRAD51C using publicly available methylation array data (three samples were previously reported as methylated; ref. 29). We were able to source tumor and germline DNA for three of the four cases (AOCS-106, AOCS-143, and AOCS-120). We also sourced tumor and germline DNA from an additional nine previously published meRAD51C samples (LS samples; ref. 8). DNA from these samples was tested by MS-HRM and targeted meRAD51C bisulfite sequencing (for quantitative meRAD51C assessment), as well as SNP arrays to determine tumor purity and RAD51C gene copy number.

SNP arrays were performed on the Global Screening Array-24 v2–0 (Illumina) and scanned on an iScan (Illumina) as described previously (29). Data processing details are available in the Supplementary Materials and Methods.

The following formula was used to calculate the adjusted (tumor) meRAD51C:

where total copy number is

where AdjMe% is the adjusted methylation, obsMe% is the observed methylation, TumCN is the tumor copy number, TotalCN is the total copy number, TumCell is the tumor cellularity, and Me is the meRAD51C.

Publicly available RNA sequencing (RNA-seq) data for AOCS patient samples (29) were analyzed using the standard pipeline. Details are available in the Supplementary Materials and Methods.

HRD scores were assessed on SNP array data using the scarHRD package on the allele-specific copy number calls. The HRD sum score cut-off of ≥42 was used to define HRD, based on the previous reports in breast and ovarian cancer (30–32).

PDX #183 was derived from chemo-naïve neoplastic cells obtained from a woman diagnosed with stage III, HGSC. The PDX was confirmed to be HGSC at transplant 3 (T3) by histopathologic and IHC review (Supplementary Fig. S1). This model was found to be cisplatin sensitive (P < 0.0001; Fig. 1A; Supplementary Table S1), reflecting the patient's response to first-line platinum chemotherapy (>11 months from cessation of platinum chemotherapy until time of first progression). This PDX also demonstrated response to the PARP inhibitors, rucaparib and niraparib (Fig. 1B and C; Supplementary Table S1), with a delay in median time to progression (TTP) of 8 days versus 47 days for rucaparib 450 mg/kg (P < 0.0001) and 8 days versus 50 days for niraparib (P < 0.0001; Supplementary Table S1). A nonsense TP53 mutation c.661G>T (p.E221*) was found by targeted sequencing using BROCA (27), but no pathogenic HRR gene mutations were detected (Supplementary Table S2). A heterozygous missense PARP1 mutation c.2008C>A (p.Q670K) in the regulatory domain was also detected in this PDX, but the effects of this mutation on PARPi activity are unclear (33). Given the observed in vivo PARPi response, we tested this model for both meBRCA1 and meRAD51C. MeBRCA1 was not detected in this model. MeRAD51C was detected initially by a nonquantitative method, MSP (Fig. 1D), and was confirmed to be homozygous using MS-HRM (Fig. 1E). PDX #183 had no detectable RAD51C gene expression by RNA-seq (Fig. 1F) or protein expression by immunoblotting (Fig. 1G) when compared with RAD51C wild-type controls PDX 13 and cell line PEO4, confirming that the homozygous meRAD51C was, indeed, causing gene silencing and thus likely causing the susceptibility to PARPi in this PDX model. Thus, PDX #183 was a cisplatin-sensitive and PARPi-responsive HGSC PDX model with meRAD51C and RAD51C gene silencing and consequent protein loss.

PDX #1240 was derived from chemo-naïve neoplastic cells obtained from the ovary of a woman diagnosed with stage IIIC HGSC. The PDX was confirmed to be HGSC at T1 by histopathologic and IHC review (Supplementary Fig. S1). Like #183, this model was found to be highly cisplatin sensitive (P <0.0001) and responsive to PARPi rucaparib (P = 0.013; Supplementary Table S1; Supplementary Fig. S2A). As the patient tumor sample was found to lack mutations in any core HRR genes (Supplementary Table S2), the PDX model was tested for HRR gene methylation, where homogeneous meRAD51C was detected on MS-HRM, whereas BRCA1 methylation was absent (Supplementary Figs. S2B and S2C), and this was associated with undetectable RAD51C gene expression (Supplementary Fig. S2D).

The initial characterization of the PDX model PH039 has been published (34). PH039 was re-established within our laboratory for further meRAD51C characterization and comparison with meRAD51C PDX #183.

Cryopreserved minced neoplastic material from the original PH039 PDX was transplanted into several NSG mice, and two PDX lineages derived from the initial transplant (T1) were used in this study: Lineage A (PH039-A) and Lineage B (PH039-B; Fig. 2A; Supplementary Fig. S1). Although both of these lineages were responsive to PARPi, PH039-A lineage was consistently less responsive to PARPi compared with PH039-B. This was observed for treatments with rucaparib 450 mg/kg (TTP: PH039-A = 50 vs. PH039-B = 71 days, P = 0.0004), and niraparib 100 mg/kg (TTP: PH039-A = 43 vs. PH039-B = 64 days, P = 0.0766; Fig. 2B; Supplementary Figs. S3A and S3B; Supplementary Table S3). Both lineages were maintained to study the basis of these differences. PH039-A and PH039-B were both cisplatin sensitive, although PH039-A had a more durable response (TTP = >120 days vs. 106 days for PH039-B, P = 0.1269; Fig. 2C; Supplementary Table S3).

When assessing the meRAD51C patterns observed for each lineage, MS-HRM results for untreated PH039-A aliquots revealed a “heterogeneous” pattern of meRAD51C (causing a broad MS-HRM peak), consistent with independent findings for this model (Supplementary Fig. S4A; ref. 35). Such heterogeneous methylation has been previously reported for other genes, including the CDKN2B gene (36). It should be noted that this heterogeneous methylation is different to a heterozygous meRAD51C pattern, which would consist of two melt peaks—one methylated and one unmethylated (example in Supplementary Fig. S4B). Instead, a broad MS-HRM meRAD51C peak is characteristic of a highly variable mixture of epialleles (i.e., meRAD51C copies) containing different patterns of CpG methylation, which then can form multiple heteroduplexes, resulting in the broadened peak. Half of the tested vehicle-treated PH039-B aliquots matched this PH039-A profile; the other half, however, displayed a more-methylated heterogeneous profile—designated “right-shifted heterogeneous” (Supplementary Table S4).

To elucidate the underlying assortment of cancer epialleles corresponding to these meRAD51C MS-HRM profiles in PH039 (“heterogeneous” vs. “right-shifted heterogeneous”), targeted meRAD51C bisulfite sequencing was performed on PDX aliquots with each profile. This analysis enables characterization of CpG methylation patterns for individual epialleles and also shows the frequency of each unique epiallele within the neoplastic sample. Sequencing revealed that the heterogeneous meRAD51C patterns in PH039-A represented a mixture of multiple epialleles, with a majority of these having reduced methylation of CpG sites at positions −114 and −69 (condensed amplicon summary in Fig. 2D, full profiles in Supplementary Figs. S4C–S4G). However, it was found that the CpG site at position −69 was, in contrast, highly methylated in the vehicle-treated PH039-B samples with “right-shifted heterogeneous” MS-HRM profiles, thereby explaining the shifts in MS-HRM curves (Fig. 2D; Supplementary Figs. S4D–S4G). Consistent with the MS-HRM data, PDX #183 tumor aliquots had highly methylated and homogeneous meRAD51C sequencing profiles relative to meRAD51C PH039 tumor aliquots (Fig. 2D; Supplementary Figs. S5A and S5B).

Interestingly, both the “heterogeneous” and “right-shifted heterogeneous” meRAD51C profiles of PH039-A and PH039-B were found to be associated with gene silencing in vehicle-treated samples, like the homogeneously methylated PDX #183 (Fig. 2E). As a previous study of heterogeneous promoter methylation at the CDKN2B locus in acute leukemia found that epialleles with <40% CpG methylation across the promoter could re-express the gene (regardless of CpG position), we were interested in the minimum degree of methylation required in this genomic region for silencing to occur (37). In our PDX samples, this was found to be seven methylated CpG sites out of 13 sites in total (54% of region CpG sites methylated; Supplementary Fig. S5C).

Both PDX PH039 lineages and PDX #183 were also found to have impaired RAD51 foci forming capacity following DNA damage, confirming that homogeneous and heterogeneous meRAD51C patterns both cause RAD51C gene silencing and HRD (Fig. 2F; Supplementary Fig. S6).

Given the complexity of the meRAD51C patterns uncovered by the targeted bisulfite amplicon assay in this study, we have summarized the definitions used to describe meRAD51C in Supplementary Figs. S7A–S7E. “Methylation zygosity” refers to the combinations of epialleles present within individual cells. These can be “homozygous methylated” (only highly/completely methylated epialleles present), “heterozygous methylated” (mixture of highly/completely methylated and unmethylated epialleles) and “unmethylated” (no methylated epialleles present). “Methylation pattern,” in contrast, describes epiallele diversity at the tumor/tissue level, with either “homogeneous” or “heterogeneous” mixtures of methylated epialleles detected.

In an effort to drive meRAD51C loss and PARPi resistance in PDX models, in vivo cyclical rucaparib retreatment experiments were carried out. PDX #183 tumor aliquots were retreated at a rucaparib dose of 300 mg/kg (Supplementary Figs. S8A and S8B), whereas PH039-A and -B xenografts required a higher dose of 450 mg/kg due to their rapid growth rate (Fig. 3A).

Rucaparib retreatment of PH039-A and PH039-B tumor aliquots resulted in recurrent cancers, which were treatment-refractory by the third cycle of treatment and thus did not reach the fourth cycle (Fig. 3B). All recurrent PH039-A and -B cancers were found to have complete loss of meRAD51C by the third cycle (Fig. 3C). Interestingly, no heterozygous meRAD51C patterns were detected in these refractory cancers. This was explained by a somatic loss of heterozygosity of the chromosome 17, including the RAD51C locus, resulting in a single gene copy in both PDX PH039 and PDX #183 (before treatment; Fig. 3D). Thus, heterozygous meRAD51C, which requires two or more copies of the RAD51C gene, would not be expected to occur in PDX models #183 or PH039, in which there is a single copy of the gene. Biallelic inactivation of BRCA1 achieved by methylation of one allele and somatic loss of the other is an accepted mechanism of functional BRCA1 loss in BRCA1 methylated breast and ovarian tumors (38), and our data now support the relevance of this mechanism for meRAD51C.

Although all PH039 tumor aliquots lost meRAD51C completely following three cycles of rucaparib 450 mg/kg, PDX #183 tumor aliquots had stable methylation patterns. Three of 12 (25%) PDX #183 tumor aliquots presented with a small degree of meRAD51C loss following four cycles of rucaparib at 300 mg/kg (Fig. 3E; Supplementary Figs. S9A and S9B; Full summary of MS-HRM screening results found in Supplementary Table S5). Rucaparib 450 mg/kg treatment data also indicated that PDX #183 had relatively stable meRAD51C under PARPi pressure compared with either PH039 lineage. Only one PDX #183 tumor aliquot had extensive loss of meRAD51C (post-niraparib, Supplementary Figs. S9C and S9D; Supplementary Table S6). Cisplatin treatment was not as effective as PARPi in driving meRAD51C loss in PDX #183 and PH039 lineage B (Supplementary Fig. S9E; Supplementary Table S6). Nonetheless, the niraparib 100 mg/kg and rucaparib 300 mg/kg treatment data established that PH039-A consistently had the least stable meRAD51C under PARPi pressure compared with either PDX #183 or PH039-B (Fig. 3E; Supplementary Table S5).

Loss of meRAD51C correlated with increased/detectable gene expression in rucaparib-treated PH039-A and PH039-B aliquots and niraparib-treated PH039-A (Fig. 3F; Supplementary Figs. S10A and S10B). This is consistent with additional findings for this model under niraparib retreatment pressure (35). MeRAD51C loss was less frequent in PDX #183, although PDX #183 tumor aliquots with meRAD51C loss also had elevated RAD51C gene expression, albeit the levels of RAD51C expression detected in this PDX were much lower relative to PH039 samples (Supplementary Figs. S10C and S10D).

Copy number profiles of unmethylated PDX PH039 lineage A and B aliquots were assessed (using SNP arrays) to determine whether unmethylated cells were derived from the same preexisting clones. It was found that resulting unmethylated PDX aliquots were derived from distinct clones within each lineage, indicating that PARPi resistance was not due to selection of rare preexisting resistant subclones within the original PH039 HGSC, but was instead acquired independently within each PH039 lineage (Fig. 3G).

To assess the relevance of the PDX findings for the clinic more broadly, we accessed additional rare cases of HGSC harboring meRAD51C. This cohort included three women recruited to the Australian Ovarian Cancer Study (AOCS) as part of the International Cancer Genome Consortium - Ovarian Cancer Project (29), one recently diagnosed woman from the WEHI Stafford Fox Rare Cancer (SFRC) program (Supplementary Figs. S11A–S11C), and eight previously reported cases from the University of Washington cohort (as assessed by MS-PCR; ref. 8). Patients were either chemo-naïve or pretreated.

DNA from 11 patient HGSC samples was tested on both the MS-HRM and targeted meRAD51C NGS assays, whereas the SFRC case was only tested by MS-HRM, revealing both homogeneous and heterogeneous patterns of meRAD51C (Table 1; Supplementary Figs. S11–S13). Heterogeneous meRAD51C, like that of PDX PH039-A, was detected in half of all patient samples [6/12 (50%); Table 1; Fig. 4A; Supplementary Figs. S11–S13]; with no obvious association with personal/family history, stage, histology, platinum response, or previous chemotherapy (Table 1; Supplementary Table S7). Merged NGS amplicon summaries of AOCS and LS patient samples revealed that heterogeneously methylated samples LS473, LS467, and AOCS-120 had reduced methylation at the same CpG sites as PDX PH039-A (CpG −69 and/or −114), whereas heterogeneously methylated patient sample AOCS-143 had reduced methylation at sites CpG −49 and −93, and LS472 had reduced methylation at sites CpG −43, −49, and −83 (Fig. 4B; Supplementary Fig. S13). Importantly, like the meRAD51C PDX samples, patient cancers also appeared to have a bimodal distribution of CpG methylation patterns—where epialleles either had >7/13 CpG sites methylated, or 0 CpG sites methylated (unmethylated epialleles; Fig. 4A; Supplementary Fig. S14). Thus, the “% methylated alleles” used to calculate zygosity in Table 1 refers to all methylated alleles detected.

Tumor purity and RAD51C copy number, estimated from SNP array data using the same DNA samples with meRAD51C sequencing available, allowed us to calculate methylation zygosity (Fig. 4C; Table 1). All tested samples had high meRAD51C (90–100%) when accounting for neoplastic cellularity, except for sample AOCS-106, which had approximately 60% meRAD51C (Table 1; Fig. 4C). Interestingly, HGSC AOCS-143 also had a pathogenic germline BRCA1 mutation c.2716A>T (p.Lys906*) detected by sequencing (29). As this mutation is in the 11q region of exon 10, which is spliced out of the D11q isoform of BRCA1, the relative contributions of this mutation and the high meRAD51C to HRD in this cancer is unclear (39). No other patient samples containing meRAD51C were found to have HRR gene mutations (8, 29).

The adjusted meRAD51C measurements for the patient HGSC sample, AOCS-106, suggested a heterozygous meRAD51C profile, with one of three RAD51C gene copies lacking promoter methylation (60% meRAD51C in neoplasm; Fig. 4C and D; Table 1). We have previously reported on such a 2/3 (∼66%) heterozygous methylation pattern in the BRCA1-methylated cell line OVCAR8, which was confirmed to have BRCA1 expression, HRR capacity and was PARPi-resistant (19). Thus, this pattern of heterozygous meRAD51C was also expected to result in restored RAD51C gene expression and HRR proficiency. This was supported by the RNA-seq data for this case, where RAD51C gene expression was elevated relative to other meRAD51C cases (Fig. 4E). The higher neoplastic purity of this sample relative to others provides even stronger evidence for restored RAD51C gene expression within the carcinoma. This sample was from a patient who had received neoadjuvant chemotherapy prior to sample collection (first-line carboplatin and paclitaxel). Given that homozygous BRCA1 methylation has been shown to become heterozygous following chemotherapy in other studies (19), it is likely that this carcinoma was initially homozygous for meRAD51C, subsequently becoming heterozygous (losing methylation of one RAD51C copy) under treatment pressure. Although no pretreatment cancer samples were available for this case to confirm the initial meRAD51C status, an HRD score analysis revealed genomic signatures consistent with historic HRD (Fig. 4F; Supplementary Figs. S15A and S15B). Thus, we hypothesize from these data that this HGSC originally had HRD caused by complete meRAD51C, which was lost following first-line chemotherapy.

A second piece of post-chemotherapy sample LS-473 was identified as another potential heterozygous meRAD51C sample, with an adjusted neoplastic meRAD51C value of 85%, compared with 97% in the first HGSC piece (Supplementary Fig. S14; Supplementary Table S8). Given this cancer was estimated to have four copies of the RAD51C gene (SNP arrays), loss of methylation in one cellular epiallele (sufficient for competent HRR DNA repair) in all cells would result in tumor meRAD51C value of 75%. It is possible that only a portion of cells within the sample had lost methylation of RAD51C copy, resulting in an adjusted meRAD51C value higher than 75% but less than 100%, however, we could not assign a heterozygous status to this cancer with confidence. This case highlights the complexity of interpreting meRAD51C zygosity in patient HGSC samples.

The impact of RAD51C gene methylation on PARPi responses in HGSC has long been unclear (7–9, 11, 12, 18). In this study we have clarified this issue, of high importance for the clinic, demonstrating that meRAD51C is a bona fide HRR defect that could sensitize HGSC to PARPi treatment, but once lost under treatment pressure, resulted in PARPi-refractory HGSC PDX. As seen for meBRCA1, a single unmethylated copy of RAD51C in HGSC cells was found to be sufficient to restore HRR DNA repair and cause PARPi resistance in PDX. Thus, the presence of fully unmethylated epialleles (those not derived from stroma) should be estimated using tumor purity and RAD51C copy number when analyzing meRAD51C in HGSC samples. Our findings have implications for therapeutic selection, as therapies preventing methylation loss should be investigated and those causing de-methylation should be used with caution in these women until additional preclinical and clinical evidence is available.

Our finding that heterogeneous meRAD51C caused gene silencing/HRD in PDX and was present in 50% of patient tumor samples screened, has implications for meRAD51C detection in HGSC. Heterogeneous meRAD51C occurs when some CpG sites show no methylation in a given allele and multiple alleles with different methylation patterns coexist in the region under investigation [reviewed by Mikeska and colleagues (40)]. Some promoter CpG sites are potentially less critical for RAD51C silencing than others, and more models of heterogeneous meRAD51C HGSC would be required to investigate this. However, we did find that only fully unmethylated epialleles were associated with active gene expression in PDX, whereas all silenced epialleles had at least 54% of the region methylated. This bimodal distribution of CpG methylation was also observed in patient samples. Thus, any degree of RAD51C promoter methylation detected in a sample should be flagged as a potential HRR defect for further validation (29), and methylation arrays that interrogate limited CpG sites should be avoided, to ensure that heterogeneously methylated samples with potential for PARPi response are not missed (6, 18). These findings are also relevant for other cancer types, such as gastric cancer, where meRAD51C is frequently reported (especially in cases positive for Epstein–Barr virus; ref. 41), and where both homogeneous and heterogeneous meRAD51C have been associated with the PARPi, Olaparib, response in cell lines (14).

Our group has previously demonstrated that loss of methylation of a single methylated BRCA1 epiallele is sufficient to restore HRR in cells and cause PARPi resistance (19). In this study, we have confirmed that the same principles apply to RAD51C in PDX models #183 and PH039, which by chance both only had one cellular copy of the gene. This single-allele haplo-sufficiency of RAD51C was also previously observed by our group for resistance-causing secondary mutations in RAD51C (24). These observations support a model in which heterozygous meRAD51C (mix of methylated and unmethylated epialleles within a cell) would also cause gene expression and, thus, PARPi resistance in HGSC. Indeed, our method of meRAD51C characterization in patient samples enabled us to detect a pretreated patient sample with apparent heterozygous meRAD51C. Although matched archival/untreated neoplastic material was not available for this case, all evidence supported a model in which homozygous meRAD51C had become heterozygous following chemotherapy. Future studies of meRAD51C zygosity in HGSC would greatly benefit from the availability of matched pre- and posttreatment patient HGSC samples, while recognizing that meRAD51C cases are rare and even large cohorts contain few, hence as matched pairs are even more difficult to locate, each pair is extremely valuable. This further highlights the need for routine meRAD51C screening in the clinic.

Importantly, given that loss of meRAD51C is associated with restored HRR and PARPi resistance, therapeutic strategies directly causing DNA methylation loss, present a potential risk in patients with meRAD51C or meBRCA1. For example, anticancer therapies, such as histone deacetylase (HDAC) and DNA methyltransferase (DNMT) inhibitors may cause permanent meRAD51C loss and remove an important therapeutically targetable feature of the patient's cancer. Indeed, decitabine (DNMT inhibitor) treatment has been shown to cause BRCA1 methylation loss in plasma of ovarian cancer patients (42), whereas 5-azacitadine treatment restored RAD51C gene expression in gastric cancer cell lines (14). These compounds are currently being trialled in recurrent ovarian cancers due to their immunogenic effects (e.g., trial NCT02915523; ref. 43). As HGSC with BRCA1 and RAD51C methylation make up 12% of all patients with HGSC, and have the potential to respond to PARPi, the possible detrimental effects of de-methylating agents on these marks should be considered in such studies.

Through our work on PDX models, we found that the model with the most heterogeneous meRAD51C (PDX PH039-A) had the least stable meRAD51C under treatment pressure, relative to other models, while the most homogeneous meRAD51C PDX (#183) had the most stable methylation. Given this observation, we initially hypothesized that heterogeneous meRAD51C is more easily lost following platinum/PARPi treatment. However, heterogeneous meRAD51C was detected in nearly half of the screened HGSC patient samples, and there did not appear to be an association with initial platinum chemotherapy response status or patient survival. Indeed, patient sample AOCS-120 still contained 100% heterogeneous meRAD51C at recurrence following six lines of chemotherapy, demonstrating that other/additional factors are influencing meRAD51C stability under platinum and PARPi pressure in HGSC. Identifying these underlying factors could be of great utility and enable design of novel therapeutic strategies to abrogate PARPi resistance due to RAD51C or BRCA1 methylation loss. Useful preclinical models for this type of study include the two lineages of PDX PH039 where meRAD51C loss and PARPi resistance were acquired independently and not via selection of a preexisting clone. Previous studies have reported that BRCA1/RAD51C methylation were not associated with platinum response and improved survival in patients with HGSC when compared with those with mutations in the same genes (8). It would also be valuable to confirm the stability of pathogenic mutations versus methylation of the RAD51C gene in future using a larger cohort of PDX or patients, to account for the heterogeneity of HGSC.

In conclusion, this study identifies both homogeneous and heterogeneous meRAD51C as bona fide HRR defects that are targetable with PARPi therapy and have potential for sustained response through multiple lines of therapy. However, promoter methylation loss can occur in both cases, leading to restored HRR and PARPi resistance. This should be considered when making treatment choices with patients with meRAD51C HGSC.

G. Chenevix-Trench, A. Green, P. Webb, D. Gertig, S. Fereday, S. Moore, J. Hung, K. Harrap, T. Sadkowsky, N. Pandeya, M. Malt, A. Mellon, R. Robertson, T. Vanden Bergh, M. Jones, P. Mackenzie, J. Maidens, K. Nattress, Y. E. Chiew, A. Stenlake, H. Sullivan, B. Alexander, P. Ashover, S. Brown, T. Corrish, L. Green, L. Jackman, K. Ferguson, K. Martin, A. Martyn, B. Ranieri, J. White, V. Jayde, P. Mamers, L. Bowes, L. Galletta, D. Giles, J. Hendley, T. Schmidt, H. Shirley, C. Ball, C. Young, S. Viduka, H. Tran, S. Bilic, L. Glavinas, J. Brooks, R. Stuart-Harris, F. Kirsten, J. Rutovitz, P. Clingan, A. Glasgow, A. Proietto, S. Braye, G. Otton, J. Shannon, T. Bonaventura, J. Stewart, S. Begbie, M. Friedlander, D. Bell, S. Baron-Hay, A. Ferrier, G. Gard, D. Nevell, N. Pavlakis, S. Valmadre, B. Young, C. Camaris, R. Crouch, L. Edwards, N. Hacker, D. Marsden, G. Robertson, P. Beale, J. Beith, J. Carter, C. Dalrymple, R. Houghton, P. Russell, M. Links, J. Grygiel, J. Hill, A. Brand, K. Byth, R. Jaworski, P. Harnett, R. Sharma, G. Wain, B. Ward, D. Papadimos, A. Crandon, M. Cummings, K. Horwood, A. Obermair, L. Perrin, D. Wyld, J. Nicklin, M. Davy, M. K. Oehler, C. Hall, T. Dodd, T. Healy, K. Pittman, D. Henderson, J. Miller, J. Pierdes, P. Blomfield, D. Challis, R. McIntosh, A. Parker, B. Brown, R. Rome, D. Allen, P. Grant, S. Hyde, R. Laurie, M. Robbie, D. Healy, T. Jobling, T. Manolitsas, J. McNealage, P. Rogers, B. Susil, E. Sumithran, I. Simpson, K. Phillips, D. Rischin, S. Fox, D. Johnson, S. Lade, M. Loughrey, N. O'Callaghan, W. Murray, P. Waring, V. Billson, J. Pyman, D. Neesham, M. Quinn, C. Underhill, R. Bell, L. F. Ng, R. Blum, V. Ganju, I. Hammond, Y. Leung, A. McCartney, M. Buck, I. Haviv, D. Purdie, D. Whiteman & N. Zeps

K. Nesic reports grants from Stafford Fox Medical Research Foundation, other support from Bev Gray PhD Top Up Scholarship and Australian Postgraduate Award (APA) stipend, and nonfinancial support from Clovis Oncology during the conduct of the study. O. Kondrashova reports personal fees from XING Technologies outside the submitted work. C.J. Vandenberg reports grants from Stafford Fox Medical Research Foundation during the conduct of the study. E. Lieschke reports grants from Stafford Fox Medical Research Foundation and nonfinancial support from Clovis Oncology during the conduct of the study. G. Dall reports grants from Stafford Fox Medical Research Foundation and Victorian Cancer Agency, and nonfinancial support from Clovis Oncology during the conduct of the study. N. Bound reports grants from Stafford Fox Medical Research Foundation during the conduct of the study. K. Shield-Artin reports grants from Stafford Fox Medical Research Foundation during the conduct of the study. D. Kee reports grants from Stafford Fox Medical Research Foundation and Medical Research Future Fund, and nonfinancial support from Clovis Oncology during the conduct of the study. N. Traficante reports grants from AstraZeneca Pty Ltd. during the conduct of the study and grants from AstraZeneca Pty Ltd. outside the submitted work. Australian Ovarian Cancer Study reports grants from AstraZeneca Pty Ltd. during the conduct of the study and grants from AstraZeneca Pty Ltd. outside the submitted work. A. DeFazio reports grants from AstraZeneca outside the submitted work. S.J. Weroha reports grants from NCI during the conduct of the study and personal fees from Kiyatec and AstraZeneca outside the submitted work. N. Waddell reports grants from Australian National Medical and Research Council (NHMRC, APP1139071) during the conduct of the study and other support from genomiQa Pty Ltd. outside the submitted work. S.H. Kaufmann reports grants from NCI and Stand Up to Cancer during the conduct of the study and grants from Takeda outside the submitted work. A. Dobrovic reports grants from National Breast Cancer Foundation of Australia during the conduct of the study. M.J. Wakefield reports grants from Stafford Fox Foundation during the conduct of the study. C.L. Scott reports grants from Stafford Fox Medical Reserch Foundation, National Health and Medical Research Council, Victorian Cancer Agency, Herman Trust, University of Melbourne, Cancer Council Victoria, Sir Edward Dunlop Cancer Research Fellow, Cooperative Research Centre Cancer Therapeutics, and Australian Cancer Research Foundation during the conduct of the study; other support from Sierra Oncology and other support from Clovis Oncology outside the submitted work; and unpaid advisory boards: AstraZeneca, Clovis Oncology, Roche, Eisai, Sierra Oncology, Takeda, MSD. No disclosures were reported by the other authors.

K. Nesic: Conceptualization, resources, data curation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. O. Kondrashova: Conceptualization, resources, data curation, formal analysis, supervision, investigation, visualization, methodology, project administration, writing–review and editing. R.M. Hurley: Conceptualization, resources, validation, writing–review and editing. C.D. McGehee: Conceptualization, resources, validation, writing–review and editing. C.J. Vandenberg: Resources, data curation, supervision, writing–review and editing. G-Y. Ho: Resources, data curation, writing–review and editing. E. Lieschke: Resources, data curation, writing–review and editing. G. Dall: Investigation, writing–review and editing. N. Bound: Investigation, writing–review and editing. K. Shield-Artin: Investigation, writing–review and editing. M. Radke: Investigation, writing–review and editing. A. Musafer: Investigation, writing–review and editing. Z.Q. Chai: Investigation, writing–review and editing. M.R.E. Ghamsari: Investigation, writing-review and editing. M.I. Harrell: Investigation, writing–review and editing. D. Kee: Resources, writing–review and editing. I. Olesen: Resources, writing–review and editing. O. McNally: Resources, writing–review and editing. N. Traficante: Resources, writing–review and editing. AOCS: Resources, data curation, writing–review and editing. A. DeFazio: Resources, writing–review and editing. D.D.L. Bowtell: Resources, methodology, writing–review and editing. E.M. Swisher: Resources, methodology, writing–review and editing. S.J. Weroha: Resources, investigation, writing–review and editing. K. Nones: Resources, investigation, writing-review and editing. N. Waddell: Conceptualization, resources, writing–review and editing. S.H. Kaufmann: Conceptualization, resources, supervision, writing–review and editing. A. Dobrovic: Conceptualization, resources, formal analysis, supervision, writing–review and editing. M.J. Wakefield: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–review and editing. C.L. Scott: Conceptualization, resources, data curation, supervision, funding acquisition, writing–review and editing.

The authors thank Dr. Paul Haluska and Mariam AlHilli (Mayo Clinic) for the cryopreserved PDX material used to re-establish PDX PH039 within our laboratory and for original PH039 characterization. The authors also thank Silvia Stoev, Rachel Hancock, Kathy Barber, Scott Wood, Lambros T. Koufariotis, and Conrad Leonard for technical assistance. They thank Clovis Oncology for providing rucaparib for in vivo experiments. This work was supported by fellowships and grants from the National Health and Medical Research Council (NHMRC Australia; project grant 1062702 for C.L. Scott, M.J. Wakefield, A. De Fazio, D.D.L. Bowtell, and Senior Research Fellowship APP1139071 for N. Waddell); the Stafford Fox Medical Research Foundation (to C.L. Scott); Cancer Council Victoria (Sir Edward Dunlop Fellowship in Cancer Research to C.L. Scott); the Victorian Cancer Agency (Clinical Fellowships to C.L. Scott CRF10–20, CRF16014; and G. Dall ECRF19003); the Herman Trust, University of Melbourne (Fellowship to C.L. Scott); NIH (2P50CA083636 to E.M. Swisher) and the Wendy Feuer Ovarian Cancer Research Fund (to E.M. Swisher); the Bev Gray Ovarian Cancer Scholarship (PhD Top-Up Scholarship) and Research Training Program Scholarship (PhD Scholarship) to K. Nesic. The Olivia Newton-John Cancer Research Institute acknowledges the support of the Victorian Government Operational and Infrastructure Support Program. This work was also supported in part by grants from the NIH (P50 CA136393 to S.H. Kaufmann and S.J. Weroha; F30 CA213737 to C.D. McGehee; T32 GM072474 to R.M. Hurley), fellowship support from the Mayo Foundation for Education and Research (to R.M. Hurley, C.D. McGehee) and Stand Up To Cancer-Ovarian Cancer Research Fund Alliance-National Ovarian Cancer Coalition Dream Team Translational Cancer Research Grant (to E.M. Swisher, S.J. Weroha, and S. H. Kaufmann). Stand Up To Cancer is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C. This work was made possible through the Australian Cancer Research Foundation, the Victorian State Government Operational Infrastructure Support, and Australian Government NHMRC IRIISS. All authors, the WEHI Stafford Fox Rare Cancer Program, and the AOCS would like to thank all of the women who participated in these research programs. The AOCS would also like to acknowledge the contribution of the study nurses, research assistants, and all clinical and scientific collaborators to the study. The complete AOCS Study Group can be found at www.aocstudy.org. The Australian Ovarian Cancer Study gratefully acknowledges additional support from Ovarian Cancer Australia and the Peter MacCallum Foundation. The Australian Ovarian Cancer Study Group was supported by the U.S. Army Medical Research and Materiel Command under DAMD17-01-1-0729, The Cancer Council Victoria, Queensland Cancer Fund, The Cancer Council New South Wales, The Cancer Council South Australia, The Cancer Council Tasmania and The Cancer Foundation of Western Australia (Multi-State Applications 191, 211, and 182), and the National Health and Medical Research Council of Australia (NHMRC; ID199600, ID400413, and ID400281).

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.