Diverse BRCA1 and BRCA2 Reversion Mutations in Circulating Cell-Free DNA of Therapy-Resistant Breast or Ovarian Cancer

BRCA1 and BRCA2 play pivotal roles in homologous recombination (HR) DNA repair, and germline mutations affecting these genes result in an increased risk of early breast and ovarian cancer development (1). Complete loss of BRCA1 or BRCA2 function results in lack of HR repair of DNA double-strand breaks (1). Cancer cells arising in germline BRCA1 and BRCA2 carriers lose the wild-type allele, and, as a consequence, lose competent HR due to biallelic inactivation of BRCA1 or BRCA2. In this context, DNA double-strand breaks are repaired by error-prone mechanisms, such as nonhomologous end joining (2–4). Tumors harboring defective HR DNA repair have been shown to be sensitive to platinum-based chemotherapy and inhibitors of the poly(ADP-ribose) polymerase (PARP) (5, 6), given that these agents induce double-strand breaks either directly or through the stalling and subsequent collapse of replication forks. BRCA1 or BRCA2 breast and ovarian cancers are reported to be sensitive to platinum-based chemotherapy and PARP inhibition (7–10), with platinum-based chemotherapy serving as the mainstay of treatment of patients with ovarian cancer. Several PARP inhibitors have recently been approved for the treatment of advanced BRCA1- or BRCA2-mutant ovarian cancer (11) and are in phase III clinical trials for patients with BRCA1- or BRCA2-mutant breast cancer (12). Importantly, several mechanisms of resistance to these agents have been reported in preclinical models and in clinical studies (3, 4). One mechanism of resistance to platinum-based chemotherapy and PARP inhibitors is in the form of reversion mutations or intragenic deletions that restore the open reading frame of the original germline BRCA1 or BRCA2 mutation, resulting in a functional protein with reacquisition of competent HR DNA repair (13–15).

Circulating tumor DNA (ctDNA) found in the plasma of cancer patients has been shown to constitute a source of tumor-derived DNA that can be employed for the analysis of sequencing-based biomarkers (16). Although ctDNA frequently comprises only a small fraction of total circulating cell-free (cf)DNA and varies according to disease burden and between cancer types (17), it is possible to detect much of the entire repertoire of somatic genetic alterations found in primary tumors or metastatic disease in cfDNA samples if high-depth sequencing approaches are employed (18–20). In addition, multiclonal BRCA2 reversion mutations associated with resistance to PARP inhibitors have been identified in cfDNA from two metastatic prostate cancer patients with germline BRCA2 mutations (21).

The aims of this exploratory, hypothesis-generating study were to define whether putative BRCA1/2 reversion mutations can be detected in the cfDNA of patients with BRCA1 or BRCA2 ovarian and breast cancer resistant or refractory to platinum-based chemotherapy or PARP inhibitors, to determine whether the putative BRCA1/2 reversion mutations found in these patients could have an impact on BRCA1/2 function, and to develop analysis techniques that could potentially be employed in the implementation of biomarkers for future patient selection.

Nineteen patients with ovarian cancer and five patients with breast cancer were prospectively accrued for this study. Inclusion criteria for the patients with ovarian cancer encompassed proven BRCA1 or BRCA2 germline mutations, stage III or IV disease resistant or refractory to platinum-based chemotherapy, and availability of archived cancer tissue (Table 1). Patients with any other concurrent stage III/IV cancer were excluded. This study was approved by the Institutional Review Board (IRB) of Memorial Sloan Kettering Cancer Center (MSKCC; New York, NY; IRB #13-128), and written informed consent was obtained from all participants. Radiologic recurrence within 6 months of last platinum administration was defined as resistant disease, whereas unresponsiveness to or progression during platinum therapy was defined as refractory disease (22). Peripheral blood samples (EDTA tubes) were collected at the time of scheduled chemotherapy following relapse or progression. Of the 19 patients with ovarian cancer included, 18 had high-grade serous, and one had endometrioid ovarian cancer (Table 1). Inclusion criteria for the patients with breast cancer entailed proven BRCA1 or BRCA2 germline mutations, metastatic disease, and prior treatment with platinum chemotherapy and/or PARP inhibitors. Samples were collected under studies approved by multicenter research ethics committees (ref. nos. 10/H0805/50 and 11/LO1595) in the United Kingdom. Peripheral blood samples (EDTA tubes) were collected upon disease progression, and serially after intervening therapy at subsequent disease progression (Table 2). This study is compliant with the Declaration of Helsinki.

To avoid sample issues related to the stability of EDTA cfDNA, blood samples collected in EDTA tubes were processed within 2 hours of sample collection, centrifuged, and plasma samples were stored at −80°C until DNA extraction as described previously (20). DNA was extracted from plasma using the QIAamp Circulating Nucleic Acid Kit (Qiagen) according to the manufacturer's instructions as described previously (16, 20). DNA was quantified using the Qubit Fluorometer (Invitrogen, Thermo Fisher Scientific).

Representative formalin-fixed paraffin-embedded (FFPE) tissue sections of all ovarian tumors and of core biopsies of metastases from the patients with advanced breast cancer obtained at diagnosis (patient 1109 patient) and at recurrence (patient L031) were stained with nuclear fast red and microdissected with a sterile needle under a stereomicroscope (Olympus SZ61) to ensure >80% of tumor cell content, as described previously (23). In nine ovarian cancer cases, histologically distinct regions of the primary tumor or distinct anatomical sites, including omental implants were available and microdissected [median of 1 (range, 1–4) anatomically distinct regions per case]. Genomic DNA was extracted from tumor samples and peripheral blood leukocytes using the DNeasy Blood and Tissue Kit (Qiagen) and quantified using the Qubit Fluorometer, as described previously (23, 24).

Tumor DNA from patients with ovarian cancer, cfDNA from patients with ovarian and breast cancer, and their respective germline DNA were subjected to targeted massively parallel sequencing in the MSKCC Integrated Genomics Operation (IGO) as described previously (24, 25) using a custom panel of baits encompassing all exons and introns of BRCA1 and BRCA2, and all exons of 141 additional genes reported to be involved in DNA repair, drug resistance, resistance to PARP inhibitors/platinum salts, and genes recurrently mutated in ovarian cancer, including TP53 (Supplementary Table S1; refs. 26–28). In addition, baits tiling common SNPs were included to allow for copy number analysis (25). Serial plasma samples from breast cancer patients L031 and 1109 were subjected also to MSK-IMPACT sequencing targeting 410 key cancer genes, as described previously (25, 29). Sequence data were analyzed as described previously (Supplementary Methods; refs. 24, 29), and in addition, variants in the cfDNAs and tumors were assessed using the SAMtools mpileup tool (30) and Varscan 2 (31). Sequence data are available at the Sequence Read Archive (SRP100525).

Putative somatic reversion mutations or intragenic deletions were defined as somatic genetic alterations that would result in a restoration of the open reading frame of BRCA1 or BRCA2 in the cfDNA from a patient harboring a known germline mutation affecting BRCA1 or BRCA2, respectively. To identify putative reversion mutations and intragenic deletions, we extracted all reads from BRCA1 or BRCA2 (i.e., the gene affected by the germline mutation in a given case). Among these reads, we used SAMtools mpileup tool (30) to search for (i) somatic small insertions and deletions (indels) that would restore the reading frame of BRCA1/2 in patients with germline indels; (ii) somatic single-nucleotide variants (SNV) that restore the BRCA1/2 reference allele in patients with germline point mutations; and (iii) intragenic deletions that delete the BRCA1/2 germline mutation and result in restoration of the open reading frame. For SNVs and indels, single reads supporting a mutation were also examined owing to the limited fraction of ctDNA in total plasma cfDNA.

To account for potential large intragenic deletions (>40 bp) that may not be aligned as single reads by BWA (32), we further included clipped reads aligning to multiple locations and spanning the germline mutation as putative intragenic deletions. All putative large intragenic deletions were visually inspected using the Integrated Genomics Viewer (33).

All putative BRCA1/2 reversion mutations and intragenic deletions were annotated using Oncotator (34), in conjunction with the respective BRCA1/2 germline mutation. The cDNA changes predicted by Oncotator were applied consecutively to the BRCA1/2 cDNA transcripts and translated into amino acids. We further inferred the Levenshtein distance, which denotes the number of insertions, deletions, and substitutions required to change one protein into the other (35). Each of the germline mutant and putative reversion mutant BRCA1/2 proteins was annotated with their respective Levenshtein distance to the wild-type BRCA1/2 protein. Any of the putative reversion mutations that differed in this metric compared with the germline mutation were flagged for manual review. Scripts to aid in the search of reversion mutations and compare their protein sequences are available online (36).

To ascertain the fraction of tumor ctDNA in the cfDNA obtained from plasma of patients with ovarian cancer, we employed (i) the TP53 variant allele fractions (VAF) of the ovarian tumors given that TP53 mutations are present, clonal and truncal in >97% of high-grade serous ovarian cancers (HGSOC; refs. 37, 38); (ii) the tumor ploidy, local TP53 copy number, and tumor purity based on FACETS (39) and ABSOLUTE (40); and (iii) the TP53 VAF from plasma. If TP53 mutations were not identified in cfDNA, a distinct clonal mutation was employed for analysis. The fraction of ctDNA in cfDNA could not be defined in cases where only subclonal mutations from the ovarian tumor were detected in the respective cfDNA.

Given the lack of matched tumor tissue from the metastatic breast cancers subjected to targeted massively parallel sequencing, a different approach was employed to infer the fraction of ctDNA in total plasma DNA. The cfDNA was quantified on a Bio-Rad QX100 droplet (d)PCR using ribonuclease P (RNase P) as a reference, as described previously (20). At least two negative control wells without DNA were included in each run. The amount of amplifiable RNase P DNA and the number of RNase P copies were calculated using the Poisson distribution in QuantaSoft (Bio-Rad) and used together with the highest VAF identified by targeted massively parallel sequencing in the cfDNAs of the patients with breast cancer to infer the fraction of ctDNA in cfDNA.

The putative c.85delG BRCA1 reversion mutation identified by massively parallel sequencing in case OCT15 was validated using the Raindrop dPCR system (RainDance Technologies) as described previously (41). Droplet PCR (dPCR) conditions were optimized as described previously (20), and assay sensitivity was tested using BRCA1 wild-type DNA library spiked in with a BRCA1 c.85delG-mutant synthetic oligonucleotide. Massively parallel sequencing libraries from the ovarian tumor samples and plasma DNA samples of case OCT15 were loaded onto the RainDrop Source instrument for droplet generation for amplification (forward 5′-ACTTTGTGGAGACAGGTT-3′, reverse 5′-TGAGCCTCATTTATTTTCTTTT-3′ PCR primers), and loaded onto the Raindrop Sense instrument for droplet counting and fluorescence intensity readout as described previously (41). Libraries from germline DNA spiked in with 10, 100, and 1,000 c.85delG BRCA1 synthetic oligonucleotide molecules were included in the run as controls and for gating purposes. Data were analyzed using the RainDrop Analyst data analysis software.

The putative BRCA2 reversion mutations and the somatic SPEN and TGFBR1 mutations identified by massively parallel sequencing were validated in three plasma samples from case L031 using targeted amplicon resequencing. In case 1109, three somatic variants affecting FAT3, ERCC4 and KDM5C were validated together with the putative BRCA2 reversion mutations identified in the posttreatment plasma sample and in a tumor metastasis core biopsy affecting the liver obtained prior to treatment. At least 10 ng of plasma DNA, microdissected tumor DNA and matching peripheral blood leukocyte-derived germline DNA were amplified using Taq Hifi polymerase (Ion AmpliSeq Library Kit 2.0, Thermo Fisher Scientific), and libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England BioLabs). As controls, matched normal DNA from cases 1109 and L031 and plasma DNA from two unrelated patients with advanced breast cancer not treated with PARP inhibitors or platinum-based chemotherapy and two tumor DNA samples obtained from unrelated breast cancer core biopsies were included. PCR conditions and primers are available on request. The quality and quantity of the amplification was tested using the Agilent 2100 Bioanalyzer and the KAPA Library Quantification Kit for Illumina(Kapa Biosystems), respectively. Amplicon resequencing of the putative BRCA2 reversion mutations in the cfDNA samples of L031 and 1109 was performed twice independently, using an Illumina HiSeq2500 (first run) and an Illumina MiniSeq (Mid output kit; second run). Sequence data analyses are described in the Supplementary Methods. Only BRCA2 reversion mutations present in plasma DNA with zero counts in the germline control and in the unrelated control samples were considered validated.

293T cells and U2OS cells were obtained from the ATCC in 2008 and 2015, respectively. The identities of the cell lines were confirmed by short tandem repeat profiling after receipt as described previously (42) and tested for mycoplasma infection using a PCR-based assay (ATCC) following the manufacturer's instructions (latest test in 2016). The cells were passaged no more than 20 to 25 times after thawing.

The U2OS double-strand break (DSB) reporter system was employed to define the ability of putative BRCA1 somatic reversion mutations to recognize DSBs, as described previously (Supplementary Methods; refs. 43–45).

293T cells were transfected with pCDNA-HA-BRCA2 plasmids(i.e., wild-type HA-BRCA2, germline c.407delA HA-BRCA2del407 and putative reversion c.402_413delTCTAAATTCTTG HA-BRCA2REV) for 72 hours and lysed in lysis buffer (0.5% NP40, 25 mmol/L Tris pH 7.5, 450 mmol/L NaCl, 0.5 mmol/L EDTA and proteinase inhibitors). The cell lysates were then incubated with anti-HA agarose beads (Sigma). After three washes with wash buffer (0.5% NP40, 25 mmol/L Tris pH 7.5, 150 mmol/L NaCl, 0.5 mmol/L EDTA), the beads were boiled in SDS sampling buffer, followed by Western blotting with antibodies against HA (Santa Cruz Biotechnology), PALB2 (Novus Biologicals), and RAD51(Santa Cruz Biotechnology).

We developed a targeted capture sequencing assay comprising the coding regions of 141 genes and all intronic and exonic regions of BRCA1 and BRCA2 (Supplementary Table S1). We first established the potential of the assay for detecting ctDNA in patients with ovarian cancer, as prior studies suggested low rates of mutation detection in ctDNA of patients with ovarian cancer (46). Massively parallel sequencing analysis of germline DNA from peripheral blood leukocytes, microdissected tumor and plasma DNA of 19 cases (BRCA1, n = 12; BRCA2, n = 7), using previously validated methods (18, 20), yielded median depths of coverage of 1,569× (range, 852×–2,272×), 823× (range, 272×–2,328×) and 1,978× (range, 1,287×–4,157×), respectively (Supplementary Table S2). Given that somatic TP53 mutations are present in >97% of all HGSOCs and that the vast majority of these mutations are clonal (i.e., bioinformatically inferred to be present in virtually all cancer cells within a sample) and truncal (i.e., present as clonal mutations in all samples analyzed; refs. 37, 38), we reasoned that the TP53-mutant allele fractions in cfDNA from patients with HGSOC could be employed to ascertain indirectly the fraction of tumor DNA in the total plasma DNA. In all 19 ovarian cancers sequenced, clonal TP53 mutations were detected (Supplementary Table S2, Supplementary Fig. S1). In nine cases, multiple anatomically distinct areas of the ovarian cancer were microdissected and/or peritoneal implants and/or metastatic sites were available and sequenced separately; the TP53 mutations in these nine cases with multiregion sequencing were found to be clonal and truncal (Supplementary Table S2; Supplementary Fig. S1). Other somatic mutations detected in the 19 ovarian cancers studied here included NF1, ERCC4, RB1 and CHEK2 (Supplementary Table S2; Supplementary Fig. S1).

Analysis of the cfDNA from these patients revealed the presence of the same somatic TP53 mutation identified in the tumors from the respective patients with ovarian cancer in 15 of 19 cases (79%). TP53 VAFs in the plasma DNA ranged from 0.06% to 32.7% (Supplementary Fig. S1; Supplementary Table S2), and only in 4 samples, these TP53 mutations were identified using our standard bioinformatics pipeline. In three cases (OCT1, OCT5, OCT12), none of the clonal TP53 mutations present in the tumors were detected in the plasma, but other somatic mutations were identified, including RB1, NF1 and FAT3 mutations (Supplementary Fig. S1). In one case (OCT3), neither the clonal TP53 nor the subclonal NF1 somatic mutations present in the primary ovarian cancer were detected in the cfDNA (Supplementary Table S2; Supplementary Fig. S1). Overall, our assay identified ctDNA in plasma of 95% (18/19) patients with advanced ovarian cancer, and the median percentage of ctDNA in cfDNA was found to be 0.31% (range, 0%–32.74%; Table 1).

Having demonstrated the high sensitivity to detect ctDNA with our assay, we investigated whether putative BRCA1/2 reversion mutations could be detected in cfDNA. Using a conservative bioinformatics strategy (see Materials and Methods), six putative somatic BRCA1 reversion mutations in four patients OCT1, OCT5, OCT10 and OCT15 were identified with VAFs ranging from 0.0314% to 0.0850% (Fig. 1; Supplementary Fig. S2; Supplementary Table S3). Four of the six putative reversion mutations were flanked by microhomology sequences (Supplementary Table S3). Using dPCR (20) and spiked-in synthetic c.85delG oligonucleotides as controls, we detected the putative somatic c.85delC BRCA1 reversion mutation in the cfDNA of patient OCT15 (Fig. 2A), which harbored a germline BRCA1 c.68-69delAG mutation. In contrast, however, analysis of the pretreatment primary tumors from the ovary and fallopian tube as well as a peritoneal implant of patient OCT15 showed no reliably detectable BRCA1 reversion mutations (Fig. 2A). Validation of the reversion mutations in the other cases was not possible given that no or insufficient amounts of residual plasma DNA were available.

To ascertain whether these putative somatic BRCA1 reversion mutations would restore the ability to recognize double-strand breaks, we employed the U2OS-DSB reporter system (43–45). Following 8 Gy of ionizing irradiation, we observed that three putative BRCA1 somatic reversion mutations, all of which were flanked by microhomology sequences, namely c.108delC(OCT5), c.113delA and c.85delC (both OCT15), resulted in an induction of BRCA1 foci to levels higher than those observed in U2OS cells expressing the respective germline BRCA1 mutation (Fig. 2B). We therefore demonstrated that ctDNA sequencing can detect putative BRCA1 reversion mutations, and that these mutations may restore BRCA1 function in in vitro assays.

We next defined whether somatic reversion mutations would be detected in ctDNA of PARP inhibitor- and/or platinum-resistant advanced breast cancer patients harboring BRCA1 (n = 1) or BRCA2 (n = 4) germline mutations. Germline DNA extracted from peripheral blood leukocytes and a single (n = 2) or two sequential (n = 3) plasma DNA samples per patient were sequenced with the same custom targeted capture sequencing assay described above to a median depth of coverage of 537× (range, 457×–630×) and 1,646× (range, 1,163×–3,153×), respectively (Supplementary Table S2). In all cfDNA samples analyzed, somatic mutations were identified (VAFs, 2.38%–54.54%), including somatic TP53 mutations in two cases (Supplementary Table S2; Supplementary Fig. S3). Although the amount of cfDNA obtained per mL of plasma was similar between the breast and ovarian cancers studied here [breast median 7.8 ng cfDNA/mL plasma (range, 5.0 ng–87 ng) vs. ovarian median 12.0 ng cfDNA/mL plasma (range, 4.8 ng–32.4 ng), P = 0.2905, Mann–Whitney U test], the percentage of ctDNA was significantly higher in breast compared with ovarian cancer patients (breast median 12.7% (range, 5.2%–54.5%) vs. ovarian 0.31% (range, 0%–32.74%, P < 0.0005, Mann–Whitney U test, Tables 1 and 2]. It should be noted, however, that the methods for the quantification of ctDNA percentage in cfDNA applied to the breast and ovarian cancer samples differed, and these methodologic differences may at least in part explain the distinct levels of ctDNA in cfDNA between the two groups. Despite this important caveat and consistent with the notion that the percentage of ctDNA in cfDNA was higher in the patients with breast cancer than in the patients with ovarian cancer studied here, SNVs and indels in the cfDNA of the patients with metastatic breast cancer were identified using our standard bioinformatics approach, whereas these were detectable using the standard bioinformatics approach in the cfDNA of only four of 19 patients with breast cancer.

Analysis of the sequencing data further revealed the presence of multiple putative somatic reversion mutations in two of the BRCA2 germline mutation carriers (L031 and 1109) at VAFs ranging from 0.0549% to 0.2273% (Fig. 3; Supplementary Table S3). To validate the multiple polyclonal BRCA2 mutations, we developed an orthogonal amplicon sequencing strategy (see Materials and Methods), which was employed twice independently and confirmed the presence of all detected reversion mutations, and identified six additional putative BRCA2 reversion mutations in the cfDNA of cases L031 and 1109 (Fig. 3; Supplementary Table S3), with no detected mutations in controls (see Materials and Methods). In the plasma sample taken after carboplatin and PARP inhibitor treatment from patient 1109, harboring a c.750_753GACAdel germline mutation, up to nine distinct putative BRCA2 reversion mutations were identified suggesting polyclonality at resistance (Fig. 3A; Supplementary Table S3). Validated putative somatic reversion mutations were confirmed to restore the reading frame of BRCA2 and were flanked by (micro) homology sequences (Supplementary Table S3). All of the reversion mutations preserved the BRC repeats, which have been shown to be essential for HR DNA repair of double-strand breaks (1, 13).

We next sought to define whether the putative BRCA2 reversion mutation identified in the cfDNA would be present in the matched tumor tissue. We obtained a tumor biopsy sample at initial diagnosis prior to carboplatin/PARP inhibitor treatment from case 1109 and a tumor sample at recurrence (i.e., synchronously with plasma sample 3) from case L031. The quality of DNA obtained from the L031 tumor biopsy was suboptimal and the targeted amplicon sequencing approach failed in this sample. Targeted amplicon sequencing of the initial diagnosis tumor sample from case 1109 confirmed the presence of the somatic FAT3, ERCC4 and KDM5C mutations identified in the cfDNA, with VAFs ranging from 16% to 55.6% (Supplementary Table S3); however, none of the putative BRCA2 reversion mutations identified in cfDNA could be detected in the tumor tissue biopsy. This suggested that reversion mutations were selected by therapy and were not detectable in the biopsied tumor prior to therapy.

Analysis of serial plasma DNA samples from one patient (L031) confirmed the presence of the putative BRCA2 reversion mutations after carboplatin treatment and prior to treatment with the PARP inhibitor talazoparib (Supplementary Table S3). The patient did not respond to talazoparib, with a differential response with some lesions unequivocally progressing. A decrease in the VAFs of the somatic putative BRCA2 reversion mutation c.402_413delTCTAAATTCTTG immediately after treatment reflected lower levels of ctDNA in the cfDNA at that time point (with a decrease in tumor-specific SPEN and TGFBR1 mutations; Fig. 4A; Supplementary Fig. S4). After subsequent treatment, a novel c.389_406delTTTCCTGTCCACTTCTAA putative BRCA2 reversion mutation, inferred to restore the open reading frame of the BRCA2 protein and initially detected at minimal levels, increased, suggesting a greater diversity in BRCA2 reversion mutations post-PARP inhibitor therapy (Fig. 4A; Supplementary Fig. S4). These results mirrored the progression of the disease and evidence of PARP inhibitor resistance in the patient.

To ascertain whether the putative somatic BRCA2 reversion mutation identified in case L031 would show interaction with PALB2 and RAD51, we expressed full-length BRCA2, the c.407delA patient-specific germline BRCA2 mutation, and the c.402_413delTCTAAATTCTTG putative BRCA2 reversion mutation in 293T cells. We observed that the BRCA2 reversion mutation but not the BRCA2 c.407delA germline mutation displayed an intact interaction with PALB2 and RAD51, which was at similar levels as those detected with the wild-type BRCA2 protein (Fig. 4B).

BRCA1 and BRCA2 reversion mutations have been documented as potential mechanisms of resistance to platinum-based chemotherapy and PARP inhibitors in cell line models and patient samples (4, 13, 14, 21, 46). Here, we report on the detection of putative BRCA1 and BRCA2 reversion somatic mutations in the cfDNA of platinum-based chemotherapy and/or PARP inhibitor–resistant/refractory ovarian and breast cancer patients harboring germline BRCA1 or BRCA2 germline mutations. We have observed these putative mutations in the cfDNA of 21% (4/19) of platinum resistant/refractory ovarian cancer patients, and 40% (2/5) of platinum and/or PARP inhibitors pretreated patients with breast cancer, suggesting that reversion mutations may not be uncommon in patients following platinum-based chemotherapy and/or PARP inhibition. The putative reversion mutations in the form of small indels restored the reading frame before the aberrant stop codon produced by the germline mutation and may have resulted in reacquisition of the DNA repair functions of BRCA1 or BRCA2 (Figs. 1 and 3). Consistent with this notion, in vitro studies revealed that three of the putative somatic BRCA1 reversion mutations identified by targeted capture sequencing of cfDNA restored, at least in part, the ability of cells to elicit BRCA1 foci following ionizing ration treatment. In addition, one of the putative BRCA2 mutations tested in vitro was found to restore the interaction with its partners PALB2 and RAD51. These putative reversion mutations could not be detected in the tumor tissue samples obtained at primary diagnosis, suggesting selection by therapy. No adequate tumor tissue was available contemporaneously with the cfDNA sample to define the frequency of these putative alterations in the resistant/refractory tumors.

Consistent with a recent report describing polyclonal reversion mutations in the cfDNA of two patients with BRCA2 prostate cancer treated with PARP inhibitors (21), our findings suggest that polyclonal reversion mutations may also be found in cfDNA of patients with BRCA1/2 ovarian and breast cancer treated with platinum-based therapy and/or PARP inhibitors, in particular in BRCA2 cancers. Importantly, however, all mutations were detected at very low allele frequencies in plasma, with more than one mutation present in 67% (4/6) of patients with reversion mutations. In two patients with BRCA2 breast cancer with disease progression after platinum-based chemotherapy and PARP inhibitor therapy, multiple concurrent somatic reversion mutations were detected by targeted capture sequencing and validated using orthogonal sequencing methods. Moreover, in one patient, the mutant allele fractions in plasma DNA increased after PARP inhibitor treatment and resistance development. These observations are consistent with the notion that resistance to targeted therapies may be polyclonal in a given patient with cancer (e.g., polyclonal ESR1 mutations as a mechanism of resistance to aromatase inhibition; ref. 47), even in therapeutic strategies based on synthetically lethal approaches. It should be noted that the VAFs of the putative reversion mutations identified in the cfDNA of subjects included here was low. Importantly, however, these VAFs were frequently similar to the allele fractions of TP53 and/or other mutations (Supplementary Table S3) and consistent with the estimated ctDNA content (Table 1), suggesting that at least, in a subset of patients, these putative reversion mutations were clonally dominant at the time of sample collection. Moreover, it is most likely that similar to other targeted therapy resistance (e.g., EGFR inhibitors in EGFR-mutant non–small cell lung cancer), resistance to platinum-based therapy and PARP inhibition may be multifactorial in a single patient (e.g., polyclonal reversion mutations and/or other mechanisms of resistance to platinum-based chemotherapy or PARP inhibitors being present in distinct subclones within a tumor), or that these putative reversion mutations may cause resistance not only in a cell autonomous manner, but also through a bystander effect.

The presence of multiple reversion mutations within a given BRCA1 or BRCA2 patient may result from the strong selective pressures imposed by platinum-based or PARP inhibitor therapy and the type of DNA repair defects cancer cells with defective BRCA1 and BRCA2 display. Consistent with this hypothesis and the more frequent reporting of polyclonal reversion mutations in BRCA2 cancer (e.g., PARP inhibitor resistant BRCA2-deficient pancreatic cancer cell line and in tumor tissue of patients with BRCA2 ovarian cancer; refs. 13, 48), loss of BRCA2 function results in defective DNA repair of double-strand breaks, as induced by platinum or PARP inhibition, and are repaired preferentially through single-strand annealing and nonhomologous end joining (1, 2). Given the selective pressure these agents provide in the context of BRCA2-deficient cells, multiple intragenic deletions could eliminate the site of the germline BRCA2 mutation and restore the open reading frame without the loss of domains essential for BRCA2 function. In contrast, nonhomologous end joining is the preferential mechanism of repair of DNA double-strand breaks in BRCA1 breast cancers (1, 2), which may be associated with a lower likelihood of multiple reversion events given the constraints of how a given germline mutation could be reversed and the maintenance of BRCA1 domains essential for its function in HR DNA repair. Further studies on the polyclonality of BRCA2 reversion mutations are required to clarify the mechanist basis of these observations.

Our exploratory, hypothesis-generating study has several limitations. Although we focused on BRCA1 and BRCA2 patients with advanced ovarian and breast cancer and performed high-depth sequencing of cfDNA, we did not detect somatic mutations in the cfDNA samples from one patient, and detected them in others at very low levels, which suggests that in a subset of platinum-based chemotherapy resistant/refractory ovarian cancer patients, the levels of tumor DNA in plasma may be limited. In fact, recent studies have found that somatic mutations, including BRCA1/2 reversion mutations, can be detected at higher allele frequencies in ascites of patients with ovarian cancer (46, 49). Second, the bioinformatics approaches employed for the identification of somatic reversion mutations were able to nominate putative somatic reversion mutations, which were successfully validated using orthogonal methods. Owing to the nature of the sequencing performed (Illumina, 100 bp reads), however, we would be unable to detect with a similar sensitivity large deletions that would result in reversion of the germline mutations. Therefore, our study may underestimate the frequency of somatic reversion mutations in the patient population analyzed. Third, we were unable to accrue tumor tissue synchronously collected with the cfDNA samples; therefore, we were unable to validate the presence of the putative reversion mutations in tumors. Fourth, the putative BRCA1 and BRCA2 reversion mutations identified in cfDNA were not tested by direct sequencing, given that the VAFs would be beyond the detection limits of Sanger sequencing. We did, however, validate these putative reversion mutations with orthogonal sequencing approaches and using distinct sequencing libraries, minimizing the likelihood of the putative reversion mutations described here constituting sequencing artifacts. Finally, owing to its small sample size, further studies are required to define the prevalence of BRCA1/2 reversion mutations detected in cfDNA and to test whether they are causative of and predict the lack of therapeutic efficacy to platinum-based chemotherapy or PARP inhibitors, ideally in the context of a prospective clinical trial or through the reanalysis of materials of a sufficiently powered prospective clinical trial.

Despite these limitations, our study, in conjunction with recent studies (21, 46), broadens the potential applications of cfDNA sequencing for the identification of somatic BRCA1 and BRCA2 reversion mutations. Further studies are warranted to define the prevalence of these reversion mutations in larger populations of patients with BRCA1 and BRCA2 ovarian and breast cancer treated with PARP inhibitors and/or platinum-based chemotherapy, to define the chronology of the emergence of these mutations and the biological impact of their potential polyclonal nature, and to ascertain their role as predictive biomarkers for these therapeutic agents.

C. Aghajanian is a consultant/advisory board member for AstraZeneca, Bayer, Cerulean, Clovis, Mateon Therapeutics, and VentiRx. D.M. Hyman reports receiving commercial research grants from AstraZeneca, Loxo Oncology, and PUMA Biotechology and is a consultant/advisory board member for AstraZeneca, Atara Biotherapeutics, Boehringer Ingelheim, Chugai, and CytomX. N.C. Turner is a consultant/advisory board member for Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B. Weigelt, J.L. Meisel, D.M. Hyman, J.S. Reis-Filho, M.E. Robson, N.C. Turner

Development of methodology: B. Weigelt, I. Comino-Méndez, I. de Bruijn, J.L. Meisel, C.K.Y. Ng, S. Piscuoglio, M.E. Arcila, J.S. Reis-Filho, R.A. Greenberg, N.C. Turner

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Weigelt, I. Comino-Méndez, L. Tian, J.L. Meisel, I. García-Murillas, C. Fribbens, L.G. Martelotto, C. Aghajanian, L. Norton, D.M. Hyman, L. Borsu, M.E. Arcila, N.C. Turner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Weigelt, I. Comino-Méndez, I. de Bruijn, L. Tian, R. Cutts, C.K.Y. Ng, R.S. Lim, P. Selenica, S. Piscuoglio, C. Aghajanian, R. Murali, M.E. Arcila, J. Konner, J.S. Reis-Filho, M.E. Robson, N.C. Turner

Writing, review, and/or revision of the manuscript: B. Weigelt, I. Comino-Méndez, I. de Bruijn, J.L. Meisel, C. Fribbens, L.G. Martelotto, C. Aghajanian, L. Norton, R. Murali, D.M. Hyman, M.E. Arcila, J. Konner, J.S. Reis-Filho, M.E. Robson, N.C. Turner

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I. Comino-Méndez, I. de Bruijn

Study supervision: B. Weigelt, J.S. Reis-Filho, M.E. Robson

We thank Julie Intrieri (MSKCC) for assistance with the dPCR assays.

Research reported in this publication was supported in part by a Basser Team Science Award by the Basser Center for BRCA, Breast Cancer Now with support from the Mary-Jean Mitchell Green Foundation, National Institute for Health Research funding to the Royal Marsden and Institute of Cancer Research Biomedical Research Centre, Meredith Israel Thomas Fund, Wooden Nickel Foundation, and a Cancer Center Support Grant of the NIH/NCI (grant no. P30CA008748). J.L. Meisel was supported by an American Society for Clinical Oncology (ASCO) Young Investigator Award (Conquer Cancer Foundation). S. Piscuoglio was funded in part by the Swiss National Science Foundation (Ambizione grant number PZ00P3_168165), L. Borsu in part by the NIH (P01-CA129243), and J.S. Reis-Filho in part by the Breast Cancer Research Foundation.

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.