Functional Ex Vivo Assay to Select Homologous Recombination–Deficient Breast Tumors for PARP Inhibitor Treatment Free

Breast cancer is the most common female cancer and the leading cause of cancer-related deaths in women (1). Great improvements have been made in breast cancer treatment with targeted therapies in estrogen receptor (ER)–positive and human epidermal growth factor receptor 2 (HER2)–positive tumors for adjuvant settings as well as treatment of metastatic breast cancer (2, 3). However, triple-negative breast cancers (TNBC), which do not express ER, progesterone receptor (PR), or HER2, have a relatively poor prognosis because of the absence of an effective targeted treatment regimen.

A particular type of TNBCs arises in familial cases of breast cancer, especially BRCA1 mutation carriers. BRCA1 mutation–associated tumors are predominantly high-grade TNBC (4). Interestingly, TNBCs arising in BRCA1 mutation carriers and sporadic TNBCs share clinicopathologic and molecular characteristics. Therefore, similar etiology is proposed for these groups of breast cancer (5–7). Hereditary breast cancer associated with BRCA1 or BRCA2 mutations show defects in the DNA damage response (DDR), which encompasses DNA damage repair, cell-cycle checkpoint signaling and apoptosis. BRCA1 or BRCA2 deficiency results in impaired double-strand break (DSB) repair by homologous recombination (HR) and chromosomal instability, which may contribute to carcinogenesis.

Dysregulation of DDR processes is a common phenomenon in cancers, sometimes highly associated with a specific type or subtype of cancer, for example, mismatch repair defects in colon carcinoma, cancer associated with crosslink repair defects in Fanconi anemia (FA), and HR defects in hereditary breast cancer (8). Interestingly, DDR defects are not only important to understand the carcinogenic process, but may also be utilized to optimize therapy response. A defective DDR pathway in tumor cells can cause dependency on another specific back up mechanism that allows cellular survival. This provides options for therapeutic intervention: specific targeting of this back up system can result in selective tumor cell death, a phenomenon referred to as “synthetic lethality” (9, 10). The major advantage of this approach is efficient tumor-specific cell killing with fewer adverse effects for the patient, because normal cells do not depend exclusively on the targeted pathway and therefore will survive the treatment. An exciting example of such a synthetic lethal approach is treatment of BRCA-deficient tumors with poly(ADP-ribose) polymerase (PARP) inhibitors (11–13).

PARP activity contributes toward signaling the presence of single-strand DNA breaks and base damage by attaching poly(ADP)ribose moieties to histones and other proteins, including itself, at the site of damage, which results in efficient repair of these types of DNA damage (14). Inhibiting PARP activity in proliferating cells results in excessive single strand and/or base lesions, which cause the collapse of replication forks and DSBs if encountered by the replication machinery (15, 16). Repair of these stalled replication fork-induced DSBs specifically requires HR (17, 18). If left unrepaired these types of lesions accumulate and cause cell death (17, 18). Therefore, HR-deficient cells associated with BRCA1 or BRCA2 mutations are extremely sensitive to PARP inhibition (8, 11, 12).

PARP inhibitor treatment showed a very effective antitumor activity in patients with BRCA mutation–associated cancers in phase I and II clinical trials (13, 19–21). Furthermore, the toxic side-effects commonly associated with conventional chemotherapy were relatively mild after PARP inhibitor treatment. Several non-BRCA mutation–associated tumor cell lines are also sensitive to this specific type of treatment, thereby extending the potential clinical application of PARP inhibitors. These cells had defects in genes, which also lead to impaired HR and/or cell-cycle checkpoints (22–26). These defects are also detected in human breast cancers (27). Unfortunately, in spite of promising experimental and clinical data, PARP inhibitors have not yet made it to breast cancer treatment in the clinic. One likely explanation for this is the lack of a marker for patient selection (except for known BRCA mutation status), because targeted treatment with PARP inhibitors can only be successful in a well-defined patient population and therefore selection of the appropriate patient population before treatment is very important.

HR activity in a tumor cell is probably the most important factor to predict whether treatment with PARP inhibitors will be successful (28). HR is mediated by the RAD51 protein that forms a nucleoprotein filament that is able to carry out the crucial strand exchange step of HR. The BRCA2 protein delivers RAD51 to DNA DSBs, where it can be detected as foci in the nucleus. Formation of RAD51 ionizing radiation induced foci (IRIF) can therefore be used as a convenient and highly informative test for HR function. Cells deficient in BRCA1, BRCA2, or a number of other HR factors do not or inefficiently form RAD51 IRIF, suggesting that this read out can be used as a PARP inhibitor sensitivity marker (28, 29). Here, we describe how ex vivo RAD51 IRIF formation capacity in primary breast tumor specimens can identify HR-deficient tumors.

Xenograft models were initiated by implanting fresh patient-derived tumor tissue subcutaneously in the thigh of immunocompromised mice. Tumors were allowed to grow out up to 2 cm in diameter and were subsequently isolated for further experiments. To maintain the specific model in vivo, tumors were isolated from mice, sliced into smaller sections, and implanted in other immunocompromised mice. BRCA1 and BRCA2 status in tumors was analyzed by immunoblotting and exon sequencing (P. ter Brugge and J. Jonkers; unpublished data).

Fresh breast tumor tissue was obtained from patients undergoing wide local excision or amputation for breast cancer at Erasmus Medical Center (Erasmus MC), Havenziekenhuis Rotterdam, and Leiden University Medical Center (LUMC) Leiden, The Netherlands. After resection, the tissue was directly transported to the Pathology department. After macroscopic investigation and determination of tumor areas for diagnostic purposes by a pathologist, left over tumor tissue was used for research purposes according to the code of proper secondary use of human tissue in the Netherlands established by the Dutch Federation of Medical Scientific Societies and approved by the local Medical Ethical committees. Specimens were coded anonymously in a way that they were not traceable back to the patient by laboratory workers. Patients receiving neo-adjuvant chemo- or radiotherapy were excluded.

Research samples were obtained within 4 hours after surgical resection and kept at 4°C during transport to the laboratory in breast medium (30) containing a 2:1 mixture of Dulbecco's modified Eagle's medium (DMEM) without phenol red and nutrient mixture F-12 (HAM) supplemented with 2% fetal bovine serum (FBS), hydrocortisone (0.3 μg/mL; Sigma), insulin (4 μg/mL; Sigma), transferrin (4 μg/mL; Sigma), 3,3′,5-triiodothyronine (1 ng/mL; Sigma), cholera toxin (7 ng/mL; Sigma), epidermal growth factor (8 ng/mL; Sigma), adenine (0.2 mg/mL; Sigma). Excess fat tissue was discarded using surgical tools and tumor specimens were depending on tissue quantity and type of experiment, sliced into 300-μm slices using a Leica vibratome 1200S or sliced manually into approximately 2-mm slices. No differences in visualization of RAD51 IRIF were observed between these slicing methods (data not shown). Slices were directly incubated in breast medium and irradiated with 5 Gy γ-radiation using a ¹³⁷Cs source (0.7 Gy/min). Incubation and irradiation of samples was performed within 6 hours after surgical resection. Subsequently, samples were incubated at 37°C and 5% CO₂ on a rotating platform (60 rpm) for 2 hours. Afterward, they were fixed in 37% neutrally buffered formalin for at least 24 hours at room temperature and subsequently embedded in paraffin (overnight procedure). Microscopy sections of 4 μm were generated and subjected to immunofluorescent staining (Supplementary Fig. S1).

Sections were deparaffinized using xylene and hydrated with declining concentrations of ethanol. Target antigen retrieval was performed using DAKO Antigen Retrieval buffer (pH 9.0 for RAD51 and pH 6.9 for others), which was heated to 100°C for 15 minutes. Cells were permeabilized using phosphate buffered saline (PBS) with 0.2% Triton X-100 for 20 minutes. For RAD51–geminin costaining, an additional DNase (1,000 U/mL; Roche Diagnostics) incubation was performed at 37°C for 1 hour. Blocking was achieved using PBS with 2% FBS and 1% bovine serum albumin (BSA). Primary antibodies [anti-RAD51 (GeneTex clone14B4 GTX70230) 1/200, anti-geminin (Proteintech Group 10802-1-AP) 1/400, anti-cleaved caspase-3 (Cell Signaling Technology 9664S) 1/100], anti-γH2AX (Millipore 2310355) 1/500, anti-53BP1 (Novus Biologicals NB100-304) 1/500, anti-P63 (Ventana clone 4A4 790-4509, ready to use)] were diluted in blocking buffer and incubated for 90 minutes at room temperature. Secondary Alexa Fluor 594 or 488 antibodies were used to visualize the primary antibody. Sections were mounted using Vectashield mounting medium with DAPI. For P63 staining, the primary antibodies were detected using DAB chromogen.

Tumor cells/areas were determined by morphology on a serial hematoxylin and eosin (H&E) stained section of the same tumor slice that was used for RAD51 foci analysis. Geminin-positive cells were counted manually. A cell was considered positive for geminin if the complete nucleus was stained by the geminin antibody. These cells were scored for the presence of RAD51 foci. A cell was considered positive for RAD51 foci if more than 5 nuclear foci were detected. The percentages of RAD51 foci–positive cells in the geminin-positive population were calculated. Approximately 100 geminin-positive cells were counted, unless sections had fewer geminin-positive cells, in which case at least 30 cells were counted in each tumor sample. To generate error bars, the standard error was estimated assuming a binomial distribution.

Statistical analysis were all 2-sided and performed using IBM SPSS statistics v21.

Genomic DNA was isolated from fresh-frozen samples of tumors using the Nucleospin Tissue Kit (Macherey-Nagel) according to the manufacturer's protocol. The percentage of tumor cells was determined by H&E staining of 5-μm cryosections of the same sample. A custom Haloplex (Agilent) Kit was used to enrich the coding regions of specific genes using 200 ng of genomic DNA to obtain sequencing libraries. Samples were tagged with a unique barcode and pooled before pair-ended sequencing 150 bp on the Illumina Miseq platform. SureCall software (Agilent) was used to detect variants. Validation of the identified mutations was performed by Sanger sequencing.

To assess promoter methylation of BRCA1, 2 MS-MLPA Kits (MRC-Holland) were used, each containing a different probe (31). Twenty-five nanograms of DNA was denatured for 10 minutes at 98°C and subsequently cooled down to 25°C. After addition of SALSA Probe-mix and MLPA buffer samples were incubated for 1 minute at 95°C followed by hybridization for 16 hours at 60°C. Next, samples were split and ligated with (methylation test) or without (copy number test) the addition of HhaI enzyme for 30 minutes at 49°C and then heated for 5 minutes at 98°C.

SALSA PCR-buffer and polymerase mix (including dNTPs, SALSA polymerase, and PCR primers) were added and samples were subjected to the following PCR reaction for 35 cycles: 30 seconds at 95°C; 30 seconds at 60°C; 60 seconds at 72°C. Finally, the PCR reaction was incubated at 72°C for 20 minutes. The amplified PCR products were separated by electrophoresis on an ABI PRISM 310 fragment analyzer (Applied Biosystems) and analyzed using Genemarker analysis software (Softgenetics).

In situ detection of BRCA1 mRNA was performed using RNAScope (ACD) using standard protocols as described by the manufacturer (32). BRCA1 probes were purchased from the same company. In short, paraffin sections were deparaffinized with xylene and samples were subjected to pretreatment steps described by the manufacturer. Next, hybridization of target probes [BRCA1 along with POLR2A (positive control) and DapB (negative control)] was achieved by incubating samples with specific probes for 2 hours at 40°C. Subsequently, several amplification steps were performed using specific amplification buffers to amplify the hybridized probe signal and visualization of this signal was achieved using Fast Red dye. Samples were counterstained with Mayer's hematoxylin solution and mounted using EcoMount mounting medium.

Cells in the S or G₂ phase of the cell cycle form RAD51 foci after DSB induction by IR. We adapted the immunofluorescence detection of these IRIF for thin breast tumor tissue slices and investigated whether we could distinguish tumors with known defects in BRCA1 or BRCA2 from other tumors by analyzing RAD51 IRIF formation. To validate our procedures, we used a collection of BRCA-positive and -negative human to mouse xenograft tumor models. Conditions for culturing tissue slices derived from these xenograft tumors were optimized (Naipal and colleagues, in preparation), slices were irradiated ex vivo, fixed after 2 hours and RAD51 IRIF formation was analyzed (Supplementary Fig. S1). The 2-hour time point was optimal for RAD51 IRIF formation (ref. 33 and Supplementary Fig. S2A). As RAD51 IRIF are only expected to occur in S- or G₂-phase cells, we also identified this cell population in the tumor slices by staining for the cell-cycle marker geminin (refs. 34 and 35; Fig. 1A). Indeed, RAD51 IRIF–positive nuclei were only found in irradiated geminin-positive cells (ref. 35 and Supplementary Fig. S2B).

In a blinded experimental setup we found all tumors without a known DDR defect (n = 4) to display prominent RAD51 IRIF in more than 50% of geminin-positive cells (Fig. 1B). On the other hand, we observed less than 10% of geminin-positive cells exhibited RAD51 IRIF in organotypic slices from tumors with known BRCA1 or BRCA2 (n = 8) defects. BRCA1 deficiency caused by a frame shift mutation or a BRCA1 promoter hypermethylation both displayed absence of RAD51 foci in xenograft tumors, showing that genetic and epigenetic modes of gene inactivation have similar RAD51 foci–deficient phenotypes in this assay. In cells without geminin expression, we did not detect RAD51 IRIF and tumors proficient for BRCA1 or BRCA2 that were not irradiated formed RAD51 foci in less than 10% of geminin-positive cells (Fig. 1A and B), showing that the assay is specific for induction of foci by IR in S- and G₂-phase cells.

We investigated whether this assay could also be used for clinical tumor specimens with a known BRCA1 defect. We obtained a tumor biopsy from a patient carrying a germline BRCA1 mutation who had developed a retrosternal recurrence after previous primary breast cancer treatment. In this tumor, only 11% of geminin-positive cells displayed RAD51 IRIF, whereas tumor slices from unselected primary tumors (n = 5) showed RAD51 IRIF in more than 50% of the geminin-positive cells (Fig. 2A and B). Results from the xenografts and the patient biopsy indicate that RAD51 IRIF can be identified in ex vivo irradiated organotypic tumor slices and that this assay can be used to discriminate HR-deficient and HR-proficient tumors.

After validation, we used this assay to identify HR defects in clinical breast cancer specimens. We collected 54 (chemotherapy naïve) tumor samples obtained from patients that underwent breast cancer surgery and generated organotypic tumor slices. Information about BRCA mutation status, family history, and pathology reports were unknown to the investigators during the analysis of the tumor samples. Pathology reports from corresponding tumors were obtained afterward (Supplementary Table S1). The majority of tumors were histologically classified as ductal carcinoma (82%; n = 44) whereas 15% (n = 8) was classified as lobular carcinoma. In addition, 93% (n = 50) of the samples expressed either ER, PR, or HER2 receptor and 7% (n = 4) of tumors had no expression of these 3 receptors (TNBC) as determined by immunohistochemical analysis. Nine tumor samples contained very low numbers of geminin-expressing cells and could therefore not be analyzed. There was no specific correlation between low geminin expression and pathologic tumor characteristics (Supplementary Table S2), suggesting that this was the result of coincidental sampling of tumor areas that were less proliferative or exhibited rapid decrease in proliferation after resection. In total, 45 tumor samples contained sufficient numbers of geminin-positive cells for RAD51 IRIF analysis.

Based on results of xenograft experiments, RAD51 IRIF formation was considered normal (positive) when more than 5 foci per nucleus were present in more than 50% of geminin-positive cells, whereas RAD51 IRIF formation was considered impaired (negative) when less than 20% of geminin-expressing cells contained more than 5 RAD51 IRIF per cell. Using these criteria, 5 tumors of 45 (11%; 95% CI, 5%–24%) showed impaired RAD51 IRIF formation (Fig. 3). We excluded technical reasons for the absence of RAD51 foci in tumor cells by showing formation of γH2AX and 53BP1 nuclear foci after irradiation (Supplementary Fig. S3A). Furthermore, normal RAD51 IRIF formation was detected in geminin-positive normal breast epithelium, stroma or fat tissue within the same tissue section (Supplementary Fig. S4). Lobular carcinomas in this cohort did not show impaired RAD51 IRIF (0 of 7), whereas tumors from the histologic subtypes classified as ductal carcinoma (3 of 35) or other (not classified ductal or lobular, 2 of 2) had impaired focus formation (Table 1). In addition, RAD51 IRIF–negative tumors were frequently classified as grade 3 carcinomas, although this correlation was not statistically significant (P = 0.419; Table 1). Three of 5 RAD51 IRIF–negative tumors were TNBC whereas the 2 other tumors expressed the ER (Table 1). Interestingly, 3 of 4 TNBC tumors in this cohort displayed impaired RAD51 IRIF formation, indicating that HR defects were more frequent in TNBC than receptor-positive BC (OR, 57; 95% CI, 3.9–825.4; P = 0.003; Table 1).

One of the 45 tumors (sample #30) in the cohort displayed great variability in RAD51 IRIF formation. RAD51 IRIF–positive cells (>5 foci/cell) were clearly recognized, but lower in number compared with other tumors. Strikingly, in some regions of the tumor RAD51 IRIF–positive cells were completely absent although geminin-positive cells were present in high numbers, similar to that of other tumors (data not shown). Overall RAD51 IRIF were observed in 38% of geminin-positive cells (Fig. 3). Notably, this tumor sample was derived from a 102-year-old patient (Supplementary Table S1); the oldest patient in this cohort. All other tumors showed little variability in RAD51 IRIF formation, indicating that analysis of a small area of the tumor will in most cases be sufficient to accurately determine RAD51 IRIF formation proficiency.

One RAD51 IRIF–negative tumor (sample #20) was incubated ex vivo with the PARP inhibitor, Olaparib. After 96 hours treatment, this tumor sample showed a clearly altered morphology with many picnotic nuclei, compared with the untreated tumor slice (Fig. 4A). The tumor cell nuclei in the treated sample were either larger in size or fragmented, shrunken, and hyperchromatic. Interestingly, the morphology of the normal mammary ducts (identified by their typical double layer of glandular cells and surrounding P63-positive myoepithelial cells) was not affected by Olaparib treatment (Fig. 4B). Subsequent staining of the sections showed increased levels of the apoptotic marker cleaved caspase-3 after 96 hours of exposure to Olaparib (Fig. 4C). In 5 tumors with normal formation of RAD51 IRIF, the altered morphology and induction of cleaved caspase-3 was not noticed after this treatment, strongly suggesting that induction of apoptosis was because of PARP inhibitor treatment and not caused by declining tissue viability (Fig. 4A and C).

PARP inhibition results in DSBs during replication. This was shown by the formation of γH2AX and 53BP1 nuclear foci (Supplementary Fig. S3B). As a consequence we expected an activated HR pathway and thus formation of RAD51 foci in HR-proficient cells after PARP inhibitor treatment. Accordingly, RAD51 foci induced by Olaparib treatment were present in normal mammary epithelium and not in tumor cells in the same tissue slice (Fig. 4B). These results indicate that the functional HR defect in these tumor cells caused sensitivity to PARP inhibitors, whereas normal epithelial cells were not affected, opening perspectives for using this assay as a functional test for clinical sensitivity.

Subsequently, we determined the basis for impaired RAD51 IRIF formation in the tumors by genetic analysis for the BRCA1 and BRCA2 genes in these tumors. We sequenced more than 99% of the BRCA1, BRCA2, and TP53 exons and flanking intron sequences and found that 3 RAD51 IRIF–negative tumors harbored a mutation in the BRCA2 gene (Table 2). Two of these tumors were ER positive and one was a TNBC (Table 2). In sample #2, we identified a G to T mutation at the splice donor site of intron 15 (c.7617 + 1G > T, NM_000059) causing aberrant splicing with skipping of exon 15 as a result. This mutation was detected hemi/homozygously in the tumor, whereas normal breast tissue from the same patient was heterozygous for this mutation. This showed that the patient carried a germline mutation in the BRCA2 gene (Supplementary Fig. S5A). Sample #54 harbored a known pathogenic missense mutation (c.9154C > T, p.Arg3052Trp, NM_000059) in the BRCA2 gene. Interestingly, this specific sample was obtained from the only male patient in this cohort. The other BRCA2 mutation was detected in sample #32, a TNBC (Table 2). This specific mutation (c.517G > C, p.Gly173Arg, NM_000059) at the intron–exon junction alters the first nucleotide of exon 7, which might abrogate splicing of exon 7. Gene sequencing of the 2 remaining tumors (sample #1 and #20), both TNBC, did not reveal mutations in the BRCA1 and BRCA2 genes but did reveal a mutation in TP53, which is very often detected in TNBC (ref. 36; Table 2).

BRCA1 promoter hypermethylation can lead to reduced BRCA1 protein expression and lack of RAD51 IRIF formation (ref. 37 and Fig. 1). Therefore, we analyzed hypermethylation of the BRCA1 promoter in the RAD51 IRIF–negative tumors. Sample #1 and #20, both displayed hypermethylation in the promoter sequence of the BRCA1 gene (Table 2 and Supplementary Fig. S5B). This was not observed in normal mammary tissue from the same patients, suggesting that impaired RAD51 IRIF formation in the tumor was caused by BRCA1 promoter hypermethylation (Supplementary Fig. S5B). BRCA1 promoter methylation was not detected in the other 3 RAD51 IRIF–negative tumors, nor in 10 random RAD51 IRIF-positive tumors (data not shown). To assess the functionality of BRCA1 promoter hypermethylation, we performed in situ detection of BRCA1 mRNA. As expected, we did not detect BRCA1 mRNA in sample #1 and #20, whereas BRCA1 mRNA was readily observed in unmethylated tumors (Supplementary Fig. S6 and data not shown). This confirms that BRCA1 promoter methylation caused BRCA1 silencing and impaired RAD51 IRIF formation.

The remaining TNBC with normal RAD51 IRIF formation (sample #62) did not harbor a mutation in BRCA1 or BRCA2, neither did it show hypermethylation of the BRCA1 promoter (Table 2). The only sample that showed intermediate levels of RAD51 IRIF also did not harbor a mutation in BRCA1 or BRCA2 and showed no hypermethylation of the BRCA1 promoter (Table 2). In conclusion, all tumors with impaired RAD51 IRIF in this cohort were found to harbor BRCA1 or BRCA2 defects.

Here we describe an assay to identify HR-deficient tumors based on RAD51 IRIF formation in organotypic tumor slices ex vivo. The assay was validated in xenograft tumors with defective or normal BRCA1/2 gene expression, where the absence of RAD51 IRIF perfectly correlated with BRCA gene status. We used this approach to identify a subgroup of HR-deficient tumors in patients with primary breast cancer and found that approximately 10% of primary breast tumors has a clearly impaired HR repair capacity based on this assay. This percentage is lower than some estimates described in other publications, which report that up to 25% of sporadic breast cancers have a BRCAness phenotype and might be related to a possible HR deficiency (38). Other research groups report even higher percentages of primary breast tumors to have impaired HR based on a RAD51 focus formation assay (35, 39, 40). There are several explanations for these discrepancies.

One explanation could be that there are differences in the methods used to induce, visualize, and characterize RAD51 foci in tumor samples. In some studies, the foci were induced by DNA damage caused by in vivo administration of chemotherapy to patients with breast cancer (35, 40). The foci were visualized in tumor biopsies of these patients obtained 24 hours after the first dose of chemotherapy. This method might result in an overestimation of HR-deficient tumors, because efficient DNA repair will result in relatively low levels of residual foci at 24 hours after treatment (33, 41). Therefore, a low number of residual RAD51 foci does not necessarily mean impaired focus formation. In contrast, we induced DNA damage by IR of tumor samples ex vivo and subsequent culturing for 2 hours before RAD51 foci were visualized. In cell culture and ex vivo tissue culture, the number of RAD51 foci peaks 2 hours after DNA damage treatment (33, 41). Thus, in contrast to detecting RAD51 foci 24 hours after the in vivo administration of chemotherapy, our assay specifically detects the ability to form RAD51 foci in the tumor samples. Moreover, after robust validation of the ex vivo irradiation approach, we can state with high confidence that this assay faithfully discriminates HR-deficient from HR-proficient tumors.

Another methodological difference that could lead to an overestimation of HR-deficient tumors is the fact that some research groups analyze the formation of RAD51 foci without adjusting for proliferating cells in the tumor samples. As HR is only active during the S and G₂ phases of the cell cycle, RAD51 focus formation is only expected in these cell-cycle phases. Thus, tumor samples with very few cells in the S–G₂ phases will have low levels of RAD51 foci, but these tumors are not necessarily impaired in RAD51 IRIF formation. Therefore, we only score RAD51 foci in cells expressing geminin, which is a marker for the S and G₂ phases of the cell cycle (34, 35, 42, 43). We identified some tumor samples having no or very few cells with geminin expression. This might be a result of a very low-proliferating tumor area or coincidental sampling of a part of the tumor that rapidly declined in proliferation after surgical resection. Therefore, we exclude samples with low geminin expression (less than 30 geminin-positive cells) from quantitative analysis to prevent inappropriate designation of tumors as HR deficient.

HR defects in ER-positive BC might be indicative for a BRCA2 mutation (38). The fact that we identify BRCA2 mutations in the 2 HR-deficient ER-positive tumors is therefore within expectations. On the other hand, TNBC more frequently harbor a mutation in BRCA genes and among TNBC the incidence of BRCA1 mutations is higher than BRCA2 mutations (38, 44, 45). In the RAD51 IRIF-negative TNBCs, we did not identify mutations in the BRCA1 gene, but instead found hypermethylation of the BRCA1 promoter as a cause for the RAD51 IRIF defect. Therefore, the absence of RAD51 IRIF in the tumor could in all cases be explained by the deficiency of BRCA1 or BRCA2. Thus, mutation screening of BRCA1 and BRCA2 in combination with methylation analysis of the BRCA1 promoter, would have been sufficient to identify these tumors. However, other causes for HR deficiency in primary BC have been described in literature (22, 46, 47). These specific defects are probably less frequently observed in the population but are expected to be present when screening a larger cohort of primary BC by the RAD51 IRIF assay.

Nevertheless, the observed frequency of RAD51 IRIF deficiency in TNBC suggests that this subgroup of breast tumors should benefit the most from PARP inhibitor treatment. However, in a phase II clinical trial, treatment of non–BRCA-associated advanced TNBC with a daily dose of Olaparib did not result in objective responses (21). There are several possible explanations for the lack of response. The sample size was small and it is likely that not all TNBC are HR deficient. In addition, the fact that the patients included in this clinical trial had previously been treated with several cycles and types of chemotherapy, probably with DNA-damaging agents, might have caused a selection of resistant tumor cells that are also resistant to PARP inhibitors.

A possible mechanism leading to PARP inhibitor resistance in BRCA1-deficient cells is 53BP1 loss, which also restores RAD51 focus formation (48). Other mechanisms causing resistance to PARP inhibitors are secondary mutations in BRCA genes that are able to restore the open reading frame and result in transcription of functional isoforms of BRCA proteins (49). We therefore argue that the ex vivo assay will also be a very useful tool to discriminate tumors that acquired resistance by these mechanisms.

Currently, phase III clinical trials are being conducted for different PARP inhibitors. However, patient selection is based on germline BRCA mutations. Other assays to determine HR status have been proposed and certain trials take this into account, for example, Myriad's HRD assay. The advantage of this assay is that it can be performed on formalin-fixed paraffin-embedded (FFPE) material and no fresh viable tissue is needed. It measures loss of heterozygosity caused by HR deficiency in the tumor (50). This gives a historic overview of genomic aberrations acquired by the tumor over time. Although the RAD51 IRIF assay can only be performed on fresh tumor material, it provides a functional analysis of HR at the moment of sampling, that might also discriminate tumors that have acquired resistance to PARP inhibitors or other DNA damaging drugs.

Concluding, we show that functional assessment of HR in breast tumors, by ex vivo determination of RAD51 IRIF formation in organotypic slices, provides a unique chance to identify a sizeable fraction of HR-deficient tumors among unselected primary breast tumors. This has a clear advantage over gene sequencing as more than only BRCA mutation associated tumors are identified as HR deficient. Based on the study presented here, we expect approximately 10% of all patients with breast cancer to be eligible for PARP inhibitor treatment. Furthermore, other chemotherapeutic treatments, causing DNA damage that requires HR for its repair, could also be considered for this subgroup of mammary tumors. Most notably, the interstrand crosslinking agent cis-Platin or the topoisomerase I inhibitor Doxorubicin are expected to have efficient cell killing capacity in this category of tumors. Therefore, this assay grants unique opportunities to select patients for clinical trials with PARP inhibitors and to facilitate optimal selection of current standard treatment options.

No potential conflicts of interest were disclosed.

Conception and design: K.A.T. Naipal, N.S. Verkaik, M.J. O'Connor, H. Vrieling, J.H.J. Hoeijmakers, R. Kanaar, M.P. Vreeswijk, A. Jager, D.C. van Gent

Development of methodology: K.A.T. Naipal, N.S. Verkaik, M. Meijers, H. Vrieling, M.P. Vreeswijk, A. Jager, D.C. van Gent

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.A.T. Naipal, N.S. Verkaik, C.H.M. van Deurzen, P. ter Brugge, A.M. Sieuwerts, J.W. Martens, J. Jonkers, M.P. Vreeswijk

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.A.T. Naipal, N.S. Verkaik, N. Ameziane, M. Meijers, J.W. Martens, H. Vrieling, M.P. Vreeswijk, A. Jager, D.C. van Gent, J. de Winter

Writing, review, and/or revision of the manuscript: K.A.T. Naipal, N.S. Verkaik, N. Ameziane, C.H.M. van Deurzen, J.W. Martens, H. Vrieling, J.H.J. Hoeijmakers, R. Kanaar, A. Jager, D.C. van Gent, J. de Winter

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.A.T. Naipal, N.S. Verkaik, N. Ameziane, M. Meijers

Study supervision: R. Kanaar, M.P. Vreeswijk, A. Jager, D.C. van Gent

The authors thank Drs. R.A. Tollenaar, W.E. Mesker, and V.T. Smit (Leiden University Medical Center) for the collection of patient tumor material, N.C. Turner (Institute of Cancer Research, UK) for assistance with the RAD51/geminin immunofluorescence staining, G. Verjans and S. Getu (Department of Viroscience, Erasmus, MC) for assisting with in situ RNA detection assays, and J. Bartek and J. Bartkova (Danish Cancer Society) for useful discussions. J.H.J. Hoeijmakers acknowledges support from the Royal Academy of Arts and Sciences of the Netherlands (academia professorship) and an advanced research grant from the European Research Council.

The research leading to these results has received funding from the European Community's Seventh Framework Programme (FP7/2007–2013) under grant agreement No. HEALTH-F2-2010-259893 and from the Dutch Cancer Society (grant EMCR 2008-4045 and a Ride for the Roses Cancer Research Grant).

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.