Functional Ex Vivo Assay Reveals Homologous Recombination Deficiency in Breast Cancer Beyond BRCA Gene Defects

Breast cancer is the most common malignancy in women with the second highest cancer-related mortality rate (1). Approximately 3% of all breast cancer cases are due to germline mutations in BRCA1/2 (2), and in triple-negative breast cancers (TNBC) this percentage is even 10% to 20% (3). The BRCA proteins play an important role in the homologous recombination (HR) pathway, the error-free DNA double-strand break (DSB) repair pathway that operates during the S- and G₂-phase of the cell cycle. HR deficiency (HRD) leading to impaired DNA DSB repair is frequently caused by, but not limited to, defects in BRCA1/2 (4).

Therapies specifically targeting tumor cells with impaired HR capacity are PARP inhibitors (PARPi), as well as classical chemotherapies such as platinum-derivates and alkylating agents (5). PARPi causes persistence of single-strand DNA breaks (SSB) by trapping PARP1 on DNA, whereas platinum-derivates cause DNA interstrand crosslinks. Both types of lesions result in replication fork stalling and/or collapse, frequently leading to DSBs that need HR for their repair (5). The targeted approach of PARPi kills tumor cells lacking HR, whereas normal cells remain unharmed, due to their normal DSB repair capacity, a phenomenon often referred to as synthetic lethality. Recently, FDA approval was granted for the use of Olaparib in germline BRCA mutated breast cancer based on the results of the Olympiad trial (6).

Although evidence is emerging that the use of PARPi could be extended beyond germline BRCA1/2 mutated cancers to sporadic cancers with BRCA-like features, a gold standard test for predicting response to treatments targeting HR is not yet available (7). Several different HRD tests exist, mostly based on genomic patterns or transcriptional predictors of BRCAness (8–12).

These genomic tests measure the accumulation of mutations and chromosomal aberrations over time, but not necessarily reflect the real-time HR status. Beyond mutational status, several other factors influence tumor behavior and therapy response, such as epigenetic changes and the microenvironment of the tumor cells. Independent of the underlying cause, the downstream effect of HR impairment (phenotype) can be assessed functionally. A functional diagnostic assay therefore has the potential for more precisely detecting patients who may benefit from PARPi than genomic assays.

A functional HRD assay was first described by Graeser et al., assessing RAD51 focus formation, a marker of HR competence, in tumor biopsies obtained 24 hours after in vivo anthracycline treatment (13). This provided the first evidence that RAD51 focus formation can serve as a predictive biomarker. To enhance clinical utility of this biomarker, test outcomes should be available before start of treatment. Therefore, we developed the homologous recombination REpair CAPacity (RECAP) test exploiting the formation of RAD51 foci in proliferating cells after ex vivo irradiation of fresh breast cancer tissue, providing a real-time HR status of the tumor (14). The aim of this study was to validate the RECAP test in an extensive cohort of primary breast cancers and provide evidence that this functional test is achievable in a pseudo-clinical setting. Additionally, thorough molecular characterization of the HRD phenotype is performed, proving that HRD tumors encompass more than only BRCA deficiencies.

Residual fresh breast cancer tissue was prospectively collected from lumpectomy of the breast or mastectomy specimens in the Erasmus MC Cancer Institute, Haven hospital, and Maasstad hospital in Rotterdam, the Netherlands, between 1 October 2011 and 1 September 2016. The first 41 patients were also included in our previous cohort (14). After macroscopic evaluation of the surgical specimen by trained pathologists, residual tumor tissue was collected for our 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 (MEC-11-098). Patients who had objected to secondary use of residual tumor material for research purposes were not included in this study. Patients with ductal carcinoma in situ (DCIS) only or patients receiving neo-adjuvant chemotherapy were excluded.

Obtained tissue samples were immediately transferred into customized breast tissue culture medium, as described in Naipal et al. (14). Processing of samples was performed within 4 hours after the tissue was resected. Microscopic analysis of hematoxylin and eosin (HE) stained sections was performed to determine presence of invasive carcinoma. The RECAP test, a functional assay exploiting the formation of RAD51 foci in proliferating cells after ex vivo irradiation of fresh breast cancer tissue, was performed and results were analyzed as described previously (14). In brief, presence of RAD51 foci was determined in S–G₂ cells only, which stain positive for Geminin. At least 30 Geminin expressing cells were counted per tumor sample. A cell was considered RAD51 positive when at least five RAD51 foci could be detected. Based on previous experiments with patient-derived xenograft (PDX) models with known BRCA status, tumors were classified as HR proficient (HRP), HR deficient (HRD), or intermediate (HRi) when more than 50%, less than 20% or between 20% and 50% of geminin positive cells showed ≥5 RAD51 foci, respectively.

To unravel the possible molecular mechanism underlying the HRD phenotype, several molecular tests were performed retrospectively (Fig. 2; Supplementary Fig. S1). As no DNA could be obtained for one HRD sample, we conducted the analyses for 23 HRD and six HRi samples. First, BRCA sequencing and BRCA1 promoter methylation analysis was performed in HRD and HRi samples, as well as in all TNBC (n = 5), ER/PR− HER2+ (n = 2), and 21 ER/PR+ HRP tumors (total n = 28; Supplementary Fig. S1). The HRD and HRi tumors without molecular explanation for their phenotype were subjected to BRCA1 and BRCA2 MLPA analysis to identify large genomic rearrangements (LGR), as LGRs are not usually identified by targeted sequencing. In addition to this targeted approach, morphologic examination of TILs and whole exome sequencing (WES) was performed on a selection of tumors to further explore molecular aspects connected to the HRD phenotype.

Isolation of DNA from 30 μm fresh frozen tissue section samples was performed using the NucleoSpin Tissue Kit (Macherey-Nagel) according to the manufacturer's instructions. Quantity and quality checks of isolated DNA were performed using the MultiNA microchip electrophoresis system (Shimadzu's Hertogenbosch), Nanodrop 2000-v.1 (Thermo Fisher Scientific), and Qubit (Thermo Fisher Scientific).

Ion semiconductor sequencing on the Ion Torrent Personal S5XL was performed according to manufacturer's instructions (Thermo Fisher Scientific). Adapter-ligated libraries were constructed using the AmpliSeq Library Kit 2.0 with amplicons designed targeting BRCA1/2 and TP53. Generation of sequence reads, trimming adapter sequences, filtering, and removal of poor signal-profile reads was performed via the Ion Torrent platform-specific pipeline software Torrent Suite v5.2.2. Initial variant calling was performed by comparison to the reference genome hg19 (build 37) using the “Torrent Variant Caller v5.2.0.34″ plug-in from the Torrent Suite Software. All BRCA2 variants were validated by Sanger sequencing and pathogenicity was evaluated using interactive Biosoftware Alamut Visual v.2.7.2. BRCA1 promoter methylation was analyzed as previously described (15). Multiplex ligation-dependent probe amplification (MLPA) analysis of BRCA1 and BRCA2 was undertaken to identify large rearrangements using the SALSA MLPA Kit P002B, and for confirmation of observed abnormalities, the SALSA MLPA Kit P087 was used (MRC Holland). Analyses were performed according to the manufacturer's instruction; products were run on an ABI automated sequencer (ABI 3730XL), and the data were analyzed by Genemarker version 2.7.0 (Softgenetics).

In situ detection of BRCA1 mRNA was performed using RNAScope (Advanced Cell Diagnostics) on the automated Ventana Discovery Ultra system (Ventana Medical Systems). BRCA1 and positive control peptidylprolyl isomerase B (PPIB) probes (product codes: 485479 and 313909) were purchased from the same company. RNAscope analysis was performed according to manufacturer's instructions using the reagent kit (VS Reagent Kit 320600; Advanced Cell Diagnostics) on proteinase K (0.1%, 5 min at 37 °C)-treated paraffin sections (4 μm).

DNA libraries for Illumina sequencing were generated using standard protocols (Illumina) and subsequently sequenced in an Illumina HiSeq 2500 system by GATC-biotech. Exome-targeting was performed using the Sureselect v5 (v6 for tumors M077, M209, M211) methods using standard protocols (Agilent Technologies). DNA libraries were whole-exome sequenced (2 × 125bp) using the HiSeq v4 paired-end sequencing protocol to a minimum depth base coverage of 90× for tumor samples and 60× for matched normal. Sequence reads were mapped against human reference genome GRCh37 using Burrows–Wheeler Aligner (v0.7.12) with default settings (16). Sequence reads originating from multiple lanes were merged after alignment using Samtools (v1.5) prior to further analysis (17). Sequence duplicates were marked using PicardTools (v1.129; ref. 18). Somatic variant calling was performed by Mutect2 (v3.7) using a matched-normal design while utilizing the dbSNP (v149, hg19) and COSMIC (v80, hg19) databases and using default settings (19–21). Variant annotation was performed by ANNOVAR (22). Heuristic filtering removed variants not passing all standard Mutect2 post-calling filters. Sequence data have been deposited at the European Genome-phenome Archive (EGA, http://www.ebi.ac.uk/ega/), which is hosted by the EBI, under accession number EGAD00001003929.

For each somatic variant, its trinucleotide context was derived from the human reference genome GRCh37 and enumerated into a mutational spectrum matrix Mij (i = 96; number of trinucleotide contexts; j = number of samples) using the MutationalPatterns R package (v1.4.0) in the R statistical platform (23). Multi-allelic and InDel variants were not included in this analysis. The 30 consensus mutational signatures, as established by Alexandrov and colleagues, (matrix Sij; i = 96; number of trinucleotide motifs; j = number of signatures) were downloaded from COSMIC (as visited on 8-11-2017; ref. 24). Per sample, a constrained linear combination of the 30 validated mutational signatures was constructed, which reconstructs the sample-specific mutational spectrum, using nonnegative least squares regression implemented in the R package pracma (v1.9.3). Signatures with lower relative contribution than 3% were summarized into a “Filtered” category.

These analyses were performed as previously described (25).

Tumor-infiltrating lymphocytes (TIL) were scored on HE stained sections, according to the consensus by the International TILs Working Group 2014 (26).

Statistical analyses were all two-sided and performed using IBM SPSS statistics v21. Significance was calculated by Fisher exact test for categorical data, by Mann–Whitney test for continuous data, and by exact binomial test for the incidence of MSI. P-values of <0.05 were considered significant.

A total of 170 samples were subjected to the RECAP test (Supplementary Fig. S1; ref. 14). In 125 of 170 (74%) primary breast cancer tissues RECAP was completed successfully (Fig. 1). In all cases, the reason for failure (n = 45) was lack of proliferating tumor cells. No differences in clinicopathologic characteristics between tumors that yielded successful versus nonsuccessful tests were observed (Supplementary Table S1). The first 41 patients were also included in our previous cohort. Here, we show that execution of a functional assay in a pseudo-diagnostic setting is achievable and validate the findings from the earlier cohort (11% HRD in cohort 1 vs. 19% HRD in cohort 2, P = 0.339). In total, we identified 95 (76%) HRP, 24 (19%) HRD, and 6 (5%) HRi samples (Fig. 1). Both HRi and HRD tumors were more frequently TN (P < 0.001) and Bloom and Richardson (B&R) grade 3 (P < 0.001) than HRP tumors. Also, HRD tumors had a larger size (P = 0.050) and were never B&R grade 1 (Supplementary Table S2).

Pathogenic BRCA2 mutations were found in six (6/23 = 26%) HRD samples, but not in the tested HRP samples (Table 1). In six tumors (two HRD, one HRi, and three HRP), BRCA2 variants were detected that were classified as benign. We did not identify any BRCA1 point mutations. Next, BRCA1 promoter hypermethylation was detected in six (6/23 = 26%) HRD tumors and in one HRi tumor, but not in tested HRP samples (Table 1). Interestingly, all six HRD tumors displaying BRCA1 promoter hypermethylation were TNBCs (Table 1; Supplementary Table S3). Vice versa, of the six HRD tumors harboring a pathogenic BRCA2 mutation, five showed hormone receptor positivity.

Thus, 12/23 HRD and 1/6 HRi samples were explained by a BRCA2 mutation or BRCA1 promoter hypermethylation. Subsequently, we proceeded with an MLPA analysis for BRCA1 and BRCA2 on the 16 HRD and HRi samples that remained unexplained to identify possible LGRs. Large BRCA1 deletions were found in four samples and BRCA2 deletions in two samples (Table 1), of which one tumor harbored both a BRCA1 and BRCA2 deletion. Two tumors [M248 (HRD) and M112 (HRi)] showed extensive chromosomal instability as they contained a mosaic BRCA1 deletion (meaning the deletion was present in a subclone of the tumor) and a BRCA2 duplication (Table 1). In total, BRCA defects have thus been identified in 16/23 HRD and 2/6 HRi samples.

Silencing of BRCA1 was validated in tumors with BRCA1 promoter hypermethylation (n = 7) and BRCA1 LGRs (n = 4) by RNAscope in situ RNA hybridization (Supplementary Fig. S2; Table 1). All tumors displaying BRCA1 promoter hypermethylation showed absence of BRCA1 RNA, except for two heterogeneous samples that contained BRCA1 positive and negative areas (Supplementary Fig. S2). As expected, the BRCA1 deletion in tumor M094 led to a total absence of BRCA1 mRNA (Supplementary Fig. S2). The mosaic BRCA1 deletions did not result in complete BRCA1 silencing (Supplementary Fig. S2). Neither did the BRCA1 deletion in M232, however this tumor also harbored a BRCA2 deletion that can explain the HRD phenotype.

After thorough analysis of the BRCA genes using several techniques, 13/23 HRD and 1/6 HRi samples could be explained by deficiencies in BRCA. Moreover, 3/23 HRD (M131, M277, and M248) and 1/6 HRi (M112) tumors also harbored BRCA deficiencies (BRCA1 promoter methylation or mosaic BRCA1 deletions), which explain the HRi and partially the HRD phenotype, as these tumors showed heterogeneous BRCA1 mRNA expression. Finally, 7/23 HRD and 4/6 HRi tumors did not show any BRCA defects and therefore remained unexplained (Fig. 2). To further characterize these HRD tumors, functional features correlating with the HRD phenotype were determined.

Recently, a subgroup of patients with TNBC was identified who showed good response to immune checkpoint inhibition through programmed death-ligand 1 (PD-L1) blockade (27, 28). This subgroup was characterized by having >10% TILs and high CD8 lymphocyte counts in the tumor centers (28). Because 11/17 TNBCs in our study showed HRD, we hypothesized that the RECAP test might select for a specific subgroup of patients with TNBC who might benefit from PD-L1 therapy. We found that significantly more HRD tumors (6/22) had >10% TILs than HRP tumors (0/28) (P = 0.004; Fig. 3). Also, tumors with BRCA defects showed more frequently >10% TILs compared with non-BRCA tumors (P = 0.001), which was also true for TNBCs compared with ER/PR+ tumors (P = 0.026).

Next, WES was performed to determine the molecular landscape of HRD tumors. A selection of HRD (n = 8) and HRi (n = 3) tumors with BRCA1/2 mutations/deletions, BRCA1 promoter hypermethylation, and BRCA WT tumors and one HRP tumor were subjected to WES. First, mutational load was determined in these tumors and high mutational load did not correlate with high numbers of TILs. Second, we did not identify commonly mutated genes other than BRCA1/2, which might explain the functional HR defect. Third, WES data were used to identify mutational signatures which are specific combinations of mutations that arise due to a certain underlying mutational or DNA repair processes (29).

Mutational signatures were derived from the WES data from HRD/HRi tumors of which matching normal DNA was available to filter out germline variants (n = 10) to explore novel mechanisms related to or underlying the HRD phenotype. Because discussion in the field exists that mutational signatures can only be faithfully obtained from WGS instead of WES data, we first carried out a pilot experiment comparing mutational signature analysis using all somatic mutations in five BrC WGS datasets (30) and filtered these WGS datasets to only contain somatic mutations on exonic regions. Both methods resulted in similar distributions of the mutational signatures (Supplementary Fig. S3).

Mutational signature 3 is related to failure of DSB repair by HR and associated with germline and somatic BRCA1/2 defects in breast, pancreatic, and ovarian cancers (31). Signature 3 was present in 6 of 10 analyzed samples (M94, M95, M119, M131, M141, and M221; Fig. 4). APOBEC-related mutagenesis (predominantly C>G or C>T substitutions in TCA or TCT motifs) is captured in signatures 2 and 13, which arise through activity of the AID/APOBEC family. BRCA-related signatures could also be identified to a lesser extent in samples (M211 and M094) having a high mutational burden of signatures 2 and 13.

One tumor (M077) showed a high mutational load and high contributions of signatures 15, 20, and 26, which are related to microsatellite instability (MSI) and mismatch repair (MMR) deficiency (http://cancer.sanger.ac.uk/cosmic/signatures). MSI is a common feature in endometrial and gastro-intestinal cancers, caused by either germline (Lynch syndrome) or somatic mutations in one of the MMR genes and/or promoter hypermethylation (32, 33). Presence of MSI in tumor M077 was confirmed using pentaplex PCR and IHC for the MMR proteins, which showed absence of MLH1 and PMS2, caused by MLH1 promoter methylation (Supplementary Fig. S4).

A set of 44 tumors (20 HRD, 6 HRi, 18 HRP) was subjected to MSI analysis by pentaplex PCR and IHC for the MMR proteins. One HRD tumor (M188) of which mutational signatures were not available, also showed MSI. MSI was never detected in any HRP tumors (Fig. 5). Tumor M188 showed absence of MSH2 and MSH6, caused by a homozygous deletion of MSH2 (Supplementary Fig. S4). The two MSI BrCs (M077 and M188) were TN and BRCA WT and ER positive and BRCA2 mutated, respectively (Fig. 2). The incidence of MSI within the HRD group (2/20, 10%) is significantly higher than the incidence of MSI in the unselected breast cancer population (1%; P = 0.017; refs. 34, 35).

Here, a unique series of fresh primary breast cancer tissues (n = 125) has been analyzed for HRD using the functional RECAP test. This first large validation study describes that functional assessment of HR in a pseudo-diagnostic setting is achievable and produces robust and interpretable results for most patients (74%). We found that the percentage of HRD tumors detected by the RECAP test is similar in this larger cohort, as compared with our previous report. Therefore, both cohorts were combined to achieve more power to thoroughly investigate the molecular mechanism underlying the HRD phenotype. Sixteen HRD samples showed deficiencies in BRCA1/2 (BRCA mutations, deletions, or promoter hypermethylation), whereas seven HRD tumors were non-BRCA related, demonstrating that HRD tumors encompass more than only BRCA-deficient tumors.

Several different HRD tests have been designed to identify HRD tumors in addition to the BRCA mutated or promoter methylated tumors to enlarge the population of patients with breast cancer that could benefit from treatments targeting the HR pathway. For example, the BRCAness classifier, which is based on specific genomic patterns derived from copy number data of BRCA1/2 mutated breast cancers that also occur in sporadic cancers (10, 11) and the Myriad MyChoice HRD test, which is a combined score of three different structural chromosomal aberrations [telomeric allelic imbalance (TAI), large-scale transition (LST), and loss of heterozygosity (LOH); ref. 36]. Both the BRCAness classifier and MyChoice are robust, easily applicable in the clinic, and have also been validated to predict in vivo response to high dose chemotherapy and neo-adjuvant platinum-based therapy, respectively, in patients with TNBC (37, 38). As opposed to the neo-adjuvant setting, the MyChoice HRD test did not predict response to PARPi therapy in platinum-sensitive recurrent ovarian cancer (39). These genomic HRD tests have the drawback that they do not determine the real-time HR status, also it remains unclear whether all HRD cases are identified. In theory, the functional RECAP test can also detect reversion of the HRD phenotype in BRCA-deficient tumors, that have been treated with various DNA damaging chemotherapies that may have induced resistance. Moreover, the BRCAness classifier focuses on TNBC only, whereas the RECAP test identifies HRD independent of hormonal status. More recently, a HRD test based on several genomic signatures has been published, HRDetect (40). This test has recently been shown to predict in vivo response to platinum-based therapies in advanced breast cancer (41). HRDetect relies on whole genome sequencing and is therefore more expensive and has a longer turnaround time for biopsy results, hampering its clinical implementation. Furthermore, the tumor cell percentage needs to be above 50% for reliable results, which is not a prerequisite for the RECAP test. However, HRDetect has the advantage that it can be performed on frozen material, whereas the RECAP test requires fresh material. The percentage of non-BRCA HRD tumors (approximately 33%) detected by the HRDetect and the RECAP test are quite comparable, although it remains elusive whether these tests identify the same tumors, therefore comparison of several HRD tests within the same patient cohort is required.

The major strength of this study is that functional diagnostics have been applied to a unprecedented large collection of tumors. The advantage of the RECAP test over genetic tests, is its functional character for exploring the HR phenotype rather than the static nature of genomic tests. Also, the RECAP test is feasible in samples with low tumor percentage, because the microscopic read-out allows differentiation between tumor and stromal cells. The RECAP test has a high success rate and results are available within one week after the biopsy procedure. In this study, reproducibility and robustness of the RECAP test is validated in an independent set of 129 tumors. For the molecular analyses to unravel the mechanisms underlying the HRD phenotype, the 41 samples included in our previous publication were also included, to achieve more power. The main limitation of this study is that although prospective trials evaluating the predictive value of RECAP for in vivo patient response to PARPi have been initiated, results are not yet available. However, previously a functional RAD51 test performed on biopsies obtained from patients 24 hours after start of therapy correlated with response to anthracycline-based therapies, indicating that functional assessment of HR can have predictive value for therapy response (13).

Among the spectrum of BRCA defects, we have not identified any BRCA1 mutations. This is somewhat remarkable but could be due to a selection bias, as the Erasmus Medical Center is specialized in hereditary BrC and all patients with TNBC are tested, therefore most families with hereditary BRCA1 mutations have been identified and carriers are offered strict screening programs or undergo prophylactic surgery in The Netherlands (42). Also, in this study there is a selection bias for tumor size, as the tumor should be large enough to provide residual material without compromising standard diagnostic procedures. Since BRCA mutation carriers are offered strict screening programs, tumors are often identified at an early stage and residual material is not available for the RECAP test. Also, many BRCA1 mutation carriers with TNBC are treated with neoadjuvant chemotherapy, which was an exclusion criterion for this study.

Clinical consequences of BRCA1 promoter methylation are unclear (43). In the current study, BRCA1 promoter methylation resulted in absence of BRCA1 RNA in four samples, but in two tumors there was still heterogeneous expression of BRCA1 RNA (Supplementary Fig. S2). In these tumors, the percentage of cells with RAD51 foci was 1% and 2%, respectively. The discrepancy between the very low HRD score and heterogeneous BRCA1 RNA could be explained by sampling from different areas of the tumor, since the tumor sample for the RECAP test was irradiated and an unirradiated tumor sample was used for molecular analyses. This sampling error is not limited to this study, but also occurs in regular diagnostics when biopsies are obtained from a certain region of a heterogeneous tumor. Tumor heterogeneity in BRCA1 promoter methylated tumors is very important for clinical decisions on PARPi use. If subsequent studies reveal that this phenomenon is observed in a large fraction of these tumors, PARPi may not be very effective in tumors with BRCA1 promoter methylation.

We identified 6 HRi BrCs in the current cohort. Only in one of the HRi tumors, distinct areas of RAD51 negative and RAD51 positive tumor cells were observed, suggesting clonal heterogeneity. The other HRi tumors all showed interspersed RAD51 negative as well as RAD51 positive tumor cells. As a BRCA defect was found in 2/6 HRi tumors, they biologically resemble HRD tumors. However, whether HRi tumors benefit from PARPi treatment remains to be elucidated.

Mutational signature analyses were performed to explore novel mechanisms related to or underlying the HRD phenotype. One HRD tumor that showed a large contribution of three signatures related to MSI and MMR deficiency proved to be truly MSI. Using pentaplex PCR and IHC, MSI was discovered in two HRD (2/20, 10%) but not in HRP or HRi tumors. This incidence is much higher than in the unselected breast cancer population (1%; refs. 34, 35), suggesting that the RECAP test may also identify MSI tumors. The relation between MSI and HRD as well as the order in which tumors develop these deficiencies remains unclear and future research is required. We hypothesize that either MSI tumors acquire HRD over time due to accumulation of mutations in genes involved in HR (44), or HRD tumors acquire MSI at a later stage of tumor development as a compensatory mechanism, to lower replication fork instability by not repairing mismatches but rather continuing DNA replication. As MSI tumors have many neo-antigens, PD-L1 blockade therapy showed antitumor activity in phase I trials (45). Recently, a first report of a patient with MSI BrC showing a profound response to PD-L1 blockade was published (46). Moreover, tumors with high numbers of TILs are generally more sensitive to immunotherapy (47). Interestingly, in our cohort, the two MSI tumors comprise a different subset of HRD tumors than the ones with high TILs. Thus, the RECAP test identifies not only tumors with BRCA defects (n = 16), but also a subgroup of breast cancers that might respond well to immunotherapy due to either MSI (n = 2) or high TIL counts (n = 6).

The RECAP test is a robust functional HR assay, detecting both BRCA1/2-deficient and BRCA1/2-proficient HRD tumors. Functional assessment of HR in a pseudo-diagnostic setting is achievable and produces robust and interpretable results. Clinical trials evaluating the predictive value of the RECAP test for in vivo response to PARPi have been initiated.

H.J. Dubbink reports receiving commercial research grants from AstraZeneca. W.N.M. Dinjens reports receiving speakers bureau honoraria from Roche, and is a consultant/advisory board member for Amgen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.G. Meijer, N.S. Verkaik, K.A.T. Naipal, A. Jager, D.C. van Gent

Development of methodology: W.N.M. Dinjens, P.M. Nederlof, H.J G. van de Werken, J.W.M. Martens, A. Jager, D.C. van Gent

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.G. Meijer, N.S. Verkaik, A.M. Sieuwerts, K.A.T. Naipal, C.H.M. van Deurzen, M.A. den Bakker, H.-J. Dubbink, T.D. den Toom, W.N.M. Dinjens, E. Lips, P.M. Nederlof, A. Jager

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.G. Meijer, J. van Riet, H.-J. Dubbink, T.D. den Toom, W.N.M. Dinjens, E. Lips, P.M. Nederlof, M. Smid, H.J G. van de Werken, J.W.M. Martens, A. Jager, D.C. van Gent

Writing, review, and/or revision of the manuscript: T.G. Meijer, N.S. Verkaik, A.M. Sieuwerts, K.A.T. Naipal, C.H.M. van Deurzen, H.-J. Dubbink, W.N.M. Dinjens, E. Lips, P.M. Nederlof, M. Smid, H.J G. van de Werken, R. Kanaar, J.W.M. Martens, A. Jager, D.C. van Gent

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Sieuwerts, J. van Riet, H.F.B.M. Sleddens, A. Jager

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

The authors thank many colleagues from the Departments of Molecular Genetics, Medical Oncology, and Pathology at Erasmus MC as well as from the Maasstad Hospital, who contributed to the collection of patient material. We thank Lindsey Oudijk for her assistance with the TILs scoring. We thank Ronald van Marion for expert technical assistance. D.C. van Gent, A. Jager, and R. Kanaar have received funding from the Dutch Cancer Society (Alpe d’Huzes grant number EMCR 2014-7048 and grant number EMCR 2008-4045). This work is part of the Oncode Institute, which is partly financed by the Dutch Cancer Society and was funded by the gravitation program CancerGenomiCs.nl from the Netherlands Organisation for Scientific Research (NWO).

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.