Ploidy and Large-Scale Genomic Instability Consistently Identify Basal-like Breast Carcinomas with BRCA1/2 Inactivation Free

Basal-like breast carcinomas (BLC) are generally described as high-grade ductal carcinomas with a so-called triple-negative phenotype [absence of estrogen receptor (ER), progesterone receptor (PR), and HER2/ERBB2 overexpression/amplification] and markers expressed by the normal basal/myoepithelial cells of the mammary gland [such as cytokeratins 5/6, 14, 17, and epidermal growth factor receptor (EGFR); for review, see ref. 1]. The BLC entity partially overlaps with the larger triple-negative breast carcinoma (TNBC) disease (1). The breast cancer susceptibility gene BRCA1 has a particular relationship to the basal-like phenotype. Firstly, BLCs represent the majority of breast carcinomas arising in BRCA1 mutation carriers, while representing less than 20% of sporadic breast tumors (2). Secondly, the high level of genomic instability observed in BLCs (3–5) fits BRCA1 involvement in double strand break signaling and repair by homologous recombination (HR; for review, see ref. 6). However, whether so-called BRCAness or HR deficiency is a general feature of BLCs (7) remains controversial, as BRCA1 inactivation (by mutation or methylation of BRCA1 promoter) is evidenced in less than 30% of BLC/TNBCs (8), and a high level of genomic instability and response to treatment exploiting HR deficiency are inconsistently found in these tumors (9–12).

Considering its importance in diagnosis and treatment stratification, many studies have tried to define clinically relevant surrogate markers of BRCAness; for review, see ref. 13. Genomic markers of BRCAness were searched for by comparing array-CGH profiles of BRCA1-mutated versus unselected hereditary or sporadic breast tumors (14–17). Studies comparing BLCs with or without BRCA1 inactivation either found no difference (10, 18, 19) or identified 3q gain as associated with BRCA1 inactivation (11). A classifier based on array-CGH profiles and trained on BRCA1 mutated tumors within unselected group of tumors (20) showed a sensitivity of approximately 80% in TNBCs in 2 independent studies (21, 22).

The goal of this study was to identify genomic markers predicting actual BRCA1 inactivation within the group of BLCs. Analysis of single-nucleotide polymorphism (SNP)-array data using a specific data processing method (genome alteration print, GAP; ref. 23) identified ploidy and large-scale chromosomal breaks to be strongly predictive of the BRCA1 inactivation regardless of the mechanism of inactivation. Moreover, BRCA2 appeared mutated in a significant number of BLCs displaying similar features to BRCA1 inactivated cases, supporting the efficiency of these genomic markers in predicting the actual BRCA1 and BRCA2 status of these tumors.

The experimental series consisted of 80 undifferentiated BLCs from patients treated by first-line surgery at the Institut Curie [some BLCs have been described previously: GEO GSE18799 (23–26)]. All tumors were negative for ER, PR [<1% nuclear staining by immunohistochemistry (IHC)], and ERBB2/HER2 [<2+ by IHC or non-amplified by fluorescent in situ hybridization]; and positive for either KRT5/6/14/17 or EGFR by IHC (24, 25). Tumor DNA and RNA extracted from frozen tissue were obtained from the Institut Curie Biological Resource Center. This series comprised a high proportion of tumors arising in patients carrying a deleterious germline BRCA1 mutation (31 tumors). The validation series comprised 60 samples with available SNP array profile including BLCs from a cohort of young women (19 cases; ref. 27), BLCs (15 cases) including 1 case with BRCA2 germline mutation, all from Institut Curie (34 cases; Affymetrix SNPchip6.0); BRCA1 BLCs from GEO GSE19177 (12 cases; Illumina 370K; ref. 18); BLCs from GEO GSE32530 (5 cases and their xenografts; Affymetrix SNPchip6.0; ref. 28); and BRCA1 BLCs from Institut Bergonié (9 cases; Illumina 660K).

This research was approved by the institutional review boards of the Institut Curie. According to French regulations, patients were informed of the research conducted using the biologic specimens obtained during their treatment and did not express opposition. Germline BRCA1/2 gene analyses were conducted exclusively in patients who attended a visit with a geneticist and a genetic counselor in a family cancer clinic, mostly at the Institut Curie, Paris, France and after informed consent.

The SNP array profiles were available for 17 basal-like cell lines [13 cell lines from American Type Culture Collection (ATCC) and BC227 cell line derived from HBCx-17 tumorgraft; ref. 29], hybridized at Institut Curie; HCC1395, HCC70, and MDA-MB-435 data obtained from the Wellcome Trust Sanger Institute Cancer Genome Project (30). The cell line series included HCC1395, MDA-MB-436, and HCC1937 bearing BRCA1 mutations (31, 32), HCC38 with BRCA1 promoter methylation (33), BC227, and HCC1599 with BRCA2 c.6033_6034delTT/p.S2012QfsX5, and c.4550_4559del10/p.K1517IfsX23 mutations, respectively [the HCC1599 mutation described in catalogue of somatic mutations in cancer (COSMIC; ref. 34) and verified by Sanger sequencing]. Deep sequencing data for the 5 triple negative cell lines and corresponding SNP arrays were retrieved from the Supplementary materials of Stephens and colleagues (31).

Breast cancer cell lines were purchased in May 2006 and May 2008 from ATCC, routinely cultured as recommended by manufacturer and used from passages 3 to 13 before analysis by SNP array. Authentification of the cell lines was conducted from the same passage than one used for experiments by verifying by Sanger sequencing the published TP53 (HCC38, HCC1143, and MDA-MB-468), BRCA1 (HCC1937 and MDA-MB-436), and BRCA2 (BC227 and HCC1599) mutations.

Methylation of the BRCA1 promoter was assessed by methyl-specific PCR after bisulfite conversion using the MethylDetector Kit (Active Motif), as described previously (24), with minor modifications (primer sequences are available upon request).

Prescreen for mutations of the BRCA1/2 gene was conducted using Enhanced Mismatch Mutation Analysis [EMMA, Fluigent (35); EMMALYS software P/N: 5331254102]. For abnormal EMMA profiles, the BRCA1/2 exons concerned were sequenced with dideoxynucleotides (BigDye Terminator V1.1, Seqscape V2.5, Applied Biosystems) according to standard protocols (primer sequences and protocols are available upon request).

Primers were designed to validate chromosomal translocations in tumor DNA from case BLC_B1_T06 (T1 from ref. 36) by translocation specific PCR. Primers and PCR program are available upon request.

Breast cancer cell lines were irradiated with a dose of 10 Grays (Gy) using a Cesium gamma-irradiator IBL137 (1.73 Gy/min for 345 s) and incubated at 37°C for 8 hours. After fixation with 4% paraformaldehyde and permeabilization with 0.1% sodium dodecyl sulfate, cells were incubated with anti-RAD51 (Abcam), and anti-53BP1 (Novus Biologicals) and then with anti-mouse Alexa Fluor 488 and anti-rabbit Alex Fluor 555 (Molecular Probes), respectively. Images were acquired using a Zeiss Axioplan 2 fluorescence microscope with a 100X/1.3 oil immersion objective, in 8-bit format and analyzed with ImageJ software (37). Images were then converted in 8-bit format for the final figures. Space resolution is 0.129 μm × 0.129 μm (xy). Scale bars, 20 μm.

Genomic profiling was conducted using 2 platforms: Illumina (32 cases; Human Hap300-Duo) and Affymetrix (48 cases; SNPChip6.0 array). Hybridization on the Illumina platform was conducted by a service provider (Integragen); raw data files were processed by BeadStudio 3.3 (Illumina, reference model file HumanHap300v2_A); normalization was conducted using the tQN algorithm (38). Hybridization on Affymetrix platform was conducted at Institut Curie; cell files were processed by Genotyping Console 3.0.2 (Affymetrix, reference model file HapMap270, version 29).

Both Illumina and Affymetrix SNP array data were mined using the previously described and validated GAP method (23). R scripts and full details of the application are available at the web site. Recognition of absolute copy number ranged from 0 to 8 copies and all segments exceeding 8-copy level were ascribed 8-copy status. Twenty-two possible segmental genotypes were therefore discriminated: A (1 copy); AA and AB (2 copies); AAA and AAB (3 copies); AAAA, AAAB, and AABB (4 copies); etc.

Chromosome number was estimated by the sum of the copy numbers detected at the pericentric regions (with an error rate less than 2 chromosomes per genome as assessed in 25 cell lines with known karyotype, see Supplementary Table S1).

Number of breakpoints in each genomic profile was estimated based on the resulting interpretable copy number profile and after filtering smaller than 50 SNPs variations.

High-quality SNP arrays were obtained for 65 BLCs (15 of the 80 SNP arrays were discarded because of low hybridization quality, low tumor content, or ambiguous profile interpretation), including 23 tumors from patients carrying deleterious BRCA1 mutations (hereafter called BRCA1 BLCs) and 42 BLCs from patients with no evidence of familial predisposition of breast/ovarian cancer or tested negative for BRCA1/2 mutations (hereafter called sporadic BLCs). Forty-one of the 42 sporadic BLCs were tested for methylation of the BRCA1 promoter and nearly 25% were found positive (11/41, hereafter called meBRCA1 BLCs; ref. 24). No evidence of methylation was observed in the remaining 31 cases (hereafter called non-BRCA1 BLCs). BRCA1 epigenetic status was consistent with BRCA1 expression in all 37 cases tested with available transcriptomic data. BRCA1 and meBRCA1 BLCs comprised the group of tumors with proven BRCA1 inactivation (34 cases) that were further compared with the group of presumably non-BRCA1 BLCs (31 cases). Loss of heterozygosity at the BRCA1 locus was observed in 61 BLCs, including all BRCA1 and meBRCA1 BLCs. As previously described, 95% of the evaluated cases were found mutated for TP53 (41/43 BLCs; the 2 TP53 wild-type cases were BRCA1 BLCs; ref. 24).

To obtain insight into the specific genomic alterations of BLCs, genomic profiling was conducted using SNP arrays, which provide 2 complementary measurements: copy number variation and allelic imbalance. The GAP methodology for mining SNP arrays (23) provided segmental genotype profiles (i.e., absolute copy numbers and allelic contents) for each sample (Supplementary Fig. S1). General genomic characteristics, such as chromosome number, DNA index, number of chromosome breaks, and proportions of genome in each genomic state, were inferred from the segmental genotype profiles.

Inferred chromosome counts per genome showed a bimodal distribution (Fig. 1A) similar to those showed for the genomes in various types of cancers (39). Tumor genomes carrying less than 50 chromosomes with a DNA index close to 1 were considered to have a “near-diploid” genome. On the basis of the hypothesis of duplication of the whole genome during cancer progression explaining the second mode of chromosome distribution (39), tumor genomes carrying more than 50 chromosomes and a DNA index higher than 1.2 were considered to have a ploidy of 4 and are hereafter called “near-tetraploid genomes.” More detailed considerations of the genomic alteration patterns justified the ploidy attribution and revealed 1 outlying case to be considered as near-diploid despite the detection of more than 50 chromosomes in its genome (Supplementary Table S2 and Fig. S2). Finally, 24 and 41 BLCs were classified as near-diploid and near-tetraploid, respectively.

Interestingly, the near-diploid tumors almost consistently carried germline mutations or epigenetic inactivation of BRCA1 (19/24) in contrast to the near-tetraploid tumors, which presented a higher proportion of non-BRCA1 BLCs (26/41; P < 0.002; Fig. 1B). Therefore, in our series of tumors, a near-diploid status had 80% [95% confidence interval (CI), 64%–96%] positive predictive value for BRCA1 inactivation.

The total number of breakpoints detected in a cancer genome characterizes the level of genomic instability (40). Overall comparison of BRCA1 versus non-BRCA1 tumors did not show any significant difference (P = .13). In the subgroup of 41 near-tetraploid BLCs, 15 BRCA1-inactivated tumors displayed an elevated total number of breakpoints (199 ± 72; range: 85–294), whereas 26 non-BRCA1 tumors were more heterogeneous (151 ± 67; range: 43–289) and presented a higher proportion of low values compared with BRCA1 tumors (P < 0.03, Wilcoxon rank test). However, the large overlap in breakpoint numbers precluded their direct application to tumor classification.

Recent advances in deciphering tumor genome complexity revealed various mechanisms of alterations associated with different sizes of genomic fragmentation (31, 41, 42), including specific patterns of BRCA1/2 mutated breast tumors (10). To determine the effective sizes of genomic alterations in BLCs (i.e., the actual distance between 2 adjacent breakpoints) the distribution of segments with the respect to their size was considered. Proportion of segments of a given size, averaged through the 65 BLCs of the experimental cohort, followed a log-linear decay starting from approximately 3 Mb (Fig. 2A, Supplementary Fig. S3). The decreasing rate for the fragments 0–3 Mb displayed a steeper decay, evidencing at least 2 populations of breaks affecting the tumor genome with the prominent cutoff at approximately 3 Mb.

Filtering and smoothing all variations less than 3 Mb resulted in a number of breakpoints more significantly associated with BRCA1 status than total breakpoint numbers (P < 0.006, Wilcoxon rank test in the subgroup of near-tetraploid BLCs, 81.3 ± 20.7 and 54.1 ± 30.1 breakpoints in BRCA1 and non-BRCA1 tumors, respectively). We then defined a state transition of the size S Mb if 2 adjacent chromosomal segments, each not less than S Mb in size, have different copy numbers and/or allelic contents. The number of state transitions in the tumor genomes displayed an approximately log-linear decay as a function of S (S = 3,…,20 Mb; Fig. 2B). When the segment size S spanned 6 to 11 Mb, near-tetraploid BLCs were split into 2 stable subgroups. The subgroup with the high numbers of state transitions was enriched in BRCA1 deficient BLCs (16/24), whereas the subgroup with the low numbers of state transitions did not contain any BRCA1 tumor (0/17) with the most significant difference in mean values observed at S = 10 Mb (Fig. 2B and C).

On the basis of these considerations, a large-scale state transition (LST) was defined as a chromosomal break between adjacent regions of at least 10 Mb; the number of LSTs in the tumor genome was estimated for each chromosome arm independently (not accounting for the centromeric breaks) and after filtering and smoothing of all variations less than 3 Mb (Fig. 2D). For the near-tetraploid tumors clear cutoff in LSTs counts separated 17 non-BRCA1 BLCs (12.3 ± 3.7 LSTs) from a group of tumors containing 15 BRCA1-inactivated and 9 non-BRCA1 BLCs (31.4 ± 5.8 LSTs; Figs. 2C and 3). In the near-diploid group mainly containing BRCA1 tumors (19/24), number of LSTs showed a unimodal distribution (25.3 ± 6.1) with 3 non-BRCA1 BLCs within 1 standard deviation (24, 28, and 32 LSTs) and 2 non-BRCA1 BLCs below 2 standard deviations from the mean (1 and 12 LSTs; Fig. 3). Therefore, in each ploidy subgroup the elevated number of LSTs was strongly associated with a BRCA1 inactivated status in BLCs whereas all tumors with reduced number of LSTs showed no evidence of BRCA1 inactivation.

SNP arrays provide the linear profiles of genetic alterations and LSTs represented a subset of chromosomal breaks, corresponding mainly to copy number alterations (Supplementary Data). To clarify the actual genomic rearrangements behind LSTs, the SNP-array profile from case BLC_B1_T06 was superimposed with its structural rearrangements identified by next generation sequencing (NGS; ref. 36). This tumor classified as near-diploid showed 34 LSTs. Most of LSTs (28 out of 34) were supported by the structural rearrangements detected by NGS representing 19 interchromosomal translocations, 5 tandem duplications, 3 probable inversions, and 1 deletion (Supplementary Table S4 and Fig. S4). Nine alterations were further validated by the Sanger sequencing (36) and 11 additional alterations by translocation-specific PCRs (Supplementary Fig. S5). The majority of LSTs (∼70%) were thus corresponding to interchromosomal translocations. In contrast, the small-scale variation (<2 Mb) was shown to be frequently observed in intrachromosomal rearrangements (31), which is consistent with the segment size of 3 Mb detected here as a threshold for the conventionally small variation and filtered out before calculating LSTs.

The number of LSTs accurately reflected the large-scale patterns of genomic instability in the tumors: all tumors with low LSTs displayed either few chromosomal breaks with a high level of aneuploidy (3 samples) or mostly regional accumulations of breaks (16 samples; see, for example, Supplementary Fig. S7). Average number of LSTs per chromosome arm was 3.1 (3.1 ± 1.0, range 0–6). Thus, LSTs avoid any overestimation of the level of genomic instability because of possible catastrophic events affecting individual chromosomes.

Distribution of LSTs along the genome was associated with increased GC content (0.42 vs. 0.399, P < 10⁻¹⁶) and gene-rich regions (OR, 1.31; 95% CI, 1.25–1.37; P < 10⁻¹⁶). Only 1 of 11 genomic regions with high incidence of LSTs corresponded to a common fragile site (FRA2H; ref. 43; Supplementary Fig. S8).

On the basis of the structural genomic features of BLCs described above, a 2-step decision rule for BRCA1-inactivated tumor selection was constructed. The first step consisted in segregating tumors according to their ploidy (into near-diploid or near-tetraploid); the second step consisted in segregating tumors according to the number of LSTs (into LST^Hi or LST^Low) using ploidy-specific cutoffs: 15 and 20 LSTs per genome in near-diploid and near-tetraploid cases, respectively. LSTs exceeding the ploidy-specific cutoff identified BRCA1 deficient tumors with 100% sensitivity and 60% specificity in the experimental set of BLCs (P < 10⁻³, Fisher test; Supplementary Table S4). However, the 12 “false-positive” LST^Hi cases may actually have presented BRCA1 inactivation (not all patients were tested for the germline mutation). The presence of BRCA1 mutation was investigated in tumor DNA for all cases with available material, including 12 LST^Hi BLCs and 15 LST^Low BLCs. Deleterious BRCA1 mutations were found in 7 LST^Hi BLCs whereas no mutation was found in LST^Low BLC. The somatic origin was shown for 2 of these cases for which germline material and patient consent were available (Supplementary Fig. S9).

The remaining 5 LST^Hi BLCs were tested for the BRCA2 mutation and deleterious mutations were found in 3 cases (Supplementary Data). With these findings, the sensitivity remained unchanged at 100% and the specificity was increased to 90% (including BRCA1 and BRCA2 tumors) in the experimental set of BLCs considered (overall accuracy 97%; P < 10⁻¹², Fisher test; Fig. 3).

A validation series of 55 BLC/TNBC (5 of the 60 samples were discarded because of low SNP array quality) included 15 cases with a BRCA1 germline mutation, 15 cases with BRCA1 promoter methylation, 1 case with a BRCA2 germline mutation, and 24 presumably non-BRCA1 cases. SNP array data were processed using the same workflow and structural genomic features were found globally similar to the experimental set (Fig. 3, Supplementary Table S5). Prediction of the BRCA1 inactivation displayed 100% sensitivity (all 30 BRCA1 and 1 BRCA2 inactivated cases were LST^Hi) and 54% specificity (11 LST^Hi cases were not annotated as BRCA1/2 inactivated, but no further screening was conducted at the tumor DNA level; P < 10⁻⁵, Fisher test).

A series of 17 basal-like cell lines was analyzed, including 4 cases with BRCA1 inactivation and 2 cases with BRCA2 mutations. The results obtained followed the trend observed in primary tumors: firstly, the only near-diploid cell line was the BRCA1-mutated MDA-MB-436 (LST^Hi); secondly, among near-tetraploid cell lines, HCC1395, HCC1937, HCC38, HCC1599, and BC227 carried the highest number of LSTs, which is again consistent with their BRCA1/2-inactivated status (Fig. 4A).

To test functionality of the HR pathway in the basal-like cell lines, formation of RAD51 foci was analyzed after ionizing radiation in 7 cell lines as a read-out of HR pathway proficiency. As expected, no foci were observed in the 4 BRCA1-deficient cell lines. Interestingly, the 2 assessed LST^Low cell lines displayed RAD51 foci accumulation, excluding undetected genetic or epigenetic BRCA1/2 inactivation in these cell lines (Fig. 4B).

To estimate the stability of LSTs over time and replication, we compared SNP array profiles of primary breast tumors and their xenograft models, at early and late passages when available (28). Number of LSTs displayed high intratumor stability, showing the LST number is an intrinsic property of the tumor, although with some variation with passages (Supplementary Table S7 and Fig. S10).

In this study, we addressed prediction of BRCA1-inactivated status based on genomic features in the basal-like subtype of invasive ductal breast carcinomas. Analysis of SNP array genomic profiles (23) of well-characterized sets of BLCs identified 2 markers, tumor ploidy, and number of LSTs, which discriminated BRCA1/2-inactivated cases with high accuracy.

Our analysis of BLCs genomic profiles provided 2 novel findings. First, tumor genomes with near-diploid and near-tetraploid genomic contents also differed in terms of BRCA1 status: more than 80% of near-diploid tumor genomes were associated with BRCA1 inactivation (because of mutation or promoter methylation), whereas non-BRCA1 tumors had mostly near-tetraploid genomes. Second, the number of LSTs (chromosomal breaks between 2 adjacent regions of at least 10 Mb in size calculated after filtering of all variation less than 3 Mb in size) was introduced as a surrogate measure of a large-scale genomic instability. A prominent cutoff in the LST distribution was observed in each ploidy subgroup that distinguished highly rearranged genomes with a high proportion of proven BRCA1/2-inactivated BLCs from discriminatively less rearranged genomes without any proven BRCA1/2-inactivated BLCs. The nature of LSTs was clarified by matching them with rearrangements detected by complete genome sequencing of the same sample (36), and 2/3 of the LSTs were found corresponding to interchromosomal translocations.

A decision rule based on the ploidy-specific LST cutoffs correctly predicted all proven BRCA1-inactivated tumors in 2 independent series of BLCs (100% sensitivity). The discriminative power of LSTs was further reinforced by deleterious BRCA1 and BRCA2 mutations found in 10 of 12 LST^Hi BLCs tested with no previous evidence of BRCA1 inactivation, whereas no mutation was detected in the LST^Low BLCs tested (90% specificity). Two BRCA1 mutations were confirmed to be of somatic origin, further supporting the role of BRCA1 somatic mutations in breast carcinomas (8).

Inactivation of BRCA2 in 6 LST^Hi samples (4 primary tumors and 2 cell lines) suggested that a high level of LST may be a marker of HR pathway deficiency more than a marker of BRCA1 status alone. According to this hypothesis, inactivation of other gene products involved in the HR pathway, such as PALB2/FANCN, BRIP1/FANCJ, or RAD51 paralogs, could potentially explain some of the non-BRCA1/2 LST^Hi cases (6).

The greater instability in BRCA1/2-associated BLCs showed in our study is in line with the observation based on array-CGH (11) showing a lower mean number of breaks in high BRCA1-expressing BLCs compared with low BRCA1-expressing BLCs. It was also observed that BRCA1 and BRCA2 deficient tumors acquired copy number alterations in longer DNA segments (10). However, estimation of genomic instability using SNP arrays presents a number of advantages over array-CGH. First, the 2 complementary profiles provided by SNP arrays (copy number and allelic difference) give a more robust estimation of breakpoint number. Second, SNP array analyses can be used to infer chromosome numbers and tumor ploidy status (23, 44). On the other hand, using large-scale chromosome breaks to characterize genomic instability is largely independent of SNP array resolution, sample quality, stromal cell contamination, and segmentation algorithm.

SNP arrays provide the linear profile of genetic alterations and LSTs represented a subset of chromosomal breaks, largely representing interchromosomal rearrangements. In HR deficiencies, double-strand DNA breaks are thought to be repaired by nonhomologous end joining, a process that may generate translocations (6). This provides a plausible explanation for the increase of LSTs in a BRCA1/2-deficient context. The exhaustive characterization of a large number of BLCs using whole genome sequencing with mate-paired strategies will be mandatory to precisely attribute patterns of large-scale instability to BRCA1/2 status. Until NGS developments make them cost- and time-effective, SNP-array approaches represent a valuable alternative in a clinical setting.

Our findings raise several questions. First, our analysis strongly suggests that BLCs actually include at least 2 different entities, differing in terms of their BRCA1/2 status. Interestingly, this dichotomy also exists in basal-like cellular models, in complete concordance with BRCA1/2 status and HR capacity. Further investigations will be required to determine whether these entities also differ in terms of oncogenic pathways. Second, the fact that BRCA1-deficient tumors are more frequently diploid than non-BRCA1 BLCs is rather intriguing, considering the numerous reports linking BRCA1 to centrosome structure and functions, and experimentally demonstrating aneuploidy and centrosome amplification as a consequence of BRCA1 inactivation (45). It is possible that near-tetraploid BRCA1-inactivated tumors are less frequently viable because of centrosome clustering dysfunction, multipolar division, and cell death in this particular genetic context (46). Alternatively, BRCA1-associated BLCs may have less selective advantages to undergo genome duplication than sporadic cases.

Interestingly enough, genomic instability as measured by number of LSTs was similar in triple negative cell lines with a long history of cell division in vitro to that in primary tumors. In the same line, numbers of LST were quite stable over passages in tumorgrafts, although some genomic changes were readily observed. A model that would account for these observations is that the instability induced by loss of BRCA1 or BRCA2 could be transient and compensated during tumor progression by various mechanisms (47–49).

Finally, our recognition of BRCAness with high sensitivity/specificity could have 2 major impacts in clinical management of breast cancer patients. Use of this genomic signature has an immediately translatable impact in identifying patients for whom a genetic test for BRCA1/2 should be proposed. This may represent a major improvement especially in absence of family history of breast cancer, a situation found in as many as half of BRCA1 mutation carriers (50). This recognition of BRCAness is inexpensive as compared with standard genetic testing, covers all possible mechanisms of BRCA1 inactivation (germinal, somatic, or epigenetic), and results can be obtained in a timeframe compatible with therapeutic decisions. Thus, this signature could be of great interest with the emerging therapeutic perspective of exploiting HR defects by targeting complementary pathways (51). The efficiency of PARP inhibitors on BRCA1/2 mutated breast cancers (52, 53) and the correlation of response to platinum salts with BRCA1 status (12) strongly support the need to more accurately stratify sporadic BLC/TNBC according to actual BRCAness, which could easily be obtained using our genomic-based signature.

A patent has been filed by Institut Curie based on the data presented in this report, with T. Popova, E. Manié, and M.-H. Stern as named inventors. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Popova, M.-H. Stern

Development of methodology: T. Popova

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Popova, E. Manié, G. Rieunier, T. Dubois, O. Delattre, B. Sigal-Zafrani, M.A. Bollet, M. Longy, C. Houdayer, X. Sastre-Garau, A. Vincent-Salomon, D. Stoppa-Lyonnet

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Popova, E. Manié, G. Rieunier, C. Tirapo, A. Vincent-Salomon

Writing, review, and/or revision of the manuscript: T. Popova, E. Manié, G. Rieunier, T. Dubois, B. Sigal-Zafrani, M.A. Bollet, M. Longy, D. Stoppa-Lyonnet, M.-H. Stern

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Popova, E. Manié, V. Caux-Moncoutier, T. Dubois, O. Delattre, C. Houdayer

Study supervision: C. Houdayer, M.-H. Stern

The authors would like to thank Khadija Abidallah for technical support, Odette Mariani for managing tumor samples, Dr. Christophe Letourneau and Pr. Jean-Yves Pierga for careful reading, Dr. Jorge Reis-Filho for full access to NGS data, The Wellcome Trust Sanger Institute Cancer Genome Project for access to SNP-array data, Pr. Alexander Gorban for fruitful discussions, and Anthony Saul for English corrections.

Institut National de la Santé et de la Recherche Médicale, Cancéropole Ile-de-France, Institut Curie and its Translational Research Department. T. Popova is a recipient of a Fellowship from the Institut National Du Cancer (INCa)—GepiG. A. Vincent-Salomon is supported by a grant “Interface INSERM.” This work is part of the “Canceropole Ile-de-France Hereditary Breast Cancer” program coordinated by D. Stoppa-Lyonnet and M.-H. Stern.

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.