The study group was derived from a well-defined population with respect to the local BRCA1 5193G-> A and BRCA2 999del5 germline mutations 2122. All patients within the study group had previously been screened for these BRCA1 and BRCA2 germline mutations. Sporadic tumours were defined as those derived from patients that were negative for the local BRCA germline mutations with no known family history of the disease.

The study group consisted of a selected set of primary infiltrating female breast tumour samples (n = 67). The samples were selected from tumours derived from BRCA1 and BRCA2 germline mutation carriers along with sporadic tumours with and without epigenetic silencing of the BRCA1 gene. At least one sporadic tumour (n = 38) without BRCA abnormalities was selected for each of the BRCA abnormal tumours (n = 29), that is, familial BRCA1 (n = 3) and BRCA2 (n = 18) tumours along with sporadic tumours displaying epigenetic silencing of the BRCA1 gene (n = 8). All tumours, sporadic and familial BRCA1 and BRCA2, were selected by their patient age at diagnosis of 61 years or younger. The DNA samples had previously been isolated from freshly frozen tumour tissue and these samples were obtained from the Biological Specimen Bank of the Icelandic Cancer Society. The tumour samples were macroscopically examined prior to DNA isolation and portions showing viable tumour tissue were identified. These portions were then selected for DNA isolation, which was performed using a standard phenol-chloroform plus proteinase K protocol. Data on clinical parameters were obtained from the Department of Pathology and Department of Oncology, Landspitali Hospital, Reykjavik, Iceland. Time to relapse refers to the time from surgical removal of the primary tumour to diagnosis of recurrence or metastasis. This work was carried out according to permits from the Icelandic Data Protection Commission (2006050307) and Bioethics Committee (VSNb2006050001/03-16). Informed consent was obtained from all patients.

Comparative genomic hybridisation (CGH) was performed using high-resolution oligonuclueotide microarrays (Roche NimbleGen, Inc., Reykjavik, Iceland) 23. The arrays used, "2006_11_01 HG17_WG_CGH" (n = 46) and "080101 HG18_WG_CGH_v2_X1" (n = 21), were of a standard design developed by Roche NimbleGen, Inc. (Reykjavik, Iceland) covering the human genome in about 7 kbp median resolution. Sample preparations and hybridisations were carried out according to manufacturer's protocols (NimbleGen Arrays User's Guide-CGH Analysis, Roche NimbleGen, Inc., Reykjavik, Iceland). Cy3 and Cy5 signal intensity distributions were then normalized using the qspline method (affy package in Bioconductor for R) 24. The array CGH data are available in the ArrayExpress repository (E-TABM-712).

Methylation at the BRCA1 promoter region was assessed in all tumours within the study group by methylation-specific PCR (MSP) as previously described 13. Allelic imbalance by microsatellite analysis at the BRCA1 and BRCA2 loci had previously been performed 13.

Core samples were removed (1.0 mm diameter) from each tumour and rearranged on empty paraffin-blocks using a manual tissue microarray device (BEECHER MTA II; Beecher Instruments, Inc., Sun Prairie, Wisconsin, USA).

Immunohistochemistry (IHC) was applied to 4 μm thick tissue microarray (TMA) sections mounted on superfrosted slides (Menzel, Germany). The slides were dewaxed and immerged in Tris-EDTA, pH 9, (HIER) in microwave oven at 99°C. Endogenous peroxidase activity was inactivated by incubation in blocking solution and the slides then incubated with primary antibody (30 minutes). Polymer conjugate was used as Visualization System (K4061, EnVision+ Dual Link, DAKO, DK-2600, Glostrup, Denmark) (30 minutes) and DAB (K-3468, DAKO, DK-2600, Glostrup, Denmark) used as chromogen (10 minutes).

Expression analysis by IHC on TMA sections was performed for oestrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor (HER)-2, epidermal growth factor receptor (EGFR), cytokeratin (CK) 5/6, CK8, CK18 and BRCA1 [see Additional data file 1]. Expression levels were estimated blindfolded to previously established tumour characteristics and BRCA abnormalities. Expression of ER and PR were scored on a discontinuous scale of 0+, 1+, 2+ and 3+ with the addition of hyper-intense staining (> 3+) being remarked for those tumours displaying extremely intense and concentrated nuclear staining. Positive nuclear ER and PR immunostaining was defined as any visible staining in more than 1% of tumour cell nuclei. Information obtained from Landspitali Hospital, Department of Pathology (ligand binding assay) was used to complement missing data on ER and PR expression thereby allowing ER and PR positivity/negativity to be established for all tumours within the study group. HER-2 positivity was defined as staining of tumour cellular membranes displaying a score of 3+ according to criteria provided by the manufacturer (DAKO, DK-2600, Glostrup, Denmark). EGFR was scored on a discontinuous scale of 0+, 1+, 2+ and 3+ which was estimated by the staining intensity of tumour cellular membranes following descriptions provided by the manufacturer (DAKO, Glostrup, Denmark). EGFR positivity was defined as tumours displaying any, weak or strong, staining of the cellular membrane whereas a score of 2+ or higher was defined as high EGFR expression. Expression of BRCA1 was estimated by nuclear staining where loss of BRCA1 expression was defined as no visible nuclear staining whereas positive expression was defined as any visible, weak or strong, nuclear staining. The CK5/6 marker was scored as positive when weak or strong cytoplasmic and/or membranous staining was visible and otherwise scored as negative. CK8 and CK18 were scored on a scale of 0+, 1+, 2+ and 3+ according to descriptions provided by the manufacturer (DAKO, Glostrup, Denmark).

Luminal phenotype was defined as positivity for either ER or PR whereas non-luminal phenotype was defined as negativity for both ER and PR. The five biomarker classification scheme given in Cheang and colleagues was used to further subdivide these two phenotypic categories into luminal, luminal-HER2, 5NP (five-marker negative phenotype), non-luminal HER2 and basal-like phenotypes 25.

Tumours derived from BRCA1 and BRCA2 germline mutation carriers were defined as BRCA1 and BRCA2 abnormal, respectively. Additionally, tumours displaying epigenetic silencing of the BRCA1 gene were defined as BRCA1 abnormal in those cases where BRCA1 promoter methylation was coupled with complete absence of nuclear BRCA1 protein expression.

Fluorescence in situ hybridisation (FISH) was performed on paraffin-embedded and formalin-fixed tumour tissue sections (4 μm). DNA probe specific for the EMSY gene (BAC human CTD 2501F13, Invitrogen, Carlsbad, CA, USA) was labelled with SpectrumOrange-dUTP (Vysis, Des Plaines, IL, USA) by nick translation, and pRB11 clone for the centromere of chromosome 11 labelled with fluorescein-12-dUTP (Enzo Life Sciences, Farmingdale, NY, USA). Slides were deparaffinised and pretreated, probes and cotI DNA denatured in t-DenHyb-2 hybridisation buffer (Insitus Biotechnologies, Albuquerque, NM, USA) and hybridised to the tumour sections overnight. Stringency wash was performed at 72°C in solution containing 2 × SSC/0.3% NP-40. Analysis was performed in a Leica DMRXA2 fluorescence microscope with at least 100 cell nuclei counted in each experiment. Thresholds for copy gains were set at gene/centromere ratio of more than 1.5 and high-level amplification at a ratio of more than 2.5.

The Cy3/Cy5 ratio signal intensities were log2 transformed following normalisation of each array CGH experiment. The data were then represented by the median of log2 ratios within a window of five probes resulting in a median resolution of about 37.5 kb. Copy number alterations were identified by the Circular Binary Segmentation (CBS) algorithm implemented in DNAcopy (Bioconductor for R) with an alpha of 0.01 to identify change points while cancelling splits having less than 1.0 standard deviation units in difference through the sd.undo procedure 2426. The threshold for determining copy number alterations was fixed at ± 0.08, which was selected to capture the level of plateaus above and below the baseline as observed within the examined tumour genomes 26. To further refine these thresholds we obtained estimates of probe noise levels for each of the arrays as described in Fridlyand and colleagues 27 to subsequently categorise the arrays by their noise levels with the discriminators being the lower and upper quartiles of the distribution. The thresholds for arrays displaying high and low noise levels were then modified to ± 0.10 and ± 0.06, respectively. The assigned thresholds were then validated by examining the association between immunostaining scores of the HER-2 gene (HercepTest, DAKO, DK-2600, Glostrup, Denmark) and copy number states, that is, gains, no change and deletions, which was found to be highly significant (Pearson's correlation, r = 0.53, P = 3.2 × 10^-5). Additionally, we found the frequency of copy number gains at the HER-2 gene locus to be 27% (18 of 67), which is in line with previous reports 20. High-level amplifications were defined as segment means that exceed 2SD units above the mean of log₂ ratios derived from segments gained in copy number. These thresholds were validated by examining the association between HER-2 overexpression (HercepTest, Score 3+) and high-level amplification of the HER-2 gene (Pearson's correlation, r = 0.64, P = 1.3 × 10^-7).

The CBS output was then used to represent each of the tumour genomes as segmented profiles in terms of copy number states, that is, +1 for copy gains, 0 for no change and -1 for deletions. Copy number alterations often cover large genomic regions within which a subset of one or more genes may be targeted 28. Thus, classification of genomic profiles is subject to a large degree of inherent biological noise. This was addressed by projecting the data to lower dimensions through principal component analysis (PCA) prior to cluster analysis making use of only the first few components. This was performed in R 2.7.2 where the prcomp function was used to obtain the components. The first three principal components explained about 40% of the variability in the genomic data [see Additional data file 2]. Each of the other components explained less than 5% of the variability and were ignored in the subsequent steps. Cluster analysis was performed on the lower-dimensional data using the k-means cluster algorithm in R 2.7.2. This was carried out in an iterative procedure where the silhouette information was estimated using the silhouette function in R 2.7.2 (cluster package) to establish strong and reproducible results. Average silhouette information was used to identify the number of clusters in the data with 1000 iterations for each k = {2, 3, ..., 10}. Tumours with a low silhouette information (silhouette < 0.20) were considered borderline instances and were classified by first filtering the data in a Kruskal-Wallis hypothesis test (P < 0.01) while leaving out all such borderline tumours and then determining their membership position by re-applying the cluster procedure. A final model was then derived through the same procedure resulting in all tumours being assigned a cluster membership indicator.

To compensate for the few samples derived from BRCA1 germline mutation carriers in our study we obtained previously published array CGH data available online through ArrayExpress (E-TABM-170). This dataset included genomic profiles derived from five familial BRCA1 tumours, which were combined with our dataset. These five familial BRCA1 tumours were analysed by first identifying copy number alterations as described in Fridlyand and colleagues 27. The output was then used to represent each of the tumour genomes as segmented profiles in terms of copy number states as described above. These segmentation profiles were then combined with our dataset by obtaining copy number states from each of the tumour genomes analysed in this study representing the nearest genomic region to those represented on the CGH arrays used in the Fridlyand and colleagues study 27. This was performed by determining the difference in genomic length for each location between the two array platforms and then selecting the minimal distance. This procedure reduces the median array resolution from about 7 kb to about 765 kb, that is, from the NimbleGen high-resolution design to that used in Fridlyand and colleagues 27.

The degree of genomic instability for each tumour was estimated by determining the fraction of the genome altered. This was computed by obtaining copy number states for each of the windowed probes and determining the number of those assigned as altered in copy number against the total number of windowed probes. This measure, referred to as the genomic instability index (GII), has been described previously 20.

Genomic alterations characterising each of the subgroups were identified by using a conservative modification of the Fisher's exact test with P ≤ 0.001, which was applied on the filtered dataset. This conservative modification of the Fisher's exact test has the advantage of penalising low P-values based on few counts 29.

Genomic alterations present within the study group were visualised by generating a frequency plot displaying the proportion of tumours with copy number gains and deletions at each genomic location analysed (Figure 1a). Examination of the frequency plot reveals that regions frequently gained are infrequently deleted and vice versa. It can also be seen that sites of recurrent high-level amplification events occur within genomic regions that are frequently gained in copy numbers (Figure 1b). These observations show that copy number alterations are not randomly distributed throughout the tumour genomes.

Variability present in the spectrum of genomic alterations within the study group was examined by unsupervised classification of the genomic profiles through cluster analysis. The purpose was to examine the resulting tumour subgroups in terms of their prevalence for BRCA1 and BRCA2 abnormalities.

Cluster analysis revealed four distinct subgroups within the set of tumours constituting the whole study group (Figure 2a; see Additional data files 3 and 4). Three of the identified subgroups displayed high levels of genomic instability as measured by the GII, whereas one subgroup was characterised by low instability levels clearly seen in the distribution of GII within this subgroup in comparison with that of the others combined (Wilcoxon rank sum test, P < 10^-11; Figure 2b). One of the GII-high subgroups (n = 11) was enriched with tumours displaying BRCA1 abnormalities (6 of 11) defined as an instance of a BRCA1 germline mutation or epigenetic silencing of the BRCA1 gene (Fisher's exact test, P = 0.006). This subgroup will hereafter be referred to as the BRCA1-related subgroup. Tumours displaying epigenetic silencing of the BRCA1 gene were also highly enriched within this subgroup when sporadic cases were considered exclusively (Fisher's exact test P = 0.007). Additionally, two other sporadic tumours within this subgroup displayed loss of BRCA1 protein expression without detectable hypermethylation of the BRCA1 gene promoter and both of these tumours were CK5/6 positive. All tumours within this subgroup analysed for loss of heterozygosity at the BRCA1 locus displayed allelic imbalance (Fisher's exact test, P = 0.003). To validate the relationship with BRCA1 abnormalities we obtained previously published CGH array data in which five familial BRCA1 breast tumour samples were analysed 27. The five familial BRCA1 tumours were combined with all the samples in our study group to subsequently re-apply the clustering procedure [see additional data file 5]. All of the five familial BRCA1 tumours clustered among the tumours that constituted the previously defined subgroup of tumours enriched with BRCA1 abnormalities. Combining the two different platforms involves reduction in array resolution for which reasons the five familial BRCA1 tumours were not included in subsequent analysis. A total of five blood-derived DNA samples from individuals with sporadic tumours displaying BRCA1-like genomic patterns were fully screened for germline mutations in the BRCA1 gene and none were found. Another GII-high subgroup (n = 9) was highly enriched of tumours derived from BRCA2 germline mutation carriers (8 of 9; Fisher's exact test P < 10^-4; Figures 2a and 2b). We will hereafter refer to this subgroup as the BRCA2-related subgroup. The third GII-high subgroup (n = 14) was not related to abnormalities in the BRCA genes and will hereafter be referred to as the GII-high-III subgroup (Figures 2a and 2b).

The genomic alterations that characterised the BRCA1-related subgroup, when compared with the rest of the cohort, were deletions at chromosomes 4p, 4q, 5p/q, Xp, Xq along with copy number gains at 10p and 16q (Fisher's exact test modified, P < 0.001) [See Additional data files 6 and 7]. Genomic regions characterising the BRCA2-related subgroup were deletions at chromosomes 1p, 3p, 6q, 8p, 11q, 13q, 14q, 16q, 17p and Xp along with copy number gains at 3p, 8q and 17q as compared with the rest of the cohort (Fisher's exact test modified, P < 0.001) [see Additional data files 6 and 7]. High-level amplifications at 1q43–q44 and 8q24 were prominent (> 20% of samples) within the BRCA2-related subgroup (Fisher's exact test, P < 0.05). The one sporadic tumour that clustered among the BRCA2-related subgroup displayed gains in copy numbers of the EMSY gene located at 11q13.5, which was confirmed by FISH analysis for two different regions of the tumour showing gene/centromere ratios of 1.9 and 3.0, respectively [see Additional data file 8]. Full sequencing of the BRCA2 gene was carried out on blood-derived DNA from this individual and no germline mutations were found.

The genomic alterations that characterised the GII-high-III subgroup were mostly small regions of copy number gains [see Additional data files 6 and 7]. High-level amplifications at 11q13.2–q13.3 were prominent (> 20% of samples) within this subgroup (Fisher's exact test, P < 0.05). All but two samples (12 of 14; 86%) within this subgroup displayed high- or low-level copy number gains at the 11q13.2–q13.3 genomic region (Fisher's exact test modified, P = 0.0003). The high-level amplifications at 11q13.2–q13.3 included two regions at which the level of significance peaks. One of these two regions covered a very small region, about 92 kb, and included a single gene, the FADD gene whereas the second region covered about 556 kb and included four genes, that is, MYEOV, CCND1, ORAOV1 and FGF19.

The subgroup characterised by tumours with low GIIs was not associated with any specific genomic alterations. Some of these tumours displayed copy number gains at 1q, 8q and 16p and deletions at 8p and 16q. These genomic alterations do not all occur within the same tumour but different combinations of them describe the observed variation in genomic profiles found within this subgroup. It is important to note that we did observe tumours that did not display any of these commonly observed alterations with some having very few copy number alterations. We propose here that these tumours may represent biologically important disease entities of breast cancers.

Visual examination of the segmentation profiles revealed clear differences in alteration patterns between each of the identified subgroups, that is, their genome architecture patterns (Figure 3). Tumour genomes within the BRCA1- and BRCA2-related subgroups were characterised by relatively long stretches of genomic alterations, deletions and copy gains along with occasional high-level amplifications (Figures 3a and 3b). Tumours within the simple-profile subgroup appear similar to the BRCA1 and BRCA2 subgroups in terms of altered segment lengths but differ in that they display considerably less complex genomes (Figure 3c). The tumours within the GII-high-III subgroup were characterised by numerous closely packed and small copy number alterations throughout their genomes with occasional high-level amplification events (Figure 3d). This is similar to the complex-firestorm patterns described by Hicks and colleagues or the amplification phenotype described by Fridlyand and colleagues whereas the BRCA1 and BRCA2-related subgroups are more similar to the complex-sawtooth patterns 3031.

The observed phenotypic features of the tumour genomes were quantitatively analysed by examining the segment lengths within each subgroup. This analysis demonstrates that the distribution of segment lengths within the GII-high-III subgroup is shifted towards smaller segments, whereas the tumours within the BRCA1-related subgroup display a shift towards longer segments (Figure 4a). Examining the segment lengths of deletions and gains separately shows that the BRCA1- and BRCA2-related subgroups are characterised by large deletions whereas the GII-high-III subgroup is characterised by small copy number gains rather than deletions (Figures 4b and 4c). Pair-wise comparisons for the distributions in deleted segment lengths between subgroups demonstrates that each of the BRCA1- and BRCA2-related subgroups are significantly different from the simple- and/or GII-high-III subgroups (Wilcoxon rank sum test, P < 10^-11).

The relation between the identified genomic subgroups and tumour phenotypes was examined using a selected panel of biomarkers analysed on TMAs. Tumour phenotypes were established as described in Cheang and colleagues by expression analysis of five biomarkers, ER, PR, HER-2, EGFR and CK5/6 on TMA sections 25. Additionally, we examined the expression of CK8 and CK18 on these TMAs.

Clear trends for particular phenotypic properties were observed for the three genomic instability groups (GII-high; Figures 5a and 5b) [see Additional data files 3 and 9]. The BRCA1-related subgroup was enriched for tumours displaying non-luminal phenotypes (9 of 11, 82%; Fisher's exact test, P = 0.0049) and grade 3+ (8 of 11, 73%). Of the nine non-luminal tumours within this subgroup a total of eight were fully interpretable for all the five biomarkers and were thus further subdivided into basal-like (4 of 8, 50%), non-luminal HER-2 (2 of 8, 25%) and tumours negative for all five biomarkers, 5NP, (2 of 8, 25%; Figure 5a). The two luminal tumours within this subgroup displayed high ER expression (3+) and negativity for HER-2 amplification. The BRCA2-related subgroup was entirely composed of luminal tumours (9 of 9, 100%). All but one of the tumours within this subgroup displayed high expression of ER (≥ 2+; 8 of 9, 89%) and almost all were of grade 3+ (4 of 5, 80%). Four of the seven tumours displaying hyper-intense ER staining (> 3+) were found within this subgroup (4 of 7, 57%; Fisher's exact test P = 0.0049). All of the nine tumours within this subgroup were HER-2 negative (9 of 9, 100%; Figure 5a). The GII-high-III subgroup was almost entirely composed of luminal tumours (12 of 14, 86%). This group of tumours displayed an unusually high frequency of high PR expression with IHC score 3+ (7 of 14, 50%; Fisher's exact test P = 0.034) and with IHC score ≥ 2+ (9 of 14, 64%; Fisher's exact test P = 0.0072; Figure 5b). Supporting this observation is the finding that two of the three tumours within the entire study group displaying hyper-intense staining of the PR gene (> 3+) were found within the GII-high-III subgroup. It can be hypothesised here that the third component shown in Figure 5b reflects differences in luminal vs. non-luminal phenotypes whereas the second component splits up two populations of luminal tumours that are different in terms of PR expression (Figure 5b) [see Additional data file 3].

The simple-profiles subgroup was found to represent a heterogeneous group of tumours in terms of their phenotypes. An important observation is that the luminal tumours within this subgroup displayed a trend towards lower tumour grade as compared with the luminal tumours within the more complex GII-high subgroups (Chi-squared test for trend, P = 0.032). However, non-luminal tumours within the simple-profile subgroup displayed a trend towards high tumour grade as compared with the luminal tumours within the same subgroup (Chi-squared test for trend, P = 0.0015).

The observed heterogeneity within the simple-profile subgroup was further examined by hierarchical cluster analysis on the genomic profiles found within this subgroup. This analysis revealed a cluster of tumours (n = 11) characterised by very low genomic instability indices (mean of 0.029 ± 0.026) as compared with the other simple-profile tumours (Wilcoxon rank sum test, P < 10^-5) and the rest of the cohort (Wilcoxon rank sum test, P < 10^-6). This cluster of tumours displaying mostly 'flat' genomes included high frequency of non-luminal phenotypes (9 of 11, 82%) of which most were basal-like (7 of 9, 78%) and grade 3+ (Figure 6). Although they are characterised by silent or flat genomes, these tumours occasionally show small spikes of alterations, including focal high-level amplifications and very small deletions. We observed that this cluster of silent-tumours (n = 11), referred to hereafter using the 'silent' prefix, more frequently displayed high expression of EGFR (≥ 2+) gene products as compared with the rest of the cohort (Fisher's exact test, P = 0.0069). Two of these five tumours with high EGFR expression displayed high-level amplification of the EGFR gene, which was not found in any other tumour within the study group [see Additional data file 10]. Importantly, the non-luminal tumours with silent genomes displayed an aggressive phenotype in terms of disease outcome (Figure 6). By contrast, the luminal tumours within the simple-profile subgroup displayed low tumour grade and few of these patients experienced relapse within 5 years suggesting non-aggressive disease (Figure 6).

Tumours derived from BRCA1 germline mutation carriers were not found within the cluster of 11 tumours displaying silent genomes. However, two familial BRCA2 tumours were found within this cluster, neither of which displayed deletion or allelic imbalance at the BRCA2 locus and both had an extremely low GII, that is less than 0.034, and were of basal-like phenotype.

Classification of genomic profiles through cluster analysis revealed four distinct subgroups of which two displayed high prevalence of tumours having either BRCA1- or BRCA2 abnormalities. These two subgroups, referred to as the BRCA1- and BRCA2-related subgroups respectively, displayed distinct patterns of genomic alterations and high instability indices. Importantly, our results show that sporadic tumours with epigenetic silencing of the BRCA1 gene develop similar patterns of genomic alterations as tumours derived from BRCA1 germline mutation carriers. This suggests that inactivation of the BRCA1 gene through epigenetic silencing is an important event in sporadic breast tumour development. We found two tumours within this genomic subgroup displaying loss of BRCA1 expression without promoter methylation of the BRCA1 gene and both of these tumours expressed the basal marker CK5/6. Given the observations described above, it is reasonable to speculate that sporadic tumours displaying BRCA1-like genomic alterations are promoted by defects linked with the BRCA1 gene function in genomic maintenance.

Interestingly, one sporadic tumour classified among the BRCA2-related subgroup, which was otherwise exclusively comprised of tumours derived from BRCA2 germline mutation carriers. This tumour displayed a deletion at the BRCA2 gene locus and gains in EMSY gene copy numbers. EMSY gene products are known to interact with and negatively regulate BRCA2 proteins and this may link the BRCA2 gene with sporadic breast tumour development. Obviously, further research is needed to examine the relation between EMSY amplification and BRCA2-like patterns of genomic alterations.

Although BRCA1- and BRCA2-related tumours develop through alterations affecting different regions in their genomes they showed similarities in their genomic architecture patterns with large segments of deletions being prominent. This suggests a similar mechanism by which these tumours acquire instability and we propose here that this might relate to the involvement of the BRCA1 and BRCA2 genes in error-free DNA repair of double-strand breaks through HR (homologous recombination). Inactivation of either BRCA1 or BRCA2 is generally thought to lead to the repair of double-strand breaks by error-prone mechanisms via non-homologous end joining 3233. DNA repair of double-strand breaks by non-homologous end joining can lead to errors leading to gains or losses of large segments of genomic material. This mechanism could underlie the characteristic type of genomic instability observed within the BRCA1- and BRCA2-related subgroups. Alternative but not mutually exclusive interpretations relate to the proposed roles of the BRCA genes in telomere maintenance and centrosome division 43435. By contrast, the third instability subgroup, GII-high-III, was found to display increased propensity to acquire small copy number gains which might relate to the previously proposed 'amplifier phenotypes' and possibly 'complex-firestorm patterns' in breast tumours 2731. The 7 kbp high-resolution array CGH analysis used in this study is crucial for distinguishing between tumour genomes characterised by small copy gains from those displaying large-scale instability patterns. This provided increased clarity in classification of breast tumours by their genomic profiles.

The genomic regions on chromosomes 4, 5 and 10 reported here to characterise the BRCA1-related subgroup overlap with those previously reported to distinguish tumours derived from BRCA1 germline mutation carriers 3637. Because familial BRCA1 tumours resemble basal-like tumours in terms of their phenotype it is of interest to note here that the genomic alterations that characterise the BRCA1-related subgroup overlap with those associated with basal-like tumours 19.

The genomic alterations that were found to characterise BRCA2-related tumour development overlap with those previously described in relation with familial BRCA2 tumours 3638. In addition, we found high-level amplifications at 1q43–q44 and deletions at chromosome 14q, which have not been described before in relation with familial BRCA2 tumour development. The detailed information on the locations of genomic alterations provided by the high-resolution CGH arrays used here allowed us to more clearly delineate the distinct genetic pathways undertaken by breast tumours displaying either BRCA1 or BRCA2 abnormalities.

Further research will be needed to examine the potential of the data presented here to predict BRCA1 or BRCA2 abnormalities in an independent population of breast tumours or cell lines. The importance of establishing a simple and effective classification scheme to identify such tumours lies in the potential benefit of targeted therapy (PARP inhibitors, platinum drugs) for a much larger group of patients than the relatively few BRCA1 and BRCA2 germline mutation carriers.

Tumours derived from BRCA1 germline mutation carriers have previously been shown to predominantly display basal-like phenotypes 1739. In line with this we observed that tumours within the BRCA1-related subgroup primarily display non-luminal phenotypes of which basal-like phenotypes were the most prominent. However, we did identify a cluster of tumours characterised by low genomic instability indices and non-luminal phenotypes. These tumours were mostly of basal-like phenotypes and displayed an aggressive phenotype in terms of disease outcome. This observation demonstrates that a subset of non-luminal breast tumours do not develop towards large-scale genomic alterations supporting the hypothesis that these tumours represent biologically important disease entities 20. Tumours derived from BRCA2 germline mutation carriers have previously been shown to primarily display luminal phenotypes and rarely overexpress HER-2 gene products and these findings were confirmed here 40. Interestingly, we observed two familial-BRCA2 tumours without deletion or allelic imbalance at the BRCA2 locus and these tumours did not display large-scale genomic instability. This raises the possibility that the natural history of some familial BRCA2 tumours does not involve loss of the wild-type BRCA2 allele or at least only partial loss as has been suggested before 41. In this relation, it has been shown that cells heterozygous for a BRCA2 mutation are associated with a phenotype 4243. Taken together, this suggests that a small subset of BRCA2 tumours could be promoted by haploinsufficiency for the BRCA2 gene.

The combined analysis of genomic alterations and tumour phenotypes, presented here, show that BRCA1- and BRCA2-related tumours develop largely through different genetic pathways in terms of the regions altered, while also displaying distinct phenotypes. In light of the common roles for BRCA1 and BRCA2 in genomic maintenance, this suggests that the observed phenotypic differences impose selective advantages for genomic alterations at distinct regions in the context of instability generated by BRCA-deficiency. This is in agreement with the results described by Melchor and colleagues showing the importance of ER status in familial and sporadic breast tumours 44. However, we found two genomically distinct populations of luminal tumours that clearly differed in terms of PR expression. This finding is novel and demonstrates the importance of this factor in breast tumour development.