Comparative Genomic Hybridization Profiles in Human BRCA1 and BRCA2 Breast Tumors Highlight Differential Sets of Genomic Aberrations

Introduction

A hallmark of all solid tumors, and therefore also of breast cancer, is genome-wide DNA copy-number instability resulting in large chromosomal gains or losses. It has been suggested that genomic instability results from diminished dsDNA repair due to mutations in DNA repair genes (1–3) . This is assumed to explain the etiology of breast cancer among carriers of mutant alleles for either of the two DNA repair genes BRCA1 and BRCA2 (4–6) , which impose a highly increased risk for hereditary or familial breast cancer. However, it is unclear whether these genes cause tumors through similar or partially overlapping pathways and thus are associated with similar or different diseases. We have previously shown that the profile of BRCA1-related tumors, as assessed by comparative genomic hybridization (CGH), is distinct from sporadic breast tumors. Application of BRCA1 CGH profiling in classification was further validated on prospective samples (ref. 7, and data not shown). In this study we aim to characterize the CGH profile for BRCA2 tumors and determine its similarity to our previously reported BRCA1 and control breast cancer CGH profiles. Comparison of the genomic profiles in BRCA1- and BRCA2-associated breast tumors may suggest whether germline mutations in BRCA1 and BRCA2 are involved in different pathways, and thus at the genomic level may represent distinct diseases. We suppose that the gains and losses that typify BRCA1 or BRCA2 mutants may yield clues about the locations of other genes involved in the development of these breast tumors. Furthermore, we aim at exploiting the potential differences in CGH profiles for the classification of breast tumors in hereditary cancer families with unconfirmed etiology.

Patients and Methods

Tumor genomic DNA isolation, probe labeling, metaphase chromosome preparation, and hybridizations were done as described (7). Tumor samples were either from verified BRCA1 or BRCA2 mutation carriers or from control breast tumors. Because control tumors were unselected for family history, we estimate their probability for containing BRCA1 and BRCA2 mutations to be around 5%, equal to their proportion in total breast cancer incidence. Data analysis and subsequent classification procedures are as described (7). In short, a series of 36 BRCA1 and 30 control tumors [most of them also included in our previous study (7)], and 25 BRCA2 tumors were analyzed by metaphase CGH against one normal female reference genomic DNA. The chromosomal copy-number loss and gain thresholds were set at fluorescence ratios 0.85 and 1.15, respectively. The experimental setup yields digital microscopic image fluorescence intensity data for 1,716 equal-size channels across the genome. With a threshold-driven statistical minimum-loss-of-information algorithm all channels were pooled into 81 features (bands) as described elsewhere (7). Band nomenclature consists of variance accounted for (VAF), the level of variance (throughout this article 0.85), and the chromosome number followed by the nth band of that particular chromosome (e.g.,VAF0.85-12.3 for the third VAF band counting from tip of the q-arm of chromosome 12, which we shall further abbreviate to VAF12.3; see Table 1 ). After hybridization, channel fluorescence ratios were used to designate for each VAF-band whether there was no change, gain (G), loss (L) or both (G and L). Subsequently, mutual information feature ranking and simple Bayesian discrete classifiers were employed in leave-one-out cross-validation procedures to construct class predictors for BRCA1 versus Controls, BRCA1 versus BRCA2, and BRCA2 versus Controls.

Statistical Procedures. The whole genome ploidy data for three classes of tumors (BRCA1, BRCA2, and Controls) across 81 features (VAF bands) was analyzed to identify tumor class-specific VAF bands. Fisher's exact two-sided P values were computed separately for the following class combinations: BRCA1 versus BRCA2, BRCA1 versus Controls, and BRCA2 versus Controls. For each of these pairwise class comparisons, two cases were considered: gains versus non-gains (the rest) and losses versus non-losses (the rest). This results in 486 P values (3 pairwise class comparisons × 81 bands ×2 aberration types). All P values are adjusted for multiple testing by executing 1,000 permutation runs of the Westfall and Young step-down resampling algorithm (8). Through this procedure, the number of (adjusted) significant features was reduced to 14 VAF bands with high likely involvement in tumor type-specific genome stability pathways ( Table 2 ).

Results

CGH results are frequently reported as frequency tables of gains and losses of chromosome arms (41 arms). In a previous study we developed a method to subdivide the genome in smaller regions (81 regions) by clustering adjacent elements based on the correlation of the CGH fluorescence ratios ( (7), Table 1). Here, we report losses and gains in those 81 bands for 36 BRCA1, 25BRCA2, and 30 control tumors ( Fig. 1 ). With an average of 34 aberrations per tumor, the BRCA1 class clearly has the highest degree of instability. BRCA2-related tumors showed on average 28 alterations per tumor, and control tumors 25 altered regions. The most pronounced differences between the classes were observed in the bin representing 11 to 20 alterations per tumor. This bin contains only 3 of 36 (8%) BRCA1-associated tumors, but 24% of BRCA2 and 33% of control tumors, respectively. The bin with 41 to 50 alterations contains 3% of the control tumors, 28% of the BRCA1-related tumors, and 12% of the BRCA2-related tumors ( Fig. 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Frequency of aberrations in all 81 VAF0.85 bands calculated per tumor class. Control tumors (n = 30), BRCA1 (n = 36), BRCA2 (n = 25).

We detected genomic gains and losses in all tumors studied. Overall, in 91 tumors with 81 channels each, gains represented 18%,losses 18%, gain-and-loss within one band 0.3% (cf. hatched in Fig. 2 ), and no aberrations 64% of all genomic regions. The tumors with highest degree of alterations (aneuploidy) contained 55 copy-number aberrations out of 81 bands. The tumor with the lowest number of aberrations showed only two gains (+VAF1.3 and +VAF3.1) and one loss (−VAF2.1). The top-three most frequent gains were +VAF1.3 (82.5% of 91 tumors), +VAF8.3 (73%) and VAF1.4 (65%). The top-three most frequent losses observed were, -VAF8.1 (52.7%), -VAF4.1 (50.5%), and -VAF16.2 (49.5%). Gains were least frequent in VAF3.2 (1%), VAF13.2 (1%), VAF13.3 (2%), and losses were never observed in three bands VAF1.3, VAF3.5 and, VAF3.6.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Individual tumor profiles in 14 significant genome VAF bands. Columns are breast tumors from proven BRCA1, proven BRCA2, or unselected breast cancer patients. Rows represent genomic regions as VAF bands. Gain (grey), loss (black), gain-and-loss (x), or unchanged (white) in 14 significant VAF0.85 bands with two-sided Fisher's exact adjusted P < 0.1.

Different Tumor Profiles of Gains and Losses for BRCA1 and BRCA2 Breast Tumors. In testing our hypothesis of differential CGH profiles in BRCA1-, BRCA2-associated, and control breast tumors, we restricted further analyses to the 14 VAF bands with significant two-sided Fisher's exact adjusted P < 0.1 ( Table 2). Pairwise comparison between BRCA1 and control tumors ( Table 2, column B1vsC) revealed 10 significant VAF bands. Nine of these show differential gains (top panel) and only VAF5.3 showed differential loss (bottom panel). In two of these bands (+8.2 and +16.1) control tumors showed the highest percentage of aberration ( Table 2, cf.B1vsC).

The comparison between BRCA2-associated and control tumors revealed just one significant VAF band, namely gain in VAF3.5 ( Table 2, column B2vsC), which is also significant in the comparison between BRCA1 and controls.

BRCA1-associated tumors differed significantly from BRCA2-associated tumors ( Table 2, column B1vsB2) in four VAF bands, +1.3, +9.4, −12.3, and −14.2. Only in +VAF9.4 the frequency of gain in BRCA2 (8/25 or 32%, see Fig. 2) significantly exceeded that ofBRCA1 (0%) whereas for the three remaining VAF bands (+1.3, −12.3, and −14.2) the highest percentage of aberrations was present in BRCA1 tumors.

Class-Specific Aberrations. Of the 14 significant VAF bands here identified, no band is uniquely associated with a specific tumor class in the sense that it is significant compared with both other classes ( Table 2). Only gain of VAF3.5 in control tumors (13%) is significantly less frequent than in BRCA1 (67%) or BRCA2 (58%) and therefore this should not be defined as a unique property of control tumors but rather a property of BRCA1 and BRCA2 tumors.

Construction of Classifiers. The other aim of this study was to evaluate the possibility to classify BRCA1 and BRCA2 breast tumors. With data from this and an earlier study, we built optimal classifiers for each tumor class according to the previously formalized procedures and concluded that, as reported (7), BRCA1 and control tumors can be classified with 83% performance. Attempts in the present study to build classifiers to distinguish BRCA2 from control tumors or distinguish BRCA1 from BRCA2 were not satisfactory based on their limited performance (B2vsC, 68%) or because too many VAF bands, each with marginal contribution, had to be included in the classifier (B1vsB2 consisted of 64 bands). Our findings suggest that BRCA2-associated tumors are a lot more similar to control tumors than BRCA1-associated tumors. This can also be seen in Table 2, where, in contrast to B1vsC, only VAF3.5 proved significantly different between BRCA2-associated tumors and controls. The subsequent differential features between BRCA1 and BRCA2 ( Table 2) were not sufficient to further distinguish these tumors from BRCA1 (data not shown).

Discussion

Classification of Individual Breast Tumors. We have identified 10 genomic regions with differential gains and losses between the BRCA1-associated and control tumors. The comparison between BRCA2 and controls yielded just one genomic region (VAF3.5) whereas four genomic regions were significantly different between BRCA1 and BRCA2 ( Table 2). These differences, however, were not enough for classification of individual BRCA2 breast tumors (results not shown). We already knew that the differences between BRCA1 and control tumors supported classification of individual BRCA1 tumors with high performance (7), but we learned in this study that the differences between BRCA2 and controls were much smaller and hardly exclusive for BRCA2 tumors because of VAF3.5 gain being also frequent in BRCA1 ( Table 2). We currently aim at collecting genomic profiles at a higher resolution using array CGH which detects aberrations beyond the resolution of metaphase CGH and could identify new significant regions between tumor classes to be employed in classification.

BRCA1 and BRCA2 Functions and Breast Cancer. Thus far, there is a limited number of publications on the genomic alterations in BRCA2 tumors of which only one uses CGH (9). Our results largely agree with those findings, namely significantly more losses in BRCA1-associated tumors compared with controls for 5q, 4q, 4p, 2q, and 12q.

Differences with the results of Tirkkonen et al. (9) include results for 6p, with gain in ∼10% of their control tumors compared with 40%gain in our control tumors. Also, we found gain of 6p in BRCA2 samples (28% in VAF6.1 and 12% in VAF6.2) in contrast to none of Tirkkonen et al. (9). Among control tumors, in the bands corresponding to 13q, we found 60% (VAF13.3) and 43% (VAF13.4) loss compared with only 25% reported by Tirkkonen et al. Finally, for BRCA1 we observed 31% loss in VAF13.3 and 25% loss in VAF13.4 as opposed to >70% by Tirkkonen et al. (9). These differences could be explained by the small sample sizes but it is not inconceivable that different genetic backgrounds between the populations of the Netherlands and Iceland play a role also, considering that 13 out of 15 BRCA2-associated tumors from the Tirkkonen study (9) represent identical mutations and belong to a single haplotype.

Comparison of our data with a CGH study of ovarian tumors using 46 BRCA1, 18 BRCA2, and 28 control tumors (10) illustrates the limitations in our current understanding of how chromosomal instability pathways are determined by tissue type and/or mutations in BRCA1, BRCA2, or other yet unidentified breast cancer predisposing genes. A comparison of frequencies of aberrations between our data and that of Ramus et al. (10) shows similar loss in BRCA1-associated tumors at 3pter-p22 (cf.VAF3.1), loss at 12q21-31 (cf. VAF12.3) compared with control tumors (10). Further research of BRCA1 and BRCA2-related tumors could prove crucial to the development of the much needed classification methodology for these high-risk familiar ovarian tumors.

Our working hypothesis that BRCA1 and BRCA2 tumors develop through a similar tumor development pathway as a result of decreased DNA repair caused by mutation seems not supported by our CGH data due to the tumors not having similar genomic aberrations, but seems in accordance with a report of distinct expression profiles for BRCA1- and BRCA2-associated breast tumors (11). Whether some of the distinct chromosomal gains and losses in BRCA1 and BRCA2 tumors account for the differences identified at the expression level remains to be evaluated in future studies that perform both DNA and messenger RNA analysis for the same tumor samples. A prominent and significant deviation from control tumor CGH profiles that is shared by both BRCA1 and BRCA2 profiles is gain of 3q, a very large region with many genes. In BRCA1 tumors gain of 3q is associated with loss on chromosome 3p containing BAP1, a BRCA1 binding protein and a tumor suppressor gene. Also located in this 3p region is MLH1, which is part of the BRCA1-associated genome surveillance complex, containing BRCA1, MSH2, MSH6, MLH1, ATM, BLM, PMS2, and the RAD50-MRE11-NBS1 protein complex (12). Other BRCA1 interacting proteins involved in DNA repair are BARD1, located at 2q34-35 and is frequently lost in BRCA2 tumors, and RAD51, a BRCA2 interacting protein located at 15q15.1 for which we found loss of 42% in BRCA1 but only 20% in BRCA2 and 20% in control samples. Interestingly, recent gene-expression profiling studies along with immunohistochemical studies indicate the existence of breast cancer subtypes (13–15) that may relate to pathology and immunohistochemistry (15–18) . A phenotype characterized as ER⁻/erbB2⁻/PR⁻/EGFR⁺/CK5/6⁺, which seems to be indicative for BRCA1 status and is referred to as “basal-like,” was initially found in unsupervised clustering of breast cancer gene expression data (13). An emerging model is that BRCA1 tumors derive from a different cell lineage (basal cells) than sporadic tumors (luminal cells) and are possibly different from BRCA2 tumors as well. However, it remains impossible to identify BRCA1 or BRCA2 tumors solely on the basis of histopathology, and development of other classification methodologies remains urgent. Recently, we have successfully employed CGH profiles to classify individual BRCA1-related tumors in a pool of otherwise unselected (control) tumors (7). Our current CGH study shows a closer molecular resemblance between BRCA2 tumors and controls than between BRCA1 and controls. This study identifies VAF3.5 with a significant difference between BRCA2 and control tumors ( Table 2) and four significant differences between BRCA1- and BRCA2-associated tumors. The optimal classifier to distinguish BRCA2-related tumors from control tumors under-performed with only 68% tumors correctly classified (data not shown). One could explain the inability to distinguish BRCA2 from controls by the limited numbers of tumors studied (n = 25), the insufficient resolution of the cCGH technology (1,716 channels), or the similar intrinsic biological properties of these tumors (16). To address these issues, and to possibly build BRCA2 classifiers in the future, we are now increasing the sample size of BRCA2-related tumors and the sampling resolution by using 3.5k Human BAC arrays similar to the recently reported mouse mammary tumor CGH from murine BRCA1 knockout lines (19). One might also argue that sporadic tumors are similar to BRCA2 tumors. Evidence for this is provided by a recent study in which a newly identified gene, EMSY, was found to bind and attenuate BRCA2 function. EMSY seemed to be exclusively amplified in about 13% of sporadic breast tumors (20), which might result in a BRCA2 phenocopy with a “BRCA2-like” molecular phenotype. Future research in our laboratory focuses on profiling non-B1/non-B2 hereditary breast tumors with array CGH, which is likely to be more sensitive and reproducible compared with metaphase CGH, and may help in the classification of familial breast cancers to enable the appropriate risk-reducing strategies for consanguineous individuals in high-risk breast and/or ovarian cancer families.