Main

More than a decade has passed since BRCA1 and BRCA2 were cloned1, 2 and their association with familial breast and ovarian cancer (FBOC) was established.3 However, recent data indicate that these two genes explain only 25% of these families.4, 5, 6

The large number of families without an identified causative gene mutation has led many a groups to pursue putative BRCAX gene(s) through different approaches, but without success. Several reports have been published since 1995 suggesting linkage of the BRCAX gene to specific chromosomal regions,7, 8 but these data have not been reproduced in larger series.9, 10, 11 Hedenfalk et al12 carried out a study using expression arrays in a small group of non-BRCA1/2 tumors, and concluded that they were heterogeneous and could be split into two main groups; but again, these results have not been reproduced. It has also been suggested that BRCAX families could be explained by a polygenic model, or that they might carry mutations in another gene or genes conferring a moderately increased risk of breast cancer.13, 14, 15

Using paraffin-embedded tissue, we and others have previously demonstrated that BRCA1 and BRCA2 tumors can be differentiated because they have a specific immunohistochemical profile.16, 17, 18, 19, 20 Based on these results, we hypothesized that the use of different immunohistochemical markers might help to group the non-BRCA1/2 tumors, and to confirm that they are heterogeneous. The confirmation of this heterogeneity and their classification would be very important for further studies searching for candidate genes.

Materials and methods

Patients

We collected paraffin-embedded tumor tissues from 50 individuals (mean age 47 years) from 50 different high-risk families, who were studied for mutations in the BRCA1 and BRCA2 genes. These individuals were from families with at least three members affected with breast and/or ovarian cancer, at least one of whom was younger than 50 when diagnosed.4 All 50 individuals were screened for mutations, including large deletions, in the BRCA1 and BRCA2 genes, and no mutations were detected.4, 21 The complete coding sequence and exon–intron boundaries of the BRCA1 and BRCA2 genes were analyzed by a combination of the following different techniques, depending on the center of origin: SSCP, PTT, CSGE, DGGE, and direct sequencing.21

We compared the profiles of non-BRCA1/2 tumors vs a group of 33 tumors from patients carrying a mutation in the BRCA1 gene, selected using the same criteria and studied with the same methodology as described above. Finally, we included a control group of 50 sporadic tumors that were selected because they were diagnosed at similar ages to the non-BRCA1/2 tumors (mean 49 years). In order to confirm that the non-BRCA1/2 and sporadic cases were genetically different, we estimated BRCA1 and BRCA2 carrier probabilities for both the groups using BRCAPRO.22 Data on the majority of the markers assessed in the present study have been previously reported for all three groups of tumors.18

Tissue Microarray Construction

The morphological subtype and grade was assessed in complete sections of each tumor stained with hematoxylin–eosin (H–E). The non-BRCA1/2 tumors consisted of 44 invasive ductal carcinomas, five in situ ductal carcinomas and one invasive lobular carcinoma.

Representative areas of the different lesions were carefully selected on H–E sections and marked on individual paraffin blocks. Two tissue cores (1 mm in diameter) were obtained from each specimen. In addition, four cores of normal breast tissue and two cores of tonsil were included as controls. The tissue cores were arrayed onto one independent new paraffin block using a tissue microarray (TMA) workstation (Beecher Instruments, Silver Spring, MD, USA). The final TMA consisted of 106 cores, each 1 mm in diameter, spaced 0.8 mm from each other. A section stained with H–E was studied to confirm the presence of morphologically representative areas of the original lesions. The BRCA1 and sporadic tumors were included in two separate TMAs using the same technique.

Immunohistochemical Studies

Immunohistochemical staining was performed by the Envision method (Dako, Glostrup, Denmark), with a heat-induced antigen retrieval step. Sections from the tissue array were immersed in 10 mM boiling sodium citrate at pH 6.5 for 2 min in a pressure cooker. Antibodies, dilutions and suppliers are listed in Table 1.

Table 1 Antibodies used in the present immunohistochemical study and threshold to consider a tumor as positive, used in χ2 analysis
Full size table

Between 150 and 200 cells per core were scored for the percentage of positive nuclei or cytoplasms, depending on the marker. We evaluated nuclear staining for estrogen receptor (ER); progesterone receptor (PR); p53; Ki-67; cyclins D1, D3, E, and A; p16; p27; p21; Skp2; retinoblastoma protein (Rb); E2F6; MDM2; topoisomerase IIα; survivin; and CHEK2, and evaluated cytoplasmic staining for BCL2, vimentin, cytokeratin 5/6 (CK5/6), cytokeratin 8 (CK8), and cyclin B1, as previously described.16, 17, 18 Only the percentage of stained cells was considered (independent of the intensity), and the positivity threshold for each marker is listed in Table 1. We and others have previously used the same threshold in the analysis of these markers.16, 18, 20, 23, 24, 25, 26, 27 HER-2 expression was evaluated according to the four-category (0–3+) DAKO system proposed for the evaluation of the HercepTest, and HER-2 expression of 3+ was the only value considered positive, as previously described.18, 19

Statistical Analysis

Hierarchical unsupervised cluster analysis was performed by means of the UPGMA (unweighted pair-group method using arithmetic averages) method using correlation distance and Euclidean distance between markers. The cluster was displayed using SOTAARRAY28 (software available at http://gepas.bioinfo.cipf.es/). The method was implemented in the GEPAS package.29 Immunohistochemical results were represented by a range of color from green to red, the most green representing the lowest, and the most red the highest percentage of positive cells for each marker. Exceptions were grades which were scaled as 33% ‘expressed’ for grade 1, 66% for grade 2, and 100% for grade 3 and HER-2, which was scaled as for 100% for positive (3+), and 0% for negative (Figure 1). We used the CAAT software based on Silhouette Width for clustering validation (software available at http://gepas.bioinfo.cipf.es/).

Figure 1
figure 1

Hierarchical clustering of 50 non-BRCA1/2 tumors. White squares correspond to data not available. The percentage of positive cells for each immunohistochemical marker is represented as a range of color between the most green (lowest percentage) and the most red (highest percentage). Intermediate colors represent percentages between the lowest and the highest.

Full size image

The χ2-test was performed to determine the differences in the distributions of the expression of each antibody and grade between the groups (Tables 2 and 3). The statistical program SPSS 13.0 for Windows (SPSS Inc., Chicago, IL, USA) was used for this analysis. Correction for multiple testing was made by a permutation method in which group membership was randomly assigned, conserving observed proportions, and the distribution of minimum P-values determined over 10 000 permutations. Differences in median BRCAPRO probabilities were tested using the Mann–Whitney rank-sum test.

Table 2 Comparison between non-BRCA1/2 tumors and a group of sporadic breast carcinomas of the number (and percentage in parenthesis) of positive cases for each immunohistochemical marker, unless otherwise indicated
Full size table
Table 3 Immunohistochemical markers that significantly differentiate non-BRCA1/2 tumors with and without somatic BRCA1 inactivation (promoter hypermethylation and LOH of the BRCA1 allele), and comparisons with BRCA1 tumors
Full size table

BRCA1 Promoter Hypermethylation

DNA methylation patterns in the CpG islands of the promoter of the BRCA1 gene were determined by methylation-specific PCR in primary tumors after bisulfite treatment of DNA.30 Placental DNA treated in vitro with SssI bacterial methylase was used as a positive control, and DNA from normal lymphocytes was used as a negative control for methylated alleles of BRCA1.

BRCA1 Loss of Heterozygosity

Loss of Heterozygosity (LOH) analysis of the BRCA1 gene was performed using the intronic microsatellite markers D17S1322 and D17S855 that localize to introns 19 and 20, respectively, and D17S1327 that localizes in 17q21.31 outside the BRCA1 gene.31 The forward primer for each set was labeled using the fluorescent dye FAM (Applied Biosystems/PE Biosystems, Foster City, CA, USA). Reactions were cycled as follows: 95°C for 5 min, then 35 cycles at 94°C for 60 s, 55°C for 60 s, and 72°C for 90 s, followed by final elongation at 72°C for 5 min.

Allele sizes were determined using an automated capillary sequencer (ABI Prism™ 310; Applied Biosystems, Perkin Elmer, Warrington, UK) and were analyzed using GeneScan 3.1 software (Applied Biosystems, Warrington, UK). LOH was determined when the difference between peaks representing alleles in the tumor and the corresponding normal DNA reactions exceeded 25%.

Results

The morphological and immunohistochemical profiles of non-BRCA1/2 tumors were established by analyzing grade and 25 immunohistochemical markers in 50 such tumors, and comparing them with 50 sporadic tumors. Non-BRCA1/2 tumors were of lower grade (adjusted P=0.04); 54% were grade 1 vs 20% of sporadic tumors (Table 2). Although there was marginal evidence that p53 and p21 expression differed between the two groups, these associations disappeared after correction for multiple testing. Overall, the expression of markers related to proliferation, cell cycle, apoptosis, hormone receptors, and epithelial proteins showed similar patterns after adjustment for multiple testing (Table 2).

Although the phenotype of both groups was similar, they were genetically very different. Using BRCAPRO we found that the median probability of not being carrier of BRCA1 or BRCA2 mutations was 0.994 (range 0.964–0.997) for the sporadic tumors and 0.519 (range 0.062–0.987) for the non-BRCA1/2 tumors (P<0.0001).

Hierarchical Unsupervised Cluster Analysis

We performed a hierarchical unsupervised cluster analysis of the non-BRCA1/2 tumors based on the 25 immunohistochemical markers and grade, and found that the 50 tumor samples were separated into two main groups with 25 tumors each, differentiated primarily by grade and ER status. The high-grade branch (Figure 1, left) included tumors of grade 2 or 3 (brown and red squares, respectively) that were ER negative and had overexpression of proteins that promote cell cycle progression and proliferation. The low-grade branch (Figure 1, right) included grade 1 tumors (green squares) that were ER positive and showed overexpression of proteins related to the inhibition of the cyclin–CDK complexes, or the overexpression of luminal epithelial proteins such as CK8.

Within the high-grade branch we distinguished the following three different subgroups: one characterized by HER-2 overexpression (Figure 1, blue branch) that corresponded to 18% of all cases; a second basal-like subgroup containing 14% of the cases; and a third group that represents 18% of all cases, defined by low expression of the luminal epithelial marker CK8 and overexpression of other proteins associated with cell cycle progression and proliferation. Furthermore, this latter subgroup contained most of the ER-positive tumors in this predominantly ER-negative branch (Figure 1, green branch), and we have named it the ‘normal breast-like’ group because of its similarity to the group described by Sorlie et al32 with this name.

The low-grade branch could be divided into two further subgroups (Figure 1, pink and brown branches). The brown branch in Figure 1 was the largest (18 tumors) and demonstrated higher expression of ER, BCL2, CK8, and proteins that inhibit cell cycle progression. The second subgroup contained the remaining seven (14%) cases (Figure 1, pink branch) and showed low, or loss of, expression of ER, PR, and BCL2, but conserved the expression of CK8. We have named them the luminal A and B groups, respectively, according to Sorlie's classification.32

We performed an unsupervised cluster analysis with the 50 sporadic breast tumors, using the same immunohistochemical markers, and identified the same five subgroups, with the same pattern of immunohistochemical expression, and similar proportions in each subgroup; 16% HER-2 positive; 15% basal like; 11% normal breast like; 42% luminal A; and 16% luminal B. We validated these results using the Silhouette Width technique (data not shown).

Somatic Inactivation of the BRCA1 Gene

We observed promoter hypermethylation of BRCA1 in 21 (42%) of the 50 cases, and these were distributed evenly over the two main groups (Figure 1). There was no evidence of an association between hypermethylation and grade, or any of the immunohistochemical markers studied.

We tested for LOH at the BRCA1 locus in 19 of the 21 cases with promoter hypermethylation of BRCA1 (there was no DNA from lymphocytes available for the other 2 cases with promoter hypermethylation), and observed LOH in seven (37%). All seven tumors with double somatic BRCA1 inactivation were in the high-grade branch and four (57%) of those were in the basal-like group (Figure 1).

Discussion

Immunohistochemical analysis of BRCA1 and BRCA2 tumors has previously shown a good correlation between genotype and phenotype,16, 19, 20 but there are very few studies on non-BRCA1/2 tumors.18, 20, 27, 33 We have previously reported that non-BRCA1/2 tumors tend to be grade 1–2 and ER and PR positive, have a low proliferative index, and express p53 to a similar extent to BRCA2 and sporadic tumors, but less than BRCA1 tumors.18, 19, 20 The present study has confirmed our previous results with new markers related to cell cycle, apoptosis, proliferation, stromal, and epithelial markers. We can say that non-BRCA1/2 tumors are of lower grade than sporadic tumors, but they are quite similar with respect to the immunohistochemical markers studied (Table 2).

Immunohistochemistry Classification of Familial Non-BRCA1/2 Tumors

By using a hierarchical unsupervised cluster analysis with 25 markers and grade, we have established a classification of the non-BRCA1/2 tumors that demonstrates their heterogeneity. Non-BRCA1/2 tumors can be divided into two main groups primarily according to their ER status and grade. The first group is characterized by higher grade; ER negativity; and the expression of proteins related to proliferation and cell cycle progression, and the second group by low grade; ER positivity; and overexpression of some cyclin–CDK complex inhibitors, antiapoptotic, and luminal (CK8) proteins (Figure 1).

The two main groups of our analysis can be further divided into five subgroups that are consistent with the classification system established by Sorlie et al32 in sporadic breast cancer, using a cDNA array study, as follows: (1) HER-2 positive; (2) basal like; (3) normal breast like, (4) luminal A, and (5) luminal B. In addition, we performed an unsupervised cluster analysis of 50 sporadic tumors using the same immunohistochemical markers used in the non-BRCA1/2 group, and obtained similar subgroups. Thus, we conclude that non-BRCA1/2 tumors are heterogeneous and that the immunohistochemical classification is very similar to that found for sporadic tumors using both expression arrays32 and immunohistochemical markers. We obtained consistent results after excluding families with any ovarian or male breast cancers (data not shown).

Somatic Inactivation of the BRCA1 Gene

We found that 42% of the non-BRCA1/2 tumors presented promoter hypermethylation of the BRCA1 gene, a percentage that is higher than in sporadic cases, where it ranges between 11 and 30%.34, 35 However these tumors were evenly distributed across both groups and hypermethylation was not associated with estrogen or progesterone receptor status, or (higher) grade, as has been suggested by some authors.35, 36 Seven (37%) of the non-BRCA1/2 tumors with promoter hypermethylation also had LOH at, and therefore total inactivation of, BRCA1. All seven were all in the high-grade group and the majority showed the basal-like phenotype (Figure 1), the latter being characteristic of BRCA1 tumors.37, 38, 39, 40 Although the sample size was small, we compared this subgroup with the rest of the non-BRCA1/2 tumors and found significant differences for the majority of the markers based on unadjusted P-values (Table 3). Consistent differences were observed when 33 BRCA1 tumors were compared with the latter group using previously published data.18 However, the same comparison between the seven BRCA1/2 tumors with somatic BRCA1 inactivation and the 33 BRCA1 tumors did not reveal significant differences even before correction for multiple testing (Table 3). In addition, six out of the seven tumors with BRCA1 promoter hypermethylation and LOH showed morphologic characteristics of medullary carcinoma, a subtype associated with BRCA1 tumors.41 All these observations suggest that in the majority of cases, a double somatic ‘hit’ in the BRCA1 gene is necessary for non-BRCA1/2 tumors to generate a BRCA1 phenotype; the first hit would be promoter hypermethylation and the second hit an LOH of the wild-type allele, and both would occur early in tumorigenesis in order to be able to mimic the immunohistochemical profile of BRCA1 tumors.

Genetic Implications of the New Familial Non-BRCA1/2 Tumor Classification

The classification by immunohistochemistry of non-BRCA1/2 tumors into five subgroups confirms that non-BRCA1/2 tumors constitute a heterogeneous group of tumors, and it supports the hypothesis that the majority of familial non-BRCA1/2 tumors might be explained by a polygenic model (that is, multiple low-penetrance genes),15 rather than by a single BRCAX gene. That is, our results could represent a practical validation of this hypothesis because we have found that non-BRCA1/2 tumors have the same immunohistochemical profile as that described in sporadic breast tumors, a type of tumor classically associated with a polygenetic model. This concept does not exclude the existence of some genes that could each explain a small number of families,13, 14 as was recently shown for CHEK2.42, 43 In fact, our classification could be very useful in defining more homogenous groups for linkage and other studies designed to identify such genes.

Clinical Implications of the BRCA1-like Group

The same mechanisms of somatic inactivation of the BRCA1 gene that we have described in 14% of our cases was recently observed in a group of sporadic breast tumors,35 and these tumors were also high grade and ER negative. These data confirm the specific characteristics (high grade and ER negative) of this group of cases that exhibit what has generically been named ‘BRCAness’, and that both sporadic44 and familial breast tumors can present as ‘BRCA like’. Some experimental studies using demethylating agents are now being conducted on different tumors with good results,45 and this opens up a new avenue for the treatment of tumors with an allele inactivated by hypermethylation that has to be explored further. On the other hand, the identification of this group of tumors with ‘BRCAness’ also provides a basis for new therapeutic strategies based on the sensitivity to DNA-damaging agents that BRCA1 tumors present.46 DNA repair protein PARP inhibitors and DNA crosslinking agents (cisplatin, mytomycin C, diepoxibutane) seem to affect tumor cells with BRCA1 mutations by inhibiting the DNA repair of single strand breaks and increasing their non-viability, while leaving normal cells or cells with a functional BRCA1 allele unaffected.47 Therefore, BRCA1 tumors occurring as a result of constitutional or somatic mutations could represent a new group for targeted therapeutic strategies.