Characterization of Familial Non-BRCA1/2 Breast Tumors by Loss of Heterozygosity and Immunophenotyping

A positive family history remains one of the most important risk factors for breast cancer, with first-degree relatives of patients having an approximately 2-fold elevated risk. About 15% of all patients have a first-degree relative with breast cancer, and although germ line mutations in BRCA1 and BRCA2 account for a substantial proportion of these cases (1), these mutations explain only 20% to 25% of the overall excess familial risk (2, 3). Mutations in other genes such as TP53 and PTEN are involved in rare multi-cancer syndromes and contribute very little to this risk. Mutations in BRCA1 and BRCA2 are strongly associated with families with at least four cases of breast cancer diagnosed before the age of 60 and one or more cases of ovarian cancer or male breast cancer (1). However, in families with four or five cases of breast cancer, and no ovarian or male breast cancer cases, BRCA1 and BRCA2 mutations were significantly less frequent. Because such a familial clustering is unlikely to have occurred by chance, this has been taken as evidence that other breast cancer susceptibility genes must exist (4).

After the identification of BRCA1 and BRCA2, several chromosomal regions have been implicated by linkage analysis to harbor a breast cancer susceptibility gene. In particular, linkage has been found with markers for 8p12-22 and 13q21 (5, 6), but although mutations in BRCA1 and BRCA2 were excluded, these studies comprised either small or heterogeneous groups of families. Accordingly, these linkage results have proven difficult to replicate by others in independently collected sets of families (7, 8). It has been argued that the inability to detect genetic linkage is largely due to a heterogeneous genetic basis for the families under study (4).

It is now well established that breast tumors arising in women carrying a BRCA1 mutation have distinct histopathologic, immunophenotypic, and genetic features (9–14). This is also true for breast tumors from BRCA2 carriers, although to a lesser extent. These findings indicate that it might be possible to subgroup the breast tumors derived from patients from families in which BRCA1 and BRCA2 have been excluded with high certainty (from now on called BRCAx families). This could possibly decrease the genetic heterogeneity within this group of families, and thereby increase the statistical power to detect linkage. Here, we used loss of heterozygosity (LOH) and immunohistochemical analyses to characterize 100 breast tumors derived from BRCAx families. The BRCAx-related tumors were characterized by more frequent LOH at 22q relative to sporadic breast cancer, and differed significantly from BRCA1- and BRCA2-related tumors in their positivity for Bcl2. However, cluster analyses of the combined data (LOH and immunohistochemistry) did not result in subgroups that would allow useful subclassification of the families.

Family selection. The families were ascertained through the Clinical Genetic Centers in Leiden, Rotterdam, and Nijmegen, as well as through the Netherlands Foundation for the Detection of Hereditary Tumors (STOET). Families were eligible if there were at least three cases of breast cancer diagnosed before the age of 60 from whom genotypes could be determined (n = 216) or inferred (n = 20). Families with cases of ovarian cancer or male breast cancer were excluded, and occurrences of other types of cancer were ignored. Pathologic reports or medical reports were retrieved where available. Blood samples and paraffin-embedded tumor tissues were collected after obtaining written informed consent. The institutional ethical committees of all of the hospitals involved approved this study.

In total, we collected 100 breast tumors derived from 92 patients from 42 selected BRCAx families. Nine of these 100 breast tumors belong to eight CHEK2*1100delC mutation carriers (15). Although the families under study were not tested for mutations in other breast cancer susceptibility genes (such as p53, E-cadherin, and PTEN), they did not show the phenotypic characteristic belonging to these cancer syndromes. We also collected 40 paraffin-embedded tumor samples from sporadic breast cancer cases unselected for family history or age, and from 31 BRCA1 mutation carriers, and 21 BRCA2 mutation carriers.

BRCA1 and BRCA2 mutation testing. In each family, the youngest breast cancer patient from whom a blood sample was available was tested for mutations in the BRCA1 and BRCA2 genes (and for many families the next youngest as well). The joint Clinical Genetic Centers applied a variety of methodologies. The largest central exons (exon 11 in BRCA1 and BRCA2, exon 10 of BRCA2) were scanned by protein truncation tests (16, 17). The small exons were scanned for mutations by denaturing gradient gel electrophoreses or direct sequencing. All of the laboratories specifically assayed the presence of large founder deletions in BRCA1 by deletion junction-PCR (18). The entire coding sequences of BRCA1 and BRCA2 were investigated by conformation-sensitive gel electrophoresis in families that were incompletely scanned at the time of ascertainment (19). Since 2002, each center has offered full sequence analysis and denaturing gradient gel electrophoreses covering the entire coding regions of both genes, and multiple ligation–dependent probe amplification to detect large deletions/duplications in BRCA1 (20).

Histology. Paraffin-embedded tumor tissues were obtained and the breast tumors were histologically classified according to the WHO criteria (21). An expert pathologist (H. Morreau) assessed the type of invasive cancer, histologic grade, the presence of in situ component, and the presence of lymphocyte infiltrate. Age of the patient at time of diagnosis was available from pathologic and medical reports.

LOH analysis. On the respective H&E stained sections, the areas of highest tumor density were selected. Four to six tissue cores (0.6 mm in diameter, Beecher Instruments, Silver Spring, MD) were punched from the designated area using a biopsy needle. DNA was isolated from these punches as described previously (17). These punches generally contain >50% tumor cells. Normal DNA was isolated from the blood samples. For the LOH analysis, we used 65 fluorescence-labeled microsatellite markers selected from Weber Screening Set 6 and covering all chromosome arms (22). Selection criteria were allele product sizes <250 bp (because PCR success rates with DNA isolated from paraffin-embedded material drops sharply with larger amplimers) and position in the telomeric half of a chromosome arm (because this will also detect mitotic recombination events; refs. 23, 24). The PCR products were visualized on an ABI PRISM 3700 DNA Analyzer (Applied Biosystems, Foster City, CA) and analyzed with the Genotyper software version 3.7 NT (Applied Biosystems). The sporadic breast tumors were analyzed only for the six different markers on chromosome 22 and marker D11S15901 on chromosome 11, and the BRCA1-related breast tumors were analyzed only for D4S1562 and D5S1471. Allelic imbalance was defined as the ratio of allele intensities in the normal versus the tumor DNA. An allelic imbalance factor (AIF) of ≥1.70 was scored as “LOH” (25). A technical limitation in the interpretation of the allelic imbalance factor is the possible contamination of tumor DNA with nonmalignant DNA. Although a biopsy needle to punch tissue cores does not prevent contamination with nonmalignant cells, in 80% of the tumor DNA samples, we detected at least one AIF > 5.0, which is only achievable when relatively high proportions of tumor cells are present in the sample (26).

Tissue microarray. Breast cancer tissue microarrays were prepared as described previously (27). From each case, three tissue cores were assembled in the tissue microarray. In total, four tissue microarray blocks were constructed. Three blocks with BRCAx samples and one block with tumor samples from BRCA1 and BRCA2 mutation carriers.

Immunohistochemistry scoring. Immunohistochemical staining was done by the labeled Streptoavidin biotin method (DAKO, Glostrup, Denmark) with a heat-induced antigen retrieval step.

One pathologist (H. Morreau) and one researcher (R.A. Oldenburg) evaluated the immunohistochemical staining results. The percentage of stained nuclei, independent of the intensity, was scored for p53, estrogen receptor (ER), progesterone receptor (PR), and cyclin D1. In the same way, the percentage of cells with cytoplasmic staining was scored for Bcl2. Her2/Neu was assessed in accordance with the DAKO HercepTest guidelines with a score of ≤1 considered negative. Cytokeratin 5/6, cytokeratin 7, and cytokeratin 19 were scored according to the presence or absence of membranous expression in the invasive component. The Chek2 staining pattern was scored as described earlier (15). For p53, we used four different categories on the basis of any level of nuclear staining: (a) negative, (b) <25%, (c) 25% to 75%, and (d) >75% positive nuclei. For ER and PR, a case was considered positive when ≥10% of the nuclei stained above background. For cyclin D1, the cutoff limit was 30%. For bcl2, the cutoff limit was 70%.

Statistics. Proportions were compared using χ² statistics. Familial aggregation of LOH status at a marker was tested using a score statistic (28). This statistic tests for the presence of an additive genetic effect. For this analysis, AIFs between 1.3 and 1.7 were regarded as missing. Empirical P values were computed by permutation of the LOH status among relatives of the same family.

Cluster analysis. For the hierarchical cluster analysis, we used the software programs Cluster and TreeView. The data was normalized, mean centered, and average linkage clustering was applied. We renumbered the LOH data of 100 tumors as follows; AIF > 1.70 were scored as “1” (LOH), AIFs between 1.0 and 1.29 (retention of heterozygosity) as “−1”, AIFs between 1.3 and 1.7 as “0”, and homozygotes as missing. The immunohistochemical data for the different markers was scored as “1” when considered positive and “−1” when considered negative.

Linkage analysis. Genotypes were generated for 19 microsatellite markers on chromosome 12, 5 on chromosome 21, and 12 on chromosome 22. The markers were derived from Linkage Mapping Set version 2 (Applied Biosystems), and amplified from peripheral blood lymphocyte genomic DNA by standard PCR methods. DNA from CEPH 1347-02 was typed as reference to ensure consistency of allele sizing. Allele frequencies for parametric linkage analyses were calculated based on one randomly chosen individual from each family. Multipoint linkage analyses were carried out using the program GENEHUNTER version 2.1-B (29). We used a model in which susceptibility to breast cancer is conferred by a dominant allele with a reduced penetrance and a population frequency of 0.003 (30, 31). The risk of breast cancer by age 80 was assumed to be 0.85 in carriers and 0.096 in noncarriers. Risks are modeled in seven age categories (<30, 30-39, 40-49, 50-59, 60-69, 70-79, and 80+) as described (31). We used the multipoint logarithm of the odds (LOD) scores for each family to compute heterogeneity LOD scores, using the standard admixture model, and hence estimated the proportion of families (α) linked to the putative “BRCAx” locus by maximizing the heterogeneity LOD score. Nonparametric linkage analyses were carried out by the program MERLIN version 0.9.12b (32).

Histology. A total of 100 paraffin-embedded breast tumor samples could be retrieved from 92 patients from 42 early onset breast cancer families (BRCAx families, defined as having at least three cases diagnosed before the age of 60, and no cases of ovarian or male breast cancer). We previously found eight patients to carry the CHEK2*1100delC mutation, representing 9 of these 100 breast tumors (15).

The histologic characteristics of this group of breast cancers, as compared with sporadic breast tumors (n = 40) and the breast tumors from BRCA1 (n = 31) and BRCA2 (n = 21) mutation carriers are listed in Table 1. The most common histologic type in all groups was infiltrating ductal carcinoma. Contrary to earlier suggestions (33), lobular carcinoma was not significantly more often found in the BRCAx tumors relative to sporadic cancers. The BRCA1 tumors were of higher grade than BRCAx tumors (P < 0.001) and the BRCA2 tumors (P = 0.01). Most of the BRCAx tumors were of grade 2, and there was a trend towards grade being lower than that of BRCA2 tumors (P = 0.07), which is consistent with previously reported results (9, 11, 33).

Immunohistochemistry. Three tissue microarray blocks were constructed with 98 of the 100 BRCAx tumors, and one with 31 BRCA1- and 21 BRCA2-related breast tumors. All tumors were stained with antibodies against ER, PR, p53, Bcl2, Her2/Neu, Cyclin D1, CHEK2, the basal cytokeratin 5/6, and the luminal cytokeratins 7 and 19, the immunohistochemical markers most commonly studied in BRCA1/2-associated breast carcinomas (Table 2). BRCAx tumors were significantly different from BRCA1 tumors for ER (P < 0.001), PR (P = 0.002), Her2/Neu (P = 0.02), cyclin D1 (P = 0.02), Bcl2 (P < 0.001), and the basal cytokeratin 5/6 (P = 0.0015) staining. There were also significant differences between the BRCA1 and BRCA2 tumors for ER (P = 0.002), Her2/Neu (P = 0.02) and the basal cytokeratin 5/6 (P < 0.001) staining. BRCAx tumors differed significantly from both BRCA1 and BRCA2 tumors only for Bcl2 (P < 0.001), whereas for cytokeratin 5/6, this difference was borderline significant (P = 0.09). As expected, the nine tumors from CHEK2*1100delC carriers were significantly more often negative for CHEK2 staining than BRCA1, BRCA2, and BRCAx tumors. Interestingly, they are also significantly more often negative for luminal cytokeratin 19 staining than BRCAx (P = 0.0008) and BRCA1 (P = 0.006) tumors.

We combined the results of the luminal marker (cytokeratin 19) together with the basal marker (cytokeratin 5/6) expression to subdivide the BRCAx breast tumors into four different cellular phenotypes: “luminal” (only expression of the luminal marker), “basal” (expression of the basal marker and no expression of the luminal marker), “mixed” (expression of the basal marker and expression of the luminal markers) and “null” (no expression of basal and luminal markers; ref. 34). In this subdivision, cytokeratin 7 was not included because of the high percentage of tumors that stained positive in all groups. The results show that a high proportion of BRCAx breast carcinomas express the mixed phenotype or have a pure luminal phenotype (Table 3). The BRCA1 tumors are more often of the mixed phenotype compared with BRCAx tumors (P = 0.0017) and with BRCA2 tumors (P = 0.0007). No significant difference was seen between the BRCA2 and BRCAx tumors. The CHEK2*1100delC related tumors showed a trend towards the null phenotype. Among the BRCAx tumors, the mixed tumors were more often positive for Her2/Neu relative to the luminal group (P = 0.02), and the pure luminal tumors are more often grade 3 than the tumors with a null phenotype (P = 0.006; data not shown).

Genome-wide LOH. The 100 BRCAx tumors were analyzed for LOH with 65 polymorphic markers representing all chromosomal arms. Of the potential 6,500 pair-wise normal/tumor comparisons, 1,698 (26.1%) failed due to PCR problems of either the tumor DNA or normal DNA. Of the remaining 4,802, 1,220 (25.4%) were homozygous (not informative). Thus, in total, 3,582 (55.1%) informative AIFs could be calculated. Using an AIF of ≥1.7 as cutoff for LOH, the mean percentage of LOH among the markers was 30% (±6.3%), which is similar to the overall average LOH rate calculated from 151 published LOH studies of breast cancer (35). LOH frequencies of ≥40% were found at 1q41, 4p16, 11q22, 11q23.3, 16p13, 16q24, 17p12, 21q22, 22q11, and 22q13 (Fig. 1), with the highest frequency found at D22S445 (59%). Whereas many of these chromosomal sites have also been highlighted in the analyses of sporadic breast tumors, we confirmed that the percentage of LOH at D22S445 and D22S315 was significantly higher in BRCAx versus the sporadic breast tumors (respectively, P < 0.02 and P = 0.035; Fig. 2). We also confirmed the high levels of allelic imbalance at 4q (7 of 12 informative cases) and 5q (4 of 9 informative cases) in BRCA1-related tumors (36).

In 28 families, we were able to assess LOH in at least two breast tumors from two patients. We tested whether there were loci at which LOH was found significantly more often within families than expected on the basis of overall LOH frequency at this locus in all our families. This was found for markers D2S125 (P = 0.007), D3S2409 (P = 0.045), D6S1552 (P = 0.03), D12S2070 (P = 0.02), D13S285 (P = 0.02), D21S1255 (P < 0.001), and D22S315(P = 0.01). Of note, marker D22S445 did not show this familial clustering (P = 0.35).

Cluster analyses. We attempted to use the LOH data of 98 tumors in a hierarchical nonsupervised clustering analysis by scoring AIF > 1.70 as “1”, AIFs between 1.00 and 1.29 (retention of heterozygosity) as “−1”, and AIFs between 1.30 and 1.70 and homozygotes as “missing” in the software package “Cluster.” Although the tumors were separated into two groups, these were not readily discernable on the basis of any single marker or combination of markers, nor did the tumors derived from the same family or the CHEK2*1100delC carrier cluster together (data not shown). Adding the immunophenotyping and histologic typing data did not resolve this.

Linkage analysis. We did a linkage analysis in 55 families, complying with our selection criteria, for chromosomes with either a conspicuous LOH score (no. 22, at D22S445) or for which LOH showed significant familial clustering (no. 12 and no. 21). For chromosomes 2, 3, 6, and 13, there were too few families for which linkage and LOH data could be combined to be statistically meaningful. The highest multipoint LOD score at chromosome 21 over all 55 families was −6.37 between markers D21S1256 and D21S1914. At the same locus, the nonparametric LOD (NPL) score was 1.72. Assuming heterogeneity, we found a nonsignificant heterogeneity LOD (HLOD) score of 0.80 (α = 0.25). Selecting the nine families in which the tumor of at least one patient showed LOH at marker D21S1255 decreased both the NPL and HLOD scores (Fig. 3). Similar results were obtained for chromosome 12 (data not shown). In agreement with the absence of linkage, we were unable to detect consistent loss of the same parental allele on either no. 12 or no. 21 in the tumors from these families.

For chromosome 22, the highest multipoint LOD score was −11.34 between markers D22S303 and D22S315, and under the admixture model, the estimated proportion of linked families was 0. When selecting the 12 families in which the tumor of at least one patient showed LOH at marker D22S445, the peak multipoint LOD score under heterogeneity was 0.06 (α = 0.2) between marker D22S303 and D22S315 (27 cM proximal of D22S445).

We have analyzed 100 breast tumors from patients strongly selected for a particular familial background for LOH and immunophenotype analysis. To our knowledge, this is the first study analyzing LOH at all chromosome arms in such an extended and highly selected group of familial tumors. The main purpose of the study was to detect patterns of LOH and/or immunophenotype that would define distinct subgroups of tumors, on the basis of which we would then be able to stratify the families from which they derive. This is one approach to address the genetic heterogeneity problem, which is commonly believed to be the main reason for the inability to detect further moderate- to high-risk breast cancer susceptibility genes (4, 37). For this reason, we have selected cases from families with a high probability of segregating a breast cancer susceptibility gene, but with a minimal residual probability that this is BRCA1 or BRCA2.

In many families, we collected tumor tissues from two or more patients, allowing us to analyze whether certain genetic, immunohistochemical, and morphologic features were more prevalent within families than predicted by chance. We did indeed observe this for LOH with several markers, but not for any of the immunohistochemical markers. However, linkage analysis in the total group of 55 families did not produce significant LOD scores for any of these chromosomes, nor did linkage analysis in subgroups of families selected on the basis of these LOH results. This suggests that LOH analysis of familial cases is unlikely to facilitate the detection of new breast cancer susceptibility loci by linkage analysis. It remains possible, however, that families in which multiple breast tumors show LOH at the same locus are caused by a shared genetic defect on another chromosome. A genome-wide linkage search in our families should address this. For example, it has been reported that breast tumors from families linked to BRCA1 show more frequent LOH on 4q and 5q relative to sporadic breast cancer, which we have confirmed here (36, 38). Hence, it might have been possible to detect linkage to BRCA1 among the families in which several tumors show LOH on 4q or 5q, rather than among families only selected on clinical phenotype. Although our LOH analysis covered all chromosome arms, certain LOH events may have escaped detection because of the limited number of markers we have used. LOH analysis of microarrays with 10,000 single nucleotide polymorphisms could indicate shared LOH regions with more accuracy, as was found for lung cancer (39). LOH analysis with polymorphic markers detects any imbalance in parental chromosomes, including trisomy (26), so that our LOH scorings in fact reflect a wide range of different chromosomal aberrations. To distinguish between these, LOH data should be combined with (array) comparative genomic hybridization. This might be relevant because we do not know at this stage whether other breast cancer susceptibility genes act according to Knudson's two-hit inactivation model (26, 40). It is conceivable, as was found for the MET oncogene in hereditary papillary renal carcinomas (41), that trisomy (or copy number gain) of the mutant allele contributes to susceptibility.

A better resolution for subgroup analysis of the tumors might be achieved by global gene expression analysis. Many different studies describe the possible classification of the heterogeneous group of sporadic breast cancers in distinct subtypes using microarray techniques (42, 43). Five different subtypes (one basal-like, one ERBB2-overexpressing, two luminal-like, and one normal breast tissue–like subgroup) have been recognized (44). These tumor subtypes may represent different biological entities and might originate from different cell types. Four distinct phenotypes (pure luminal, mixed luminal/basal, pure basal, and null) have been defined by immunostaining 1,944 sporadic breast tumors with antibodies for both the luminal and basal phenotypes (34). These subgroups were significantly different in their biological features and clinical course of the disease. In addition, another study (14) showed that the expression patterns from 15 fresh frozen tumor samples from seven non-BRCA1/2 families clustered within their respective families, suggesting an underlying common genetic basis. The recently developed DASL assay technique (45), which makes gene expression analysis possible in archival paraffin-embedded tissues, may extend this observation to larger numbers of cases.

The hypothesis that genetic predisposition to breast cancer might preferentially give rise to certain subtypes is also supported by histopathologic findings in BRCA1-related tumors. These are generally of higher grade, show pushing margin growth patterns, and high lymphocyte infiltration in comparison to sporadic cases (33). They are also more often ER-negative, PR-negative, Bcl2-negative, P53-mutated, and negative for Her2/Neu amplification (our data, and refs. 9, 11, 46). In gene expression profiling, a basal-like gene expression pattern has been associated with BRCA1 carriers (13). We found most BRCA1 tumors (81.5%) to belong to the pure basal or mixed phenotype category, based on cytokeratin 5/6 and cytokeratin 19 expression, as opposed to the BRCA2 tumors which were mostly (75%) of the luminal or null phenotype. Intriguingly, BRCAx tumors were almost equally distributed over both categories. However, we noted that different tumors within the same family frequently belonged to different phenotype categories, indicating that it is unlikely that the basal/luminal phenotype is genetically determined in these cases.

The morphologic and immunohistochemical results from BRCAx breast carcinomas and those arising in BRCA1 and BRCA2 mutation carriers are similar to those recently reported by others (9, 11, 47, 48). Only Bcl2 displayed a significant difference between BRCAx tumors and BRCA1 or BRCA2 tumors (both P < 0.0001), but the proportion of positive BRCAx tumors is not conspicuously different from what is observed in a series of unselected sporadic breast tumors (9). In general, the patterns of immunostaining and LOH in BRCAx tumors closely resemble those of sporadic breast tumors, with the possible exception of the “mixed” phenotype (as defined by cytokeratins 5/6 and cytokeratin 19) and LOH at chromosome 22. Two recent studies (49, 50) have used classical comparative genomic hybridization to analyze a small number of BRCAx-related breast tumors. Both these studies also found chromosomal aneuploidy patterns broadly resembling those of sporadic breast tumors, but did not identify chromosome 22 as a frequent target for aneuploidy. Conversely, regions on chromosome 8 and 19, identified by comparative genomic hybridization (50), were not observed by us. It should be noted, however, that a direct comparison of the BRCAx cases in these studies and ours is difficult due to differences in the applied selection criteria for BRCAx families. For example, the occurrence of ovarian cancer was not used to exclude families in the comparative genomic hybridization studies (49, 50), increasing the probability that some are caused by undetected mutations in BRCA1.

Thus, in our families, a clustering of sporadic, or sporadic-like breast cancer is seen. Yet, it has been argued that such familial clustering is unlikely to occur by chance but instead is more likely to have a genetic basis (1). Therefore, if our families indeed have a genetic basis, our results suggest that this basis is the same as that for sporadic breast cancer. Analyses of genetic models to explain familial breast cancer have indicated that, after correction for BRCA1 and BRCA2, the polygenic model incorporating multiple interacting low penetrance genes is the most likely explanation (51–53). Such genes are also suspected to explain a substantial proportion of sporadic breast cancer. If more detailed analyses of this group of patients by high-resolution array comparative genomic hybridization or gene expression profiling confirms that these tumors strongly resemble sporadic tumors, then this is in agreement with the idea that the remainder of familial risk to breast cancer is caused in a polygenic way. Finding these genes will be a challenge for years to come, but family studies will remain valuable in this regard because one is enriching for genetic susceptibility (54), as was convincingly shown with the identification of the CHEK2*1100delC variant (15, 55).

We thank Annemieke M. van der Wal for assistance with the immunohistochemical staining.