[Frontiers in Bioscience E3, 125-136, January 1, 2011]

BRCA1-related gene signature in breast cancer: the role of ER status and molecular type

Katarzyna Marta Lisowska1, Volha Dudaladava1,2, Michal Jarzab1, Tomasz Huzarski3, Ewa Chmielik4, Ewa Stobiecka4, Jan Lubinski3, Barbara Jarzab5

1Department of Tumor Biology, Maria Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Gliwice Branch, Gliwice, Poland, 2currently: Department of Medical Biology and Genetics, Grodno State Medical University, Grodno, Belarus, 3International Hereditary Cancer Center, Pomeranian Medical University, Szczecin, Poland, 4Department of Pathology Maria Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Gliwice Branch, Gliwice, Poland, 5Department of Nuclear Medicine and Endocrine Oncology, Maria Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Gliwice Branch, Gliwice, Poland

TABLE OF CONTENTS

1. Abstract
2. Introduction
3. Material and methods
3.1. Clinical samples
3.2. BRCA1 mutation-testing
3.3. RNA isolation
3.4. Detection of BRCA1 promoter methylation
3.5. Oligonucleotide micorarrays
3.6. Quantitative RT-PCR
3.7. Methods of data analysis
3.8. Data analysis workflow
4. Results
4.1. Search for the BRCA1-mutation signature in breast cancer
4.2. BRCA1-related expression signature is strongly influenced by ER status of the tumor
4.3. The biological context of the gene expression differences between BRCA1-mutated and sporadic tumors
4.4. BRCA1 expression signature in the context of differences between basal and luminal subtype of breast cancer
4.5. Q-PCR verifies whether changes in gene expression are related to the BRCA1 mutation status or to other factors
4.6. The impact of BRCA1 inactivation (mutation or methylation) on gene expression profile in breast cancer
4.7. Unsupervised analysis confirms that main sources of molecular variability are ER status and molecular subtype
5. Discussion
6. Acknowledgments
7. References

1. ABSTRACT

There is an ongoing debate whether hereditary breast cancer is a clinical entity distinct from sporadic breast cancer. We tried to shed some light on this issue by comparing the molecular profiles of these two types of cancer using DNA microarrays. Our results show that a previously reported marked difference between BRCA1-mutation linked and sporadic breast cancer was probably due to uneven stratification of samples with different ER status and basal-like versus luminal-like subtype. We observed that apparent difference between BRCA1-linked and other types of breast cancer found in univariate analysis was diminished when data were corrected for ER status and molecular subtype in multivariate analyses. In fact, the difference in gene expression pattern of BRCA1-mutated and sporadic cancer is very discrete. These conclusions were supported by the results of Q-PCR validation. We also found that BRCA1 promoter hypermethylation had similar effect on global gene expression as mutation-induced protein truncation. Thus, in the molecular studies of hereditary breast cancer, BRCA1 promoter methylation should be recognized and considered together with gene mutation.

2. INTRODUCTION

Since wide implementation of mutation screening and genetic counseling, breast cancer has been frequently regarded either as a sporadic or a hereditary disease (hereditary breast cancer, HBC). Hereditary cancers may arise due to germline mutation in BRCA1 or BRCA2 genes, and rarely also due to other known mutations (e.g. in TP53, ATM, PTEN or CHEK2), or they may be of unknown etiology (called non-BRCA1/2 or BRCAx cases). There is an ongoing debate whether pathology and clinical behavior of HBC is distinct from those of sporadic breast cancer. Undoubtedly, patients with HBC develop the disease at a younger age, but it is not clear whether they have worse prognosis, as suggested in some studies. When assessing pathology and immunophenotype, BRCA1 mutation-linked breast cancer is regarded as the most distinct category of HBC. Among its characteristics are: high tumor grade (these cases are often characterized by high proliferative activity, resulting in tumors with pushing margins and high mitotic index), elevated lymphocyte infiltration and low estrogen receptor expression (1-4).

The question of putative molecular differences between hereditary BRCA1 or BRCA2 mutation-linked and sporadic tumors was first analyzed by Hedenfalk et al. who used early generation cDNA microarrays (5). The authors claimed that these three categories of breast cancer could be easily distinguished on the basis of distinct gene expression pattern. This issue was also indirectly addressed in two further microarray studies, one concerning the multi-gene signature correlated with clinical outcome (6) and the other describing molecular subtypes of breast cancer (7). Van't Veer et al. observed that tumor samples from patients with BRCA1 mutation fall within ER-negative cluster while Sorlie et al. found that BRCA1-mutated tumors associated with basal-like subtype. In our opinion, the results of these two studies give a clear indication that molecular differences between mutation-induced and sporadic breast cancers are less pronounced than originally proposed in (5). However, this discrepancy was not discussed therein and its causes have not been systematically investigated so far.

The aim of our study was to verify the magnitude of differences in gene expression profile between BRCA1-associated and sporadic breast cancers. In our analyses we took into account the estrogen receptor status and molecular subtype of the tumor, the two most significant features affecting global gene expression pattern in breast cancer. We also checked for BRCA1 promoter methylation in tumor samples and analyzed gene expression profile in tumors with BRCA1 gene inactivated either by mutation or epigenetic silencing..

3. MATERIALS AND METHODS

3.1. Clinical samples

We analyzed 35 breast cancer specimens (Table 1). Surgical samples obtained during mastectomy were flash-frozen in liquid nitrogen and stored at -80�C. Only samples from patients without neoadjuvant chemotherapy were used in this study as chemotherapy may seriously affect gene expression profile. All tissue samples were collected at the Pomeranian Medical University in Szczecin.

Seventeen tumor samples were collected from patients with hereditary breast cancer: 12 were derived from tumors affecting women with hereditary BRCA1 mutation, the only one from a woman with BRCA2 mutation, while another eight cases had familial history of breast/ovarian cancer, but were negative for the BRCA1/2 mutations (so called BRCAx cases). Proportion of BRCA1 and BRCA2 mutated tumors was typical for the Polish population (8-10). Ten samples were derived from patients with apparently sporadic disease (no familial history of cancer) while 4 patients had a history of familial cancer aggregation (FCA) but without prevalence of breast/ovarian cancers. Thus, these samples were merged with sporadic samples in most of the analyses.

All BRCA1 mutation-linked tumors in our study were negative for estrogen receptor (by immunohistochemistry, standard procedures for ER, PGR and HER2 staining were applied), while the only BRCA2-mutated tumor was ER-positive. As we considered hormone receptor status being a very important source of variability in gene expression profile, we carefully matched the control group of sporadic cancers to the group of hereditary breast tumors. Thus, 10 sporadic cases were selected with respect to the estrogen receptor status: 9 tumors were ER-negative, while only one was ER-positive. The larger proportion of ER-positive tumors was observed only in BRCAx group and FCA group (3 and 2 ER-positive samples, respectively). Average age of the patients was 51.8 years. There were 26 ductal and 5 medullary carcinomas within the study group, which is consistent with the distribution of histopathological types in BRCA1 mutation carriers. Patients were diagnosed at stage T1-2, N0-1 and M0.

3.2. BRCA mutation-testing

All patients with familial history of breast/ovarian cancer or familial cancer aggregation (FCA) were diagnosed and tested at the International Hereditary Cancer Center of the Pomeranian Medical University in Szczecin. The most common founder mutations that account for over 90% of all BRCA mutations in the Polish population (5382insC, 300T/G and 4153delA in BRCA1 gene) were checked by multiplex PCR (patent no. P-335917, Poland). Patients who were negative for these three mutations were further tested for 185delAG in BRCA1 and 6174delT in BRCA2 gene by allele specific PCR, according to (11). Tumor samples from patients with strong familial history of breast or breast/ovarian cancers, but who tested mutation-negative, were assigned to the BRCAx group for the purpose of microarray data analysis.

3.3. RNA isolation

Total RNA was isolated according to Chomczynski and Sacchi (12), as it worked best in our hands for fat-rich breast tissue. Frozen tissue (20-40 mg) was homogenized in 600 microliters of ice-cold Solution D with Lysing Matrix D ceramic spheres using a FastPrep instrument (MP Biomedicals, Irvine, CA, USA). After extraction RNA was cleaned up with simultaneous on-column digestion of DNA traces using RNeasy Mini Kit and DNAse I (Qiagen), according to the manufacturer's recommendations. RNA quantity was estimated with ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). RNA quality was assessed using Agilent platform: RNA 6000 Nano LabChip Kit, RNA Integrity Number software and the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA).

3.4. Detection of BRCA1 promoter methylation

DNA was extracted from frozen tissue with Genomic Mini kit (AA Biotechnology). Sodium bisulfite modification of DNA was performed according to (13). Methylation-specific PCR was done with primers specific for methylated sequence and, separately, with primers specific for unmethylated sequence (14). PCR conditions for the unmethylated sequence were as follows: initial denaturation: 10 min at 95�C, then 40 cycles at 95�C (for 30 s), at 60�C (for 55 s) and at 72�C (for 30 s); final elongation at 72�C for 7 min. For methylated sequence: initial denaturation at 95�C for 10 min, then 2 cycles at 95�C (for 30 s), at 64�C (for 55 s) and at 72�C (for 30 s), 2 cycles at 95�C (for 30 s), at 62�C (for 55 s), and at 72�C (for 30 s), 36 cycles at 95�C (for 30 s), at 60�C (for 55 s) and at 72�C (for 30 s), then final elongation at 72�C for 7 min. For each sample PCR was repeated three times.

3.5. Oligonucleotide microarrays

We used HG U133 Plus 2.0 Gene Chip oligonucleotide arrays (Affymetrix) allowing detection of 47 000 human gene transcripts. The hybridization target was prepared according to the recommendations from microarrays' manufacturer. Total RNA (8 micrograms) was used for synthesis of double stranded cDNA. Half of the cDNA volume was used for synthesis of biotinylated cRNA with the BioArray High Yield RNA Transcript Labeling Kit (Enzo Diagnostics). Both cDNA and cRNA were purified with Gene Chip Sample Cleanup Module (Affymetrix). cRNA (16 micrograms) was fragmented and hybridized to the microarray for 16 h at 45�C. After washing and staining the microarrays were scanned with GeneChip Scanner 3000 (Affymetrix). Data were acquired using GCOS 1.2 software (Affymetrix). The preprocessing was performed by Robust Multi-array Analysis (RMA, Bioconductor). Raw pre-processed dataset is available at www.genomika.org/publications/hereditarybreastcancer, together with descriptions of the samples.

3.6. Quantitative RT-PCR

Quantitative RT-PCR analysis was done using the ABI 7700 Sequence Detection System and dedicated software (Applied Biosystems, Foster City, CA, USA). The reactions were performed using MasterAmp Real-Time RT-PCR Kit (Epicentre) according to the manufacturer's recommendations. Primers for the SYBR Green system were designed using Primer3 online software (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi). The sequences of the PCR primer pairs are shown in Table 2. As a reference gene to normalize RNA quantity we used the Eukaryotic Translation Initiation Factor 4 gamma 2 (EIF4G2), that appears to be equally transcribed in all tissues analyzed by microarrays. Gene expression values were obtained by a standard delta-delta Ct method. Primers specificity was verified by sequencing of selected RT-PCR products for each gene.

3.7. Methods of data analysis

Gene expression comparisons by Welch t-test were performed using GeneSpring 7.2 software (Agilent), with non-corrected threshold of p-value less than 0.001. False Discovery Rate (FDR) was estimated by Benjamini-Hochberg algorithm. Two-way analysis of variance (ANOVA), with random variance assumption and global testing were carried out by procedures implemented in BRB Array (developed by Richard Simon and Amy Peng Lam and available on the National Cancer Institute website). Principal Component Analysis of microarray data was carried out by prcomp package of R environment. The reanalysis of the data of Hedenfalk et al. was done on raw microarray data for 3226 clones and 22 tumor samples (among them 7 sporadic and 7 BRCA1-mutated tumors were analyzed), provided by BRB Array Tools repository. Data were analyzed by the same approach and software as our dataset. Class prediction procedure was carried out using Support Vector Machines (SVM) class prediction engine with leave-one-out cross-validation (BRB Array Tools). Real-time PCR gene expression values were compared by non-parametric Kolmogorov-Smirnov test by SPSS 13 software (SPSS), with two-sided p-value threshold of less than 0.05.

3.8. Data analysis workflow

Searching for the BRCA1-mutation signature we performed several supervised analyses (Table 3). The results of each univariate analysis were further verified by the global test. This allowed to estimate the probability of getting that number of genes by chance, despite lack of real differences between the classes. In the multivariate comparisons, the analysis carried out with respect to BRCA1 truncation was corrected for other features that were suspected to influence the molecular profile of breast cancer. Full gene lists obtained in all comparisons are freely available as a supplementary data at www.genomika.org/publications/hereditarybreastcancer Not all gene lists are discussed here, due to the large number of comparisons and low statistical significance of some lists. However, we show the numbers of genes obtained in consecutive analyses (Table 3), as in our opinion, that numbers illustrate which features are truly related to significant changes in molecular profile and which analyses are biased due to samples stratification.

Class prediction was applied in order to estimate whether the BRCA1-mutation linked breast cancers can be distinguished on the basis of their gene expression profile. Finally, we applied Principal Component Analysis (PCA) and searched for the features which correlate with most distinct gene expression profiles.

4. RESULTS

4.1. Search for the BRCA1-mutation signature in breast cancer

First, we compared gene expression profile of BRCA1-mutation linked breast cancer samples to that of all other breast cancer samples without BRCA1 mutation (sporadic, BRCAx, FCA and the only one BRCA2-mutated). Univariate analysis revealed 234 differentially expressed probesets (selected with non-corrected p less than 0.001). This result proved significant also in the global test (p=0.001). However, when BRCA1-mutated samples were compared exclusively to sporadic cases, we obtained only 41 probesets (univariate non-corrected p less than 0.001) and this result turned out non-significant in the global test (p=0.065).

Than we reanalyzed of the data of Hedenfalk et al. (5) using the same approach and software as for our dataset. Raw microarray data for 3226 clones and 22 tumor samples, among them 7 sporadic and 7 BRCA1-mutated tumors were retrieved from BRB Array Tools repository. Comparison of BRCA1-linked tumors vs. sporadic ones revealed only 11 differentially expressed cDNA clones (non-corrected p-value less than 0.001, non-significant in the global test: p=0.065). Thus, it appeared that the difference in gene expression pattern between BRCA1-mutated and sporadic tumors was similarly weak in both datasets.

To further analyze the magnitude of difference between BRCA1-mutated samples and other tumors, class prediction was carried out using 423 probesets with non-corrected p-value p less than 0.001 and Support Vector Machine engine with leave-one-out cross-validation. The maximum achieved accuracy of discrimination of two classes ("BRCA1-mutated tumors" and "other tumors") was only 69%. The same method was applied to the dataset from (5). Maximal possible accuracy (64%) in discrimination of two classes was obtained with 96 probesets (p-value less than 0.001). In both cases the classification accuracy was not much higher than could be achieved by chance, suggesting that there is no significant difference between the tumor classes defined by presence or absence of BRCA1 mutation. There was also almost no overlap between the two gene lists, what may indicate that majority of selected genes were false positives. The only two genes occurring on both lists were TOB1 and ALCAM. One can speculate that these genes may be of functional significance for BRCA1 mutation-linked breast cancer.

4.2. BRCA1-related expression signature is strongly influenced by ER status of the tumor

ER status is well known factor affecting both, clinical course of breast cancer and gene expression pattern in breast tumor samples. Thus, we were aware that a difference in the frequency of ER-positive tumors between groups of "BRCA1-mutated" and "non-mutated" cancers (Table 1) may influence the results of analysis. Indeed, the univariate comparison revealed 589 probesets differentially expressed in ER (+) and ER (-) tumors (non-corrected p-value less than 0.001, global test p-value less than 0.001; data not shown).

To avoid the impact of ER-positive samples on gene selection procedure we excluded from further analysis all ER-positive samples. When we repeated a comparison of "BRCA1-mutated" and "sporadic" tumors in the homogenous ER-negative group of tumors, the difference between both classes was not significant (27 genes at p-value less than 0.001, global test probability p=0.16). A similar conclusion was reached after two-way ANOVA that included both variables: BRCA1 mutation and ER status. In this analysis, ER status was associated with significant changes in gene expression (1380 probesets) while no genes showed altered expression in the context of BRCA1 status at FDR less than 20%. Also at FDR less than 10% none of genes passed these criteria in the comparison of BRCA1-mutated and non-mutated tumors. When the less stringent criteria were applied (non-corrected p-value less than 0.001, nota bene the threshold similar to the that used in (5).), we obtained 375 probesets related to ER status and 101 probesets potentially associated with BRCA1 mutation status.

We found also that the majority of genes selected in our first comparison ("BRCA1-mutated" versus "all other" tumors) were related to ER status. This suggests that in fact we analyzed the difference between ER-negative and ER-positive tumors in this first comparison, what could account for statistical significance of the result. In our opinion, the results described in this chapter suggest that the majority of variance attributed previously to the presence of hereditary BRCA1 mutation was dependent on ER-positive sample imbalance between BRCA1-mutated and remaining tumors.

4.3. The biological context of the gene expression differences between BRCA1-mutated and sporadic tumors

Although the diversity in molecular profile of BRCA1-mutated and non-mutated breast tumors appeared rather discrete and nonsignificant, we tried to analyze its biological background. Thus, we carried out gene set analysis based on two repositories: Biocarta (gene sets related to cellular signaling pathways) and BROAD (different gene sets). Taking into account the results of previous analyses, we used the data exclusively from ER-negative samples. Ten gene sets from Biocarta repository showed significantly changed expression (p-value less than 0.005) between BRCA1-mutated and sporadic tumors, in at least one of four statistic tests used. One gene set (called Msp/Ron receptor signaling), with up-regulation of immunity-related genes TNF, IL1B, CCL2 and CSF1, was found to be significant in all 4 tests. Two further gene sets showed differences in 3 tests; this were Antigen Processing and Presentation, with over-expression of TAP2/1 and HLA-A/DRA genes, and interestingly, BRCA1-dependent Ub-ligase activity, relying on the differences in FANCA/FANCE and BRCA1/BARD genes.

In analysis of curated gene sets from BROAD repository, only two gene sets were significant in all 4 tests (gene set described by Kang in (15) and Msp/Ron pathway). The "Kang gene set" (17 genes in total) was described as downregulated in gastric cancer cell lines resistant to doxorubicin, comparing to parent chemosensitive line. We observed a coordinated change of some genes from this set in BRCA1-mutated samples. Following genes were downregulated: PCYOX1, TMEM106B, ATP2B1, TBL1X, PTP4A2 and, less significant, also NCOA3. Among gene sets changed in 3 of 4 tests used, we found one set directly related to BRCA1 in breast cancer (BRCA1 reporter gene set from (6)). Other gene sets were related to induction of gene expression by TNFA or IFNA. In addition, two chromosome locations showed coordinated gene expression change: 16q24, with up-regulation of at least 25 genes, among them LOC348180, SPG7, FANCA, C16orf7, GALNS, SLC7A5, FBXO31 in BRCA1-mutated samples; and 5q13, with down-regulation of at least 17 genes like LOC653080, MARVELD2, SERF1ACOL4A3BP, TNPO1 and - of great interest - PIK3R1.

4.4. BRCA1 expression signature in the context of differences between basal and luminal subtype of breast cancer

Molecular subtype (mainly basal-like and luminal-like) of breast cancer is also among the features that strongly influence its gene expression profile. Thus, we used microarray data to identify main molecular subtypes in our group of breast tumors, by clustering samples with subtype-related gene set specified by Sorlie et al. (7). It appeared that both subtypes are disproportionately distributed between BRCA1-mutated and non-mutated cancers (Table 1). Thus, we used two-way ANOVA to analyze BRCA1 mutation effect in the context of molecular subtype. Both basal-like and luminal-like subtypes had distinctly different gene expression profile (5705 probesets with changed expression), while no transcripts showed changed expression between tumors with and without BRCA1 mutation at FDR less than 20% and even at FDR less than 10%. When less stringent criteria were applied (non-corrected p-value less than 0.001), we obtained 1222 and only 37 probesets, respectively. The attempts to analyze signaling pathways and functional gene clusters differentially expressed between BRCA1 mutated and non-mutated basal tumors gave no significant results. Thus, it seems that BRCA1-mutated tumors probably do not differ significantly from other basal breast cancers, or the difference is so subtle that much larger numbers of samples are required to achieve informative results.

4.5. Q-PCR verifies whether changes in gene expression are related to the BRCA1 mutation status or to other factors

We have chosen three genes for Q-PCR validation: HIP2, TOB1 and FANCA. All three genes appeared on the list of 423 genes used in class prediction procedure in our dataset. HIP2 gene occurred also among 37 genes obtained in the analysis of BRCA1 mutation status in the context of the molecular subtype. TOB1 gene was one of the only two genes which appeared both in the analyses done on our dataset, and on the dataset from (5).

In the validation step we first used only samples from ER-negative tumors to avoid potential bias related to the differences in gene expression between ER-positive and ER-negative tumors (Table 5). We found out that both TOB1 and HIP2 were differently expressed in BRCA1-mutated tumors compared to other tumors (p=0.04 in both analyses), confirming the results obtained by microarray study. However, when ER-positive samples were also considered, TOB1 showed strong difference between ER-negative and ER-positive tumors (p=0.001), which resulted in high overall significance (p=0.004) between BRCA1-mutated and remaining tumors (non-mutated, both ER-positive and ER-negative). On the contrary, HIP2 showed no differences between ER-positive and ER-negative samples and the overall difference between BRCA1-mutated and remaining tumors was also non-significant in this comparison (p=0.07).

Third gene, FANCA, was significantly changed in the context of BRCA1 mutation status in the microarray data analysis corrected for the influence of ER status. It also appeared in the analyses of molecular pathways done on ER-negative samples (section 4.4.). Q-PCR analysis confirmed that FANCA expression does not depend on estrogen receptor status, however the difference between hereditary and sporadic cancers was also non-significant (expression changes were non-significant in all analyses).

4.6. The impact of BRCA1 inactivation (mutation or methylation) on gene expression profile in breast cancer

In the next step of our analysis we checked by MS-PCR for BRCA1 promoter methylation. In six breast cancer samples that were negative for BRCA mutations, we found BRCA1 promoter methylation. This were: one sample of BRCAx cancer, two FCA samples and 3 samples of sporadic cancer. All tumors with BRCA1 methylation were estrogen-negative and showed basal-like molecular profile (Table 1). There were no significant differences in gene expression pattern between samples with BRCA1 mutation and samples with BRCA1 promoter methylation (43 genes at non-corrected univariate p-value less than 0.001; insignificant in the global test: p=0.24). We thus assumed no biological difference between samples with the BRCA1 gene truncated by either of two molecular events.

Next, we analyzed BRCA1 inactivation in the context of ER-status of the tumor by two-way ANOVA. In this analysis BRCA1 status was a stronger factor determining gene expression pattern than ER status: for BRCA1 inactivation 259 probesets passed the criteria of FDR less than 10% (supplementary Table 4), while only 101 showed such differences in the context of ER. When non-corrected p-values were applied, these numbers were 260 and 185 probesets, respectively (p-value less than 0.001).

When BRCA1 inactivation was analyzed in the context of molecular subtype, 2441 genes passed the FDR threshold of 20% for molecular cancer subtype while none of them reached this limit for BRCA1 inactivation. With non-corrected p-value less than 0.001, 526 genes were found for the subtype, while 57 genes were significantly associated with BRCA1 inactivation (supplementary Table 6).

4.7. Unsupervised analysis confirms that main sources of molecular variability are ER status and molecular subtype

Principal Component Analysis is an unsupervised algorithm that produces graphical representation of samples in which the distance between the samples reflects differences in gene expression profile. Figure 1 shows that according to the first two principal components the tumor samples were split into two clusters. To investigate the nature of these two clusters we analyzed several features defined within the studied group of tumors. It appeared that all BRCA1 mutated tumors as well as BRCA1 methylated samples, located within that cluster which was generally higher in first component and lower in second component (Figure 1A). All 5 medullary carcinomas were also located in this cluster (Figure 1D.). On the contrary, most BRCAx samples and the only BRCA2-mutated sample located within the other cluster (Figure 1A.). When the samples were color-coded according to ER expression level as measured by microarrays, we observed that the clusters clearly differed in this aspect: the first cluster contained samples with low or negative ER expression, while the second consisted mostly of samples with high or medium expression level (Figure 1B.) When we marked the samples according to their molecular subtype (Figure 1C), it appeared that the basal versus luminal difference was the major factor that determined the distribution of the cancer samples into the two clusters. Thus, PCA confirmed that major sources of gene expression variability in breast cancer are ER expression level and molecular subtype, while tumors with BRCA1 gene inactivated by mutation or promoter methylation build the subgroup among other ER-negative, basal-like cancers.

5. DISCUSSION

The analysis presented by Hedenfalk et al. (5) is a landmark microarray study concerning hereditary cancer. These authors showed that hereditary breast tumors with proven mutation of BRCA1 gene differ significantly in terms of general gene expression from samples with BRCA2 mutation. The authors also pointed out that tumors from both BRCA1 or BRCA2 mutation-carriers differ from sporadic breast cancers. This problem was not directly addressed in any further microarray studies.

Thus, we aimed to verify the hypothesis of Hedenfalk and coauthors. However, in the Polish population BRCA2-induced breast cancer cases are very rare (e.g. sequencing of 100 DNA samples from patients with strong familial history of breast/ovarian cancer performed at the M. Sklodowska-Curie Memorial Cancer Center in Gliwice revealed 29 mutations in BRCA1 gene and only 3 mutations in BRCA2 gene, see also (17)). For this reason, we decided to ask, whether sporadic tumors differ from BRCA1-induced hereditary breast cancer in their gene expression profile.

In our study, the difference between BRCA1-mutated and sporadic tumors appeared to be of borderline significance: one could not exclude that the list of 41 probesets, selected in this comparison, was obtained by chance (p=0.065 in the global test). When we applied the same method of data analysis to the data from (5), we obtained an even shorter list of genes (11 clones) and its significance was also below the limit (p=0.065 in the global test). The same class prediction algorithm was used to classify BRCA1-mutated and sporadic tumors in both datasets. Classification accuracy was similarly relatively poor for our dataset and that from (5): 69% and 64%, respectively. Moreover, only two genes were common for the two classifiers: TOB1 and ALCAM. Thus it may be concluded that the difference in gene expression profile of BRCA1-mutated and sporadic breast cancer is very discrete.

We were curious how it happened that in the previous study (5) the authors observed such a distinct difference in gene expression pattern between BRCA1-mutated, BRCA2-mutated and sporadic tumors. We assumed that other sources of variability and sample stratification might have significant impact on their results. One should take into account that BRCA1 mutation-evoked breast cancer is typically ER-negative, while over a half of the population of sporadic breast cancer is ER-positive. Numerous microarray studies have shown ER status of breast cancer to be a very strong source of variability in gene expression pattern (18-20). In the meantime it also became obvious that ER-status and molecular subtype (as initially described by Sorlie et al. (7)) are tightly connected: lack of estrogen receptor expression is one of the hallmarks of basal-like tumors. In our study all tumors with BRCA1 mutation were ER-negative. Among the remaining samples there were 6 ER-positive tumors (17% of all samples); however, we excluded them from most comparisons or corrected the analyses for ER status. On the contrary, in the study of Hedenfalk et al., 45% of samples were ER-positive (5). In addition there were distinct differences in the frequency of ER-positive tumors between BRCA1-mutated group (none of 7 samples), sporadic group (4 of 7 samples) and BRCA2-mutated group (6 of 8 samples) in their study. We conclude that a weak concordance between our and their results comes from the fact that the pivotal influence of ER status have been regarded in the first study while not in the second.

The results of Q-PCR validation further supported our general conclusion that the differences in gene expression profile between BRCA1-mutation linked and sporadic breast cancers are of rather small scale and are strongly influenced by other sources of variability. The two genes, TOB1 and HIP2, showed slightly different expression in BRCA1-mutated tumors in comparison to non-mutated ones (p=0.04) when we analyzed only the samples from ER-negative tumors. However, TOB1 showed strong difference between ER-negative and ER-positive tumors (p=0.001). If ER-positive samples were also included in the analysis, the apparent difference between BRCA1-mutated and other tumors seemed to be an order of magnitude higher than previously (p=0.004). This phenomenon illustrates our opinion that majority of difference in gene expression profile usually ascribed to the presence of BRCA1 mutation may be a derivate of the difference in ER status and/or basal versus luminal difference. Second gene, HIP2 showed no differences between ER-positive and ER-negative samples. Thus, when analyzing its expression in a mixed population of tumors, the difference between BRCA1-mutated and non-mutated samples was of the same significance like in the analysis done on ER-negative group (p=0,04). In conclusion, HIP2 appears to be truly related to the BRCA1 mutation status but difference in its expression level is weak.

Despite its probably higher dependence on the ER status than on BRCA1 mutation status, TOB1 gene may be of special interest. It was selected in the study of Hedenfalk and confirmed in our reanalysis of their dataset as well as in the analysis of our own dataset. TOB1 is an antiproliferative protein that probably acts via transcriptional repression of several signaling pathways and by controlling post-transcriptional stability of target mRNAs (21-23). Following an in vitro studies, TOB1 was proposed as a novel radio-sensitizer, suitable for breast cancer therapy (24). Altogether, these facts suggest that TOB1 may be considered as a potential marker and/or therapeutic target in a selected cases of breast cancers.

Interestingly, our BRCA1-related signature contained also several genes from the gene set described by Kang et al., in the gastric cancer cell line resistant to DNA-damaging agent, doxorubicin (15). However, overexpression of these genes not obviously must correlate with doxorubicin resistance in BRCA1 mutated breast tumors, as they have impaired DNA repair and generally should be more sensitive to DNA-damaging chemotherapy. Thus, it may be proposed that this gene set may confer not only chemoresistance but some more general properties of cancer cells, although it requires confirmation in further studies.

It should be underlined that the results of our molecular analysis support the opinion of distinguished pathologists who have already suggested close relationship between BRCA1-mutated and basal breast cancers (24, 25). In fact, pathological and immunohistochemical characteristics of both types of breast cancer are nearly identical: both are described as predominantly grade 3 ductal or medullary carcinoma, ER-negative, PR-negative, almost without HER2neu amplification, frequently characterized by p53 mutation and p53 positive IHC staining, lymphocyte infiltrate, comedo-like necrosis, pushing margins and high mitotic indices. Similar are also prognostic uncertainties in patients with BRCA1-linked or basal-like breast cancer, each group being of not uniformly poor prognosis (26, 4). At the molecular level, the kinship between BRCA1-mutated and ER-negative basal-like breast tumors is reflected by a nearly common gene expression pattern. Already van't Veer et al. noted that BRCA1-mutated breast cancer samples, included in their patient outcome analysis, all fell into the cluster of ER-negative tumors. When Sorlie et al. (7) reanalyzed the dataset of van't Veer according to the molecular subtype, they noted that BRCA1-mutated cancers clustered together with basal-like tumors. Similar results were probably achieved by Desper et al. (27) who analyzed the data from (5) using phylogenetic classification tree. They observed that the BRCA1-mutated tumors all clustered together in one sub-tree, while BRCA2-mutated tumors, in another sub-tree. These clusters were located far from each other, on the distal parts of the tree, while sporadic tumors laid between them, mostly as leaves off main branch. With our present knowledge we may suppose that sporadic tumors of ER-negative and/or basal-like phenotype were located closer to the BRCA1-mutated cluster while ER positive and/or luminal-like sporadic tumors were closer to the BRCA2-mutated cluster. However, this cannot be confirmed as Desper et al. used sample descriptions that differ from the original.

To explain why BRCA1-mutated breast tumors have molecular profile nearly identical as other basal-like tumors one can assume that in the latter the BRCA1 pathway is truncated as well. This may be caused either by mutations or epigenetic inactivation of several crucial genes. Indeed, it appeared that 6 of non-mutated basal-like tumors had BRCA1 promoter methylation. Further studies are needed to identify potential other molecular events affecting BRCA1 pathway in basal-like breast cancer. However, even if we assume that all basal-like tumors have somehow truncated BRCA1 pathway, we must take into account that these tumors are not absolutely homogenous. This conclusion emerged from the analysis of essential biological traits that could be affected in BRCA1-mutated versus other breast tumors. This approach revealed differentially expressed several pathways and gene sets known previously to be connected with the presence of BRCA1 mutation. This were immune response-specific genes and the genes engaged in the pathway "BRCA1-dependent Ub-ligase activity" or belonging to the BRCA1 reporter gene set specified by (6). Thus, it is possible that different events leading to BRCA1 pathway truncation are expressed as slightly different phenotypes of ER-negative and/or basal-like breast cancer.

In conclusion, it may be suggested that the majority of BRCA1-linked breast cancers have a molecular profile of basal-like cancer. As a practical consequence, clinical observations and therapeutic recommendations that are true for basal-like breast cancer may also apply to the BRCA1-mutated tumors. Further studies are needed to unravel whether and which of BRCA1 partners could be affected in non-mutated basal-like breast cancer, how such putative distinct molecular events account for the heterogeneity of basal-like tumors and why tumors with truncated BRCA1 pathway develop mostly as basal-like subtype.

5. ACKNOWLEDGMENTS

Katarzyna Marta Lisowska and Volha Dudaladava equally contributed to this work. This work was supported by grants from the Polish Ministry of Science and Higher Education no. 3 P05A 060 25 and PBZ-MNiI-2/1/2005. We gratefully acknowledge Prof. Ewa Grzybowska, Ph.D., who participated in fund raising. We thank Aleksandra Pfeifer, M.Sc. for her help in data analysis and visualization.

6. REFERENCES

1. Breast Cancer Linkage Consortium: Pathology of familial breast cancer: differences between breast cancers in carriers of BRCA1 or BRCA2 mutations and sporadic cases. Lancet 349, 1505-10 (1997)
doi:10.1016/S0140-6736(96)10109-4

2. S. R. Lakhani, B. A. Gusterson, J. Jacquemier, J. P. Sloane, T. J. Anderson, M. J. van de Vijver, D. Venter, A. Freeman, A. Antoniou, L. McGuffog, E. Smyth, C. M. Steel, N. Haites, R. J. Scott, D. Goldgar, S. Neuhausen, P. A. Daly, W. Ormiston, R. McManus, S. Scherneck, B. A. Ponder, P. A. Futreal, J. Peto, D. Stoppa-Lyonnet, Y. J. Bignon, M. R. Stratton: Multifactorial analysis of differences between sporadic breast cancers and cancers involving BRCA1 and BRCA2 mutations. J Natl Cancer Inst 90, 1138-45 (1998)
doi:10.1093/jnci/90.15.1138
PMid:9701363

3. M. J. van de Vijver: The pathology of familial breast cancer: The pre-BRCA1/BRCA2 era: historical perspectives. Breast Cancer Res 1, 27-30 (1999)
doi:10.1186/bcr9
PMid:11250679    PMCid:138507

4. L. S. Teng, Y. Zheng, H. H. Wang: BRCA1/2 associated hereditary breast cancer. J Zhejiang Univ Sci B 9, 85-9 (2008)
doi:10.1631/jzus.B0710617
PMid:18257128    PMCid:2225488

5. I. Hedenfalk, D. Duggan, Y. Chen, M. Radmacher, M. Bittner, R. Simon, P. Meltzer, B. Gusterson, M. Esteller, O. P. Kallioniemi, B. Wilfond, A. Borg, J. Trent, M. Raffeld, Z. Yakhini, A. Ben-Dor, E. Dougherty, J. Kononen, L. Bubendorf, W. Fehrle, S. Pittaluga, S. Gruvberger, N. Loman, O. Johannsson, H. Olsson, G. Sauter: Gene-expression profiles in hereditary breast cancer. N Engl J Med 344, 539-48 (2001)
doi:10.1056/NEJM200102223440801
PMid:11207349

6. L. J. van't Veer, H. Dai, M. J. van de Vijver, Y. D. He, A. A. Hart, R. Bernards, S. H. Friend: Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530-6 (2002)
doi:10.1038/415530a
PMid:11823860

7. T. Sorlie, R. Tibshirani, J. Parker, T. Hastie, J. S. Marron, A. Nobel, S. Deng, H. Johnsen, R. Pesich, S. Geisler, J. Demeter, C. M. Perou, P.E . Lonning, P. O. Brown, A. L. Borresen-Dale, D. Botstein: Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci U S A 100, 8418-23 (2003)
doi:10.1073/pnas.0932692100
PMid:12829800    PMCid:166244

8. E. Grzybowska, H. Zientek, A. Jasinska, M. Rusin, P. Kozlowski, K. Sobczak, A. Sikorska, E. Kwiatkowska, L. Gorniak, E. Kalinowska, B. Utracka-Hutka, J. Wloch, E. Chmielik, W. J. Krzyzosiak: High frequency of recurrent mutations in BRCA1 and BRCA2 genes in Polish families with breast and ovarian cancer. Hum Mutat 16, 482-90 (2000)
doi:10.1002/1098-1004(200012)16:6<482::AID-HUMU5>3.0.CO;2-O

9. E. Grzybowska, M. Sieminska, H. Zientek, E. Kalinowska, J. Michalska, B. Utracka-Hutka, J. Rogozinska-Szczepka, M. Kazmierczak-Maciejewska: Germline mutations in the BRCA1 gene predisposing to breast and ovarian cancers in Upper Silesia population. Acta Biochim Pol 49, 351-6 (2002)
PMid:12362976

10. B. Gorski, T. Byrski, T. Huzarski, A. Jakubowska, J. Menkiszak, J. Gronwald, A. Pluzanska, M. Bebenek, L. Fischer-Maliszewska, E. Grzybowska, S. A. Narod, J. Lubiński: Founder mutations in the BRCA1 gene in Polish families with breast-ovarian cancer. Am J Hum Genet 66, 1963-8 (2000)
doi:10.1086/302922
PMid:10788334

11. J. P. Struewing, P. Hartge, S. Wacholder, S. M. Baker, M. Berlin, M. McAdams, M. M. Timmerman, L. C. Brody, M. A. Tucker: The risk of cancer associated with specific mutations of BRCA 1 and BRCA 2 among Ashkenazi Jews. NEJM 336, 1401-1408 (1997)
doi:10.1056/NEJM199705153362001
PMid:9145676

12. P. Chomczynski and N. Sacchi: The single-step method of RNA isolation by acid guanidinium thiocyanate-phenol-chloroform extraction. Anal Biochem 162, 156-9 (1987)
doi:10.1016/0003-2697(87)90021-2

13. M. Esteller, J. M. Silva, G. Dominguez, F. Bonilla, X. Matias-Guiu, E. Lerma, E. Bussaglia, J. Prat, I. C. Harkes, E. A. Repasky, E. Gabrielson, M. Schutte, S. B. Baylin, and J. G. Herman: Promoter hypermethylation and BRCA1 inactivation in sporadic breast and ovarian tumors. J Natl Cancer Inst 92, 564-569 (2000)
doi:10.1093/jnci/92.7.564
PMid:10749912

14. E. Matros, Z. C. Wang, G. Lodeiro, A. Miron, J. D. Iglehart, and A. L. Richardson: BRCA1 promoter methylation in sporadic breast tumors: relationship to gene expression profiles. Breast Cancer Res Treat 91, 179-186 (2005)
doi:10.1007/s10549-004-7603-8
PMid:15868446

15. H. C. Kang, I. J. Kim, J. H. Park, Y. Shin, J. L. Ku, M. S. Jung, B. C. Yoo, H. K. Kim, J. G. Park: Identification of genes with differential expression in acquired drug-resistant gastric cancer cells using high-density oligonucleotide microarrays. Clin Cancer Res 10 (1 Pt 1), 272-84 (2004)
doi:10.1158/1078-0432.CCR-1025-3
PMid:14734480

16. R. Desper, J. Khan, A. A. Schaffer: Tumor classification using phylogenetic methods on expression data. J Theor Biol 228, 477-96 (2004)
doi:10.1016/j.jtbi.2004.02.021
PMid:15178197

17. B. Gorski, A. Jakubowska, T. Huzarski, T. Byrski, J. Gronwald, E. Grzybowska, A. Mackiewicz, M. Stawicka, M. Bebenek, D. Sorokin, Ł. Fiszer-Maliszewska, O. Haus, H. Janiszewska, S. Niepsuj, S. Gozdz, L. Zaremba, M. Posmyk, M. Pluzanska, E. Kilar, D. Czudowska, B. Wasko, R. Miturski, J. R. Kowalczyk, K. Urbański, M. Szwiec, J. Koc, B. Debniak, A. Rozmiarek, T. Debniak, C. Cybulski, E. Kowalska, A. Toloczko-Grabarek, S. Zajaczek, J. Menkiszak, K. Medrek, B. Masojc, M. Mierzejewski, S. A. Narod, J. Lubinski: A high proportion of founder BRCA1 mutations in Polish breast cancer families. Int J Cancer 110, 683-6 (2004)
doi:10.1002/ijc.20162
PMid:15146557

18. S. Gruvberger, M. Ringner, Y. Chen, S. Panavally, L. H. Saal, A. Borg, M. Ferno, C. Peterson, P. S. Meltzer: Estrogen receptor status in breast cancer is associated with remarkably distinct gene expression patterns. Cancer Res 61, 5979-84 (2001)
PMid:11507038

19. A. Weisz, W. Basile, C. Scafoglio, L. Altucci, F. Bresciani, A. Facchiano, P. Sismondi, L. Cicatiello, M. De Bortoli: Molecular identification of ERalpha-positive breast cancer cells by the expression profile of an intrinsic set of estrogen regulated genes. J Cell Physiol 200, 440-50 (2004)
doi:10.1002/jcp.20039
PMid:15254972

20. V. N. Kristensen, T. Sorlie, J. Geisler, A. Langerod, N. Yoshimura, R. Karesen, N. Harada, P. E. Lonning, A. L. Borresen-Dale: Gene expression profiling of breast cancer in relation to estrogen receptor status and estrogen-metabolizing enzymes: clinical implications. Clin Cancer Res 11, 878-83 (2005)

21. S. Jia, A. Meng: Tob genes in development and homeostasis. Dev Dyn 236, 913-21 (2007)
doi:10.1002/dvdy.21092
PMid:17304515

22. T. Miyasaka, M. Morita, K. Ito, T. Suzuki, H. Fukuda, S. Takeda, J. Inoue, K. Semba, T. Yamamoto: Interaction of antiproliferative protein Tob with the CCR4-NOT deadenylase complex. Cancer Sci 99, 755-61 (2008)
doi:10.1111/j.1349-7006.2008.00746.x
PMid:18377426

23. N. Ezzeddine, T. C. Chang, W. Zhu, A. Yamashita, C. Y. Chen, Z. Zhong, Y. Yamashita, D. Zheng, A. B. Shyu: Human TOB, an antiproliferative transcription factor, is a poly(A)-binding protein-dependent positive regulator of cytoplasmic mRNA deadenylation. Mol Cell Biol 27, 7791-801 (2007)
doi:10.1128/MCB.01254-07
PMid:17785442    PMCid:2169145

24. Y. Jiao, C. M. Ge, Q.H. Meng, J. P. Cao, J. Tong, S. J. Fan: Adenovirus-mediated expression of Tob1 sensitizes breast cancer cells to ionizing radiation. Acta Pharmacol Sin 28, 1628-36 (2007)
doi:10.1111/j.1745-7254.2007.00647.x
PMid:17883950

25. S. R. Lakhani, J. S. Reis-Filho, L. Fulford, F. Penault-Llorca, M. van der Vijver, S. Parry, T. Bishop, J. Benitez, C. Rivas, Y. J. Bignon, J. Chang-Claude, U. Hamann, C. J. Cornelisse, P. Devilee, M. W. Beckmann, C. Nestle-Krämling, P. A. Daly, N. Haites, J. Varley, F. Lalloo, G. Evans, C. Maugard, H. Meijers-Heijboer, J. G. Klijn, E. Olah, B. A. Gusterson, S. Pilotti, P. Radice, S. Scherneck, H. Sobol, J. Jacquemier, T. Wagner, J. Peto, M. R. Stratton, L. McGuffog, D. F. Easton; Breast Cancer Linkage Consortium: Prediction of BRCA1 status in patients with breast cancer using estrogen receptor and basal phenotype. Clin Cancer Res 11, 5175-80 (2005)
doi:10.1158/1078-0432.CCR-04-2424
PMid:16033833

26. N. C. Turner, J. S. Reis-Filho: Basal-like breast cancer and the BRCA1 phenotype. Oncogene 25, 5846-53 (2006)
doi:10.1038/sj.onc.1209876
PMid:16998499

27. L. Da Silva, C. Clarke, S. R. Lakhani: Demystifying basal-like breast carcinomas. J Clin Pathol 60, 1328-32 (2007)
doi:10.1136/jcp.2006.041731
PMid:17496191

Key words hereditary breast cancer, BRCA1 gene mutation, gene expression signature, ER-status of the tumor, molecular subtypes of breast cancer

Send correspondence to: Katarzyna Marta Lisowska, Department of Tumor Biology, Maria Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Gliwice Branch, ul. Wybrzeze Armii Krajowej 15, 44-101 Gliwice, Poland, Tel.: 48322789888, Fax.:. 48322313512, E-mail kasial@io.gliwice.pl