Establishment and characterization of two breast cancer xenografts in immunodeficient mice for studies on hormone-dependent and hormone-independent tumor growth, progression and invasion

Many genes/proteins have been involved in cellular transformation. However, a systems-level understanding of this pathological process is still absent. To address this question we developed a strategy to generate preliminary models of the networks around known cancer gene products. By examining functional genomic information as gene expression profiles, disease-associated genetic networks and systems-level integrated networks, we defined a 'breast cancer gene module' with predicted novel functional relationships to known breast cancer tumor suppressors. Genes within this module encoded for novel functional relationships with BRCA1 and BRCA2. Among the novel components identified we functionally characterized the hyaluronan-mediated motility receptor (HMMR, human Rhamm), which defines a BRCA1/BRCA2 protein network involved in the control of centrosome number and chromosome segregation. Biochemical data reveal that BRCA1/BRCA2 and HMMR form complexes, that HMMR is ubiquitinated by BRCA1/BARD1, and that BRCA1 and HMMR together regulate centrosome duplication in tissue culture cell lines derived from breast tissue. Our results indicate that similar strategies could help to build and complete other cancer-related cellular networks, and thus to understand how they are affected and/or contribute to cellular transformation.

loss of TGF-β responsiveness in fibroblasts resulted in intraepithelial neoplasia in prostate and invasive squamous cell carcinoma of the forestomach with high penetrance by 6 weeks of age. Both epithelial lesions were associated with an increased abundance of stromal cells. Activation of paracrine hepatocyte growth factor (HGF) signaling was identified as one possible mechanism for stimulation of epithelial proliferation. TGF-β signaling in fibroblasts thus modulates the growth and oncogenic potential of adjacent epithelia in selected tissues. More recently, we have examined the effects of Tgfbr2 fspKO fibroblasts on normal and transformed mammary epithelium. We analyzed the role of TGF-β signaling by stromal cells in mammary tumor progression. To avoid the possibility of endogenous wild-type fibroblasts masking potential effects of Tgfbr2 fspKO cells on tumor progression, we implanted PyVmT mammary carcinoma cells with Tgfbr2 fspKO or wildtype fibroblasts in the subrenal capsule of nude mice. Mammary tumor cells implanted with Tgfbr2 fspKO cells exhibited an increase in tumor growth and intravasation associated with an increase in tumor cell survival, proliferation and an increase in tumor angiogenesis compared with tumor cells implanted with control fibroblasts. We demonstrated increased expression of several growth factors by Tgfbr2 fspKO fibroblasts compared with control fibroblasts in primary culture. These included HGF, MSP and TGF-α. There was an increase in tumor cell activating phosphorylation of the cognate receptors, c-Met, RON, erbB1, and erbB2 in carcinomas accompanied by Tgfbr2 fspKO fibroblasts. The Tgfbr2 fspKO mouse model illustrates that a signaling pathway known to suppress cell-cycle progression when activated in epithelial cells can also have an indirect inhibitory effect on epithelial proliferation when activated in adjacent stromal fibroblasts in vivo. Loss of this inhibitory effect can result in increased epithelial proliferation and may even progress to invasive carcinoma in some tissues.

S.03 Genomic analysis of human breast cancer in families and populations
cancer in the general population remains unclear. Furthermore, the methodology of ATM mutation detection is still laborious and costly. Because the ATM protein kinase phosphorylates such a wide array of downstream targets, many pathways to oncogenesis are possible and largely unexplored. What seems clear is that: A-T heterozygotes are at a fourfold to fivefold increased risk of breast cancer, although confidence intervals are large; and the spectrum of ATM mutations is distinct for A-T families versus breast cancer cohorts. Only a handful of mutations have been identified in both A-T families and breast cancer cohorts. Missense mutations represent <10% of mutations in A-T patients and >80% in breast cancer cohorts. ATM missense mutations are also more common in some leukemias and lymphomas. Experimental data suggest that some missense mutations represent dominant interfering mutations [1][2][3][4][5]; however, clinical support for a dominant interfering model is minimal in family studies, suggesting either that the model is flawed or that penetrance of these mutations is very low. Histological classifications of breast cancer are largely grouped as genetically homogeneous models, although expression microarray data suggest otherwise. Other studies have associated ATM-SNPs with increased breast cancer risk; however, just three SNP haplotypes across the ATM locus include ~95% of a global population, and this must be factored into such association models. Without the benefit of mRNA analyses, of minigene experiments, of Maximum Entropy Scores, of site-directed mutagenesis or of functional assays of ATM activity, most 'missense' mutations cannot be reliably distinguished from polymorphisms or from other types of mutations, such as splicing variants that lead to secondary stop codons. Our recent analyses have focused on two ATM missense mutations, 7271T>G and IVS10-6T>G. For each of these mutations, there are published functional data suggesting that they act as dominant interfering mutations, and epidemiological data suggesting a role in breast cancer. Some family studies of the 7271T>G mutation suggest that it is a highly penetrant breast cancer susceptibility allele. However, its infrequency in the population means that its contribution to breast cancer risk is slight and it is possible that 7271T>G represents only one of a diverse array of uncommon ATM mutations leading to increased cancer risk. We found that the frequency of the IVS10-6T>G mutation was not increased in breast cancer cases as compared with controls. Furthermore, the evidence that IVS10-6T>G is an A-T mutation is called into question by our recent evidence that, in the one known example of a homozygous IVS10-6T>G individual with A-T, a homozygous mutation at 5644C>T was also present (Purayidom and colleagues, submitted). Taken together, these studies suggest that whereas no single ATM mutation impacts significantly upon breast cancer risk, it may be possible to group mutations that do modulate risk for breast cancer based on their phenotypic effects. This group of patients might benefit substantially from a therapeutic approach to correct missense mutations.
S5 markers in breast cancer samples. This means that under the conditions of a constant high level of E2, ERα does not reappear in the nucleus. A similar situation exists during pregnancy when there is a constant high level of E2 and there is no ERα in the mammary epithelium. This resistance to the proliferative response to E2 in the presence of a constant high dose of E2 probably explains the very successful use of high-dose E2 in the treatment of breast cancer. ERβ, on the other hand, appears to have a differentiative role not a proliferative role in the mammary gland, and the lactating rodent mammary gland of ERβ -/mice does not express gap junction and adhesion proteins, typical indicators of fully differentiated cells. In recent years there have been several publications showing that ERβ is expressed in human breast cancer, and conclusions and speculations about a causative role for ERβ in breast cancer development and/or progression have been made. We have studied 500 frozen breast biopsies in collaboration with Prof. RC Coombes, London, in order to clarify the role of ERβ in normal and malignant breast. In this study we measured ERα and ERβ proteins by several techniques (immunohistochemistry, western blotting, ligand binding in sucrose gradients, and RT-PCR) in various human samples obtained from both benign breast and malignant breast. We found that ERβ is the predominant estrogen receptor in the normal mammary gland and in benign breast disease. There is very little ERα in the normal mammary gland. This low expression of ERα is one of the striking differences between rodents and humans. This is in stark contrast to ERβ, which is expressed in 80% of epithelial cells and is also present in the stroma. We found that ERα is abundantly expressed in invasive and in situ ductal carcinoma but not in medullary cancer. ERβ is also expressed in breast cancer, both ductal and medullary. In this study we also found that, in the human breast, the major ER in breast stroma is ERβ. This surprising finding has necessitated several new lines of investigation about the function of ERβ in the breast. It has long been thought that ERα in the stroma was responsible for secretion of growth factors in response to E2 and that these growth factors were responsible for epithelial cell proliferation. The discovery that it is ERβ that is present in the stroma might suggest a role of ERβ in growth factor secretion.

S.12
Molecular approaches to understanding pregnancyinduced protection against breast cancer S.15 A breast cancer progression model: the importance of three-dimensional tissue architecture and metalloproteinases MJ  Previous studies from our laboratory have shown that non-malignant and malignant cells can be distinguished easily and rapidly by their morphology and growth rate when cultured in three-dimensional (3D) laminin-rich basement membrane but not when cultured on traditional tissue culture plastic (two-dimensional [2D]) [1,2]. In addition, we have shown that cellular responses to signaling inhibitors and apoptotic agents differ in cells cultured in 2D versus 3D [3,4]. This applies also to our finding with reverted tumor cell lines [3][4][5][6][7][8]. In this presentation, I will address two inter-related topics. First, we asked how the 3D morphology and gene expression profiles for a panel of 60 breast cancer cell lines for which the Gray laboratory has obtained 2D expression as well as CGH profiles may differ, and whether any of the surrogate genes or phenotypes could track with response to therapy. The cell lines examined so far fell into four distinct morphologies of 'round', 'mass', 'grape-like' and 'stellate'. An ANOVA analysis of Affymetrix gene expression profiles for each of these cell lines was used to identify genes, the expression profiles of which could distinguish the other known parameters of the cultured cells. Of the 22,283 genes on the Affymetrix 133A chip, ~5800 genes were identified where expression patterns differed between different cell lines both in 2D and 3D, and ~2000 genes were identified where expression differed between the non-malignant and malignant cell lines. About 700 genes differed between 2D and 3D, and ~800 correlated with the morphological differences seen in 3D. These genes fall into a number of functional classes, which we are currently analyzing to identify common signaling themes and/or morphological regulators that will be tested by manipulation of expression and correlated with therapeutic response of these cell lines in 2D and 3D to Herceptin and other chemotherapeutic drugs. Second, we have also shown previously that loss of basement membrane in both cultured mammary mouse cells [9] and in transgenic animals led to epithelial to mesenchymal transition (EMT) and mammary tumors [10]. We have now determined the molecular pathways induced by MMP-3 to lead to EMT and genomic instability via production of reactive oxygen species [11]. These mechanisms will be discussed.  (DOI 10.1186/bcr1059) Advances on several fronts have led to increases in survival duration and to reduced mortality in patients with breast cancer. These include improved procedures for earlier detection, optimization of combined surgical and radiotherapy, and use of optimized selective estrogen receptor modifiers (SERMS) and new chemotherapeutic strategies including gene-targeted therapies. In addition, molecular stratification strategies have been developed that stratify patients according to outcome. Stratification based on measurement of expression 'signatures' have been particularly effective and seem likely to improve treatment strategies. Patients at increased risk of progressive disease can be offered standard of care chemotherapy. However, some of these patients do not respond well to these treatments and current stratification strategies provide little information to guide treatment of these patients. This study of tumors from patients treated according to standard of care identifies genomic and coordinated transcriptional aberrations -especially amplification at 11q, and 20q in tumors with the luminal A expression phenotype, and at 17q in tumors associated with the ERBB2 expression phenotype -that are strongly associated with poor response to such treatment. Our study identifies genes in these regions of amplification that can be assessed to identify patients that will respond poorly to the current standard of care and that are targets for therapies that will be effective against these poorly responding tumors. Interesting, this study also shows that patients with basal-like tumors do not have substantially shorter survival durations than patients with luminal-like tumors, suggesting that basal-like tumors respond well to the adjuvant adriamycin and cyclophosphamide therapies employed during their treatment. We performed comprehensive molecular analysis of each cell type composing normal breast tissue and in situ and invasive breast carcinomas. Gene expression profiles were analyzed using serial analysis of gene expression, genetic changes were analyzed by single nucleotide polymorphism arrays, while epigenetic changes were analyzed using methylation-specific digital karyotyping. Based on these data we determined that gene expression and epigenetic changes occur in all cell types during breast cancer progression, while genetic alterations were only detected in tumor epithelial cells. Many of the differentially expressed genes encode for secreted proteins and receptors suggesting alterations in autocrine and paracrine interactions in breast tumorigenesis. Two of these genes, the CXCL14 and CXCL12 chemokines, overexpressed in tumor myoepithelial cells and in myofibroblasts, respectively, bind to receptors on epithelial cells and enhance their proliferation, migration, and invasion. Chemokines may thus play a role in breast tumorigenesis by acting as paracrine factors. The role of these chemokines, and myoepithelial and stromal cells in the progression of in situ carcinomas to invasive carcinomas was investigated using a xenograft model of human ductal carcinoma in situ. Based on our studies we determined that changes in the tumor microenvironment and epithelial-myoepithelial and epithelial-stromal cell interactions play an important role in breast tumor progression. Methods The estrogen receptor (ER) and proliferation (Ki67) were evaluated and compared in TDLUs and HELUs in the same breasts (n = 250) by immunohistochemistry. Apoptosis was also assessed by the TUNEL assay. The rate of ER expression in proliferating cells was assessed by dual-labeled immunofluorescence. Comprehensive gene expression profiling was performed in a subset of samples (currently six matched pairs of TDLUs and HELUs) using RNA isolated from microdissected formalin-fixed paraffin-embedded breast tissue samples and Affymetrix U133-X3P microarrays analyzed by dCHIP software. Xenografts of human TDLUs and HELUs were prepared by implanting isolated epithelial cells into cleared mammary fat pads of estrogenstimulated immune-compromised mice 'humanized' by prior local injection of immortalized (h-tert transfected) human mammary fibroblasts.

Results
The average ER expression was significantly elevated in HELUs compared with adjacent TDLUs (85% vs 30% positive cells, respectively; P < 0.0001). Proliferation was significantly higher (6% vs 2%; P < 0.0001) and apoptosis was significantly lower (0.6% vs 0.2%; P < 0.001) in HELUs than TDLUs. There was a large increase in the proportion of ER-positive proliferating cells in HELUs compared with TDLUs (35% vs 4%; P < 0.0001). In preliminary analysis of the microarray results, HELUs and TDLUs segregated perfectly in unsupervised hierarchical comparisons. In supervised comparisons, many (n = 74) genes showed >3-fold (P < 0.05) differences in expression, with 45 relatively up (from 3.5-fold to 9.5-fold) and 29 relatively down (from 3.5-fold to 12.5-fold) in HELUs versus TDLUs. Especially prominent elevations in HELUs included several genes involved in G-protein signaling, the retinoic acid pathway, and detoxification. Prominent decreases included genes involved in cell cycle inhibition, apoptosis, differentiation, and water transport. Differences were also noted in the expression of genes for ligands of the epidermal growth factor receptor between HELUs and TDLUs. Several fresh human samples of TDLUs and HELUs are in various stages (currently up to generation four) of implantation in mice in attempts to establish stable xenografts.
Conclusions HELUs are one of the earliest histologically recognizable lesions in the human breast with premalignant potential. They show striking elevations of ER, which may partially explain the hyperplasia leading to their development from TDLUs through increased proliferation and decreased apoptosis, which are both regulated by estrogen. DNA microarrays reveal many additional differences in the expression of genes involved in growth and differentiation. Human xenograft models are under development to support mechanistic studies of these genes to understand their roles in the development and progression of HELUs and how to prevent it.  10.1186/bcr1062) Pathogenesis of cancer begins as hyperplastic lesions; some lesions remain benign, while others progress to malignancy. An increase in cell proliferation rates and changes in tissue architecture are two properties commonly observed in hyperplastic lesions. A great deal is known about the molecular events that regulate cell proliferation and the knowledge gained is widely used for development of diagnostic and treatment tools. Our understanding of the mechanisms that deregulate tissue architecture is poor, and hence it is understandable that the use of architectural features to determine prognosis of early lesions has varying success. We used polarized epithelial cells and an inducible method of ErbB2 activation to investigate whether the cell architecture influences ErbB2-induced gene expression and to investigate how activation of ErbB2 disrupts epithelial cell architecture. Activation of ErbB2 in three-dimensional epithelial acini-like structures leads to expression of a unique set of genes that was not observed when ErbB2 was activated in cells grown on plastic dishes, suggesting that the cell architecture can have significant influence on ErbB2-induced gene expression. To investigate the effect of ErbB2 activation on epithelial architecture, we activated ErbB2 in polarized epithelial cells. ErbB2 induced a loss in apical-basal polarity, re-initiated proliferation and induced multilayering of epithelial sheets. These changes correlate with the ability of ErbB2 to regulate the Par complex, a protein complex known to regulate establishment of epithelial cell polarity. Inactivation of atypical protein kinase C, a component of the Par complex, cooperates with ErbB2 to disrupt polarized epithelial cells, suggesting that the Par complex is a mediator of ErbB2-induced effects on polarized epithelial cells. In addition, we identify tricellular junctions, and not bicellular junctions, as a novel site for ErbB2 action in cultured epithelial cells and in primary breast cancer. We are thus beginning to gain novel insights into the molecular mechanisms that regulate early lesions. Existing methods to detect breast cancer in asymptomatic patients have limitations, and there is a need to develop more accurate and convenient methods. Especially, an accurate method for breast cancer detection based on peripheral blood as a clinical sample will be highly desirable because of the easy accessibility and less-invasive nature by which samples can be obtained. Results demonstrating that peripheral blood can be used to develop a gene expression based test for early detection of breast cancer will be presented. The rationale for using blood cells as monitors for a malignant disease elsewhere in the body is based on the hypothesis that a malignant growth will cause characteristic changes in the biochemical environment of blood. These changes will affect the expression pattern of certain genes in blood cells.
We initially conducted a pilot study where the expression pattern of 1368 genes in peripheral blood cells of 24 females with breast cancer and 32 females with no signs of this disease were analyzed using macroarrays and the expression data analyzed by PAM. The results were validated using a standard leave-one-out cross-validation approach. We were able to identify a set of genes that correctly predicted the diagnostic class in at least 82% of the samples. The majority of the identified genes had a decreased expression in samples from breast cancer patients, and predominantly encoded proteins implicated in ribosome production and translation control. In contrast, the expression of some defence-related genes was increased in samples from breast cancer patients. In order to revalidate these findings and to increase the repertoire of informative genes, we have now extended the study with a larger number of breast cancer and non-breast cancer samples and used Agilent WG oligo arrays for large-scale gene expression analysis. The preliminary analysis of the data supports our previous finding that a blood-based gene expression test can potentially be developed to detect breast cancer in asymptomatic patients. The epithelial components of the breast are thought to arise from a stem cell population that is capable of both self-renewal and lineagespecific differentiation. We and others have hypothesized that mammary stem cells or their immediate progeny are targets for transformation during carcinogenesis. Normal stem cells and carcinoma cells share many characteristics including self-renewal capacity, telomerase expression, ability to differentiate, resistance to apoptosis, and ability to home to specific sites. Mammary transformation may require dysregulation of pathways that control normal stem cell selfrenewal such as Notch, Wnt, Hedgehog, and Bmi-1. In order to study these pathways in normal mammary development, we have developed an in vitro culture system in which primary human epithelial cells isolated from reduction mammoplasties are cultured as 'mammospheres' on non-adherent surfaces. Cells within mammospheres are able to self-renew, as well as to differentiate into all the lineages found in the mammary gland. Utilizing this system, we demonstrate bi-directional interaction between Notch and Hedgehog signaling and Bmi-1 in the regulation of stem cell self-renewal. When mammospheres are admixed with irradiated human mammary fibroblasts and implanted into the cleared fatpads of NOD/SCID mice, they are able to reconstitute the ductal alveolar structures found in the human mammary gland. The stem cell model of carcinogenesis may also provide a partial explanation for the generation of cellular heterogeneity seen within mammary tumors. Using flow cytometry, we have identified a small population of cells within primary or metastatic breast cancers that bear the cell surface phenotype ESA + CD44 + CD24 -/low Lineagethat have the properties of human tumor stem cells. As few as 200 of these cells are able to reproducibly generate tumors in NOD/SCID mice, while the vast majority of cells in these tumors that lack this phenotype are incapable of tumor formation even when tens of thousands of cells are injected. Consistent with a stem cell model, tumorigenic cells generate tumors that recapitulate the phenotypic heterogeneity found in the original tumors. We have demonstrated that pathways that control normal stem cell self-renewal, such as Hedgehog, are activated in mammary tumor stem cells, compared with their differentiated progeny. Despite progress in breast cancer therapeutics, metastatic breast cancer remains an incurable disease. Current therapies that have been developed by virtue of their ability to induce tumor regression may selectively target more differentiated cells in tumors, while leaving the tumor stem cell population intact, accounting for treatment resistance and relapse. Multiple mechanisms may account for this resistance to apoptosis, including increased expression of antiapoptotic genes, increased DNA repair mechanisms, and transporter proteins such as BCRP found in the tumor stem cell population. The targeting of stem cell self-renewal pathways such as Hedgehog or Notch may thus provide a novel and more effective approach for the treatment of advanced breast cancer.  10.1186/bcr1065) HER2 or c-ERBB2/neu is a member of the epidermal growth factor receptor (EGFR) family and encodes a tyrosine kinase receptor. Overexpression of HER2 protein is generally attributable to gene amplification. HER2 is overexpressed in 20-30% of primary invasive breast carcinomas and in a greater proportion of in situ breast cancers. Invasive breast cancers that overexpress HER2 are generally higher stage, show lymph node positivity, and have higher S-phase. Moreover, they are often associated with poor prognosis, particularly in node-positive patients. Microarray studies have subdivided breast cancers into several subtypes. HER2-overexpressing ER-negative tumors are generally classified within a single subtype denoted ERBB2-overexpressing. However, ERpositive HER2-overexpressing tumors are usually intermixed with other ER-positive tumors that do not show HER2 overexpression. Our recent population-based study evaluating HER2 overexpression and hormone receptor status has unexpectedly found that the majority of HER2-overexpressing tumors are hormone receptor-positive and are more common than HER2-overexpressing ER-negative breast cancers. This implies that the ERBB2-overexpressing molecular subtype, which is associated with ER-negative status, only includes a minority of HER2-overexpressing tumors. We therefore studied gene expression patterns of HER2-overexpressing breast cancers and found several tumor subtypes with distinctive molecular signatures. These ERBB2overexpressing subtypes spanned the range of hormone receptor status and highlighted different biological characteristics. Since the clinical course varies among patients with HER2-positive tumors, as does their response to targeted therapy, differences in global gene expression among HER2-overexpressing tumors could be important in distinguishing patients for the design and delivery of individualized targeted therapies. To assess the oncogenic role of cyclin E-LMW as compared with fulllength cyclin E, we examined the consequences of overexpressing these isoforms in the mammary glands of transgenic mice using the MMTV promoter. Four constructs were generated: MMTV-M46A coding for the full-length cyclin E (EL1), MMTV-EL1/EL4 coding for EL1 and the isoform translated at methionine 46 (EL4), and MMTV-T1 and MMTV-T2 coding for the isoforms generated by elastase cleavage at the first site (EL2 + EL3) and at the second site (EL5 and EL6), respectively. For each construct at least two transgenic lines were established. Transgene expression was demonstrated by RT-PCR, northern blotting and western blotting. Overexpression of cyclin E was seen in more than 90% of ductal and lobular cells of the mammary glands for each independent line. Mammary-specific LMW cyclin E overexpression induced extensive abnormalities at 2 months, including perturbed architecture, polyploidy, anysocytosis and apoptosis. Wholemount preparations of mammary glands at different development stages showed that overexpression of EL1/EL4 and cyclin E-T1 induced growth delay, while at 6 months of age an increased proportion of cells in the S phase was found (25.6 ± 5.6% for EL1/EL4, 9.0 ± 2.7% for T1 compared with 3.9 ± 1.9% for nontransgenic animals). We observed a 34% (13/38) incidence of mammary adenocarcinomas in the EL1/EL4 transgenic lines with a mean latency of 18.3 months, and observed a 20% (5/25) incidence in the T1 transgenic lines with a mean latency of 17.1 months. The tumor incidence rate of the other transgenic lines, M46A and T2, are still unknown due to the young age of the mice (all under 7 months of age) and the long latency of cyclin E-mediated tumor generation. Thirty percent (4/13) of the EL1/EL4 and 40% (2/5) of the T1 tumor-bearing animals developed lung metastasis. The tumors induced by the EL1/EL4 and T1 transgenes were mainly solid adenocarcinomas with very little differential to glandular for EL1/EL4 and mostly glandular for T1. Since p53 alterations are common in human breast carcinomas, we bred a T1 line with p53 +/mice. The T1 × p53 +/cross generated tumors that are much more malignant than the T1 tumors; the incidence increased to 100%, with a much shorter latency of 11 months. Biochemical analysis of the tumors revealed that 64% (9/14) retained cyclin E expression and that, on average, the cyclin E-overexpressing tumors had threefold higher cyclin E kinase activity than the non-cyclin E-expressing tumors. Taken together, these data indicate that tumor progression in cyclin E transgenic mice follow sequential steps of dysplasia, mammary intraepithelial neoplasia and invasive/metastatic tumors. Collectively, the biochemical and biological differences between the full-length and the LMW isoforms of cyclin E provide a molecular mechanism for the poor clinical outcome observed in breast cancer patients harboring tumors expressing high levels of the LMW forms of cyclin E. The transgenic mouse model system can serve as a useful system in which to study the mechanisms responsible for LMW cyclin E-induced genetic instability and may help identify those factors that promote tumor progression and metastasis. The properties of the LMW forms of cyclin E suggest that they are not just surrogate markers of poor outcome, but that they are bona fide mediators of aggressive disease and potential therapeutic targets for patients whose tumors overexpress these forms.

S.22
drugs that produce specific apoptotic chemotherapeutic mechanisms that cause tumor shrinkage in mice without deleterious side effects. The small natural product b-lapachone is specifically apoptotic to a variety of cancer cells. It synergizes strongly with taxol. It seems to have several mechanisms of lethality depending on the tumor type and the drug concentration. One mechanism is to elevate the major S-phase transcription factor E2F-1, to an apoptotic concentration [1]. It is now in clinical trial. Tumor cells often mutate to apoptosis resistance; for example, by inactivating the p53 protein. We have reported that Go6976, a kinase inhibitory small molecule, can decrease activation of the anti-apoptotic transcription factor NF-κB [2]. These two novel therapies thus specifically cause cancer cell apoptosis; one by increasing an apoptotic factor, and the other restoring apoptosis by decreasing an anti-apoptotic factor. Conclusion Diploid tumors are particularly useful subjects for this approach, revealing complex rearrangements and repeated sequential amplification events on certain chromosomes that provide unique insights into the genomic progression of the disease. First, the fine structure of these amplification clusters, as detected by ROMA and quantitatively validated by FISH, provides extremely high-resolution 'pointers' to potential novel oncogenes, since many of the detected amplicons contain only one or two known or prospective genes. Second, FISH patterns provide a means for interpretation of the mechanism of these events. Third, the reproducibility and frequency of these events, especially in very early stage tumors, provides insight into the earliest chromosomal events in breast cancer. Finally, we have identified correlations between certain sets of rearrangement events and clinically relevant parameters such as long-term survival. These correlations may enable novel and powerful prognostic indicators for breast cancer and other cancers when more samples can be examined. We used gene expression profiling with DNA microarrays harboring 25,000 genes on 78 primary breast cancers of young lymph-nodenegative patients to establish a signature, predictive for a short interval to distant metastases. This 'poor prognosis' signature consists of genes involved in the cell cycle, invasion and angiogenesis. The prognosis signature is superior to currently available clinical and histopathological prognostic factors in predicting a short interval to distant metastases (odds ratio = 18 [95% confidence interval = 3.3-94], P < 0.001, multivariate analysis). We have validated our findings of this poor prognosis profile on a large unselected consecutive series of LN0 as well as lymph-node-positive (LN+) young breast cancer patients (n = 295). The analyses confirm that the profile is a strong independent factor in predicting outcome of disease for LN0 patients in general (10-year overall survival for the good prognosis profile 96% vs 50% for the poor prognosis profile). Furthermore, the profile is also powerful for LN+ patients. At present, the prognostic significance of the 70 genes is tested in older breast cancer patients. Nowadays, consensus guidelines in the management of breast cancer select up to 95% of lymph-node-negative young breast cancer patients for adjuvant systemic therapy (e.g. NIH and St Gallen consensus criteria). As 70-80% of these patients would have remained diseasefree without this adjuvant treatment, these patients are 'overtreated'. The 'poor prognosis' signature provides a novel strategy to accurately select patients who would benefit from adjuvant systemic therapy and can greatly reduce the number of patients that receive unnecessary treatment.

S.27
Our data revealed that already small tumors display the metastatic signature, and recent results show that the molecular program established in a primary breast carcinoma is highly preserved in its distant metastasis. These findings suggest that metastatic capability in breast cancer is an inherent feature, and is not based on clonal selections. The results further imply that neo-adjuvant treatment given to patients based on (yet to be established) response expression profiles of their primary breast tumor might indeed prevent the outgrowth of micrometastases. Currently, the EORTC breast group is preparing a 5000-patient randomized trial to compare the efficacy of guidance of breast cancer patients for adjuvant chemotherapy based on either 'conventional' St Gallen consensus criteria or the microarray prognosis test (MINDACT trial within the EU-TRANSBIG program). The aim of the study is to confirm that the microarray test will save up to 30% of the patients from unnecessary chemotherapy and to identify 5% of them who are nowadays 'undertreated'.

S.29
Genomic profiling of breast cancer

S.30
A single nucleotide polymorphism in the HDM-2 gene regulates the p53 apoptotic response and influences the age of onset of cancers in humans: the SNP 309 HDM-2 polymorphism S12 mRNA is translated more efficiently than mRNA that starts at the first exon. The coding region of the HDM-2 protein starts in the third exon. At residue 309 in this first intron is a single nucleotide polymorphism, with 12% of people being a G/G homozygote, 40% being a G/T heterozygote and 48% of people being T/T wild-type homozygotes (the G/G genotype is lower in black Americans and the sample size is now over 300 people). We have found that the G/G genotype creates a better SP-1 transcription factor binding site, raises the level of m-RNA in unstressed cells and produces threefold to sixfold more HDM-2 protein in cells (cancer cells in culture) with the G/G genotype. This mRNA starts at the second exon, and is probably translated better in unstressed cells. After DNA damage or other stresses, P53 activity in cells with the G/G genotype is lower and the percentage of cells undergoing apoptosis is lower when compared with cells in culture with T/T genotypes. We have reproduced these observations with lymphocytes taken from human volunteers and placed in culture, with EBV-immortalized B cells in culture, with primary fibroblasts in cell culture and with cancer cell lines in culture. In 92 individuals that have donated lymphocytes we see individuals forming a distribution of apoptotic responses between 20% and 60% after gamma radiation, with individuals being quite reproducible in repeated experiments. The lower half of the distribution is heavily weighted with the G/G genotype, while the upper half of the distribution has mainly the T/T genotype. The higher HDM-2 levels in cells thus result in a lower apoptotic index in cells from these volunteers. It has become clear in recent studies that SNP 309 has a clinical impact. We have genotyped two cancer cohorts, one at MD Anderson and one in Germany, containing patients with sarcomas and breast cancers. The results have been statistically significant (P = 0.01-0.02) and clear in both cohorts, and the average age of onset of these cancers is 10-15 years earlier in people with the G/G genotype than in people with the same cancer with the T/T genotype. The interpretation is then that the probability of eliminating pre-cancerous clones of cells via a p53 mechanism is lower in people with a G/G genotype (high HDM-2 levels) and the probability of developing a cancer at an earlier time in life is higher. In addition, in patients that have a germline mutation in the p53 gene (this yields one-half of the p53 protein level in a cell) those individuals that have a G/G genotype or a G/T genotype develop multiple cancers (three, four or five cancers) over their lifetimes, while no T/T homozygotes develop that many independent cancers. Both processes are frequently disrupted in cancer cells, and each act as potent barriers to tumorigenesis. Since radiation and many chemotherapeutic agents induce apoptosis or senescence, the integrity of these programs can influence the outcome of cancer therapy. Our laboratory strives to understand how cancer genes control apoptosis and senescence in normal cells, and how mutations that disrupt these processes impact tumor development and therapy. The goal of these efforts is develop therapeutic strategies based on an understanding of drug action and cancer genotype. We currently are using genetically engineered mouse models to understand how apoptosis and senescence are controlled in tumor cells, as well as the response of tumors to conventional and targeted therapeutics. Recent work exploring the action of tumor-derived myc mutants in oncogenesis and the role of the p53 tumor suppressor network in the action of targeted therapeutics will be discussed. We previously identified mutations affecting the L2 and/or L3 domains of the TP53 gene to predict resistance to anthracycline as well as mitomycin therapy [1,2]. However, while TP53 mutations were significantly associated with therapy failure, we observed tumours resistant to therapy despite harbouring wild-type p53. We also saw responding tumours among those harbouring TP53 mutations affecting the L2 or L3 domains. Based on these assumptions, we postulated that chemoresistance could be due to failure of the 'p53 pathway' acting in concert with one, or more, redundant pathways [3]. In a recent paper we thus reported a mutation of the CHEK2 gene among one of the tumours resistant to therapy despite harbouring wild-type TP53 [4]. In addition, we are searching for redundant pathways that may compensate for the p53 mechanism. Strikingly, looking at genetic alterations associated with resistance to other drugs with respect to other malignancies, this seems to concentrate on drugs known to be involved in so-called 'family cancer syndromes', meaning genes involved either in growth arrest, apoptosis or DNA damage repair [3]. This may seem logical, as much of the damage created by chemotherapeutic drugs resemble genetic events involved in carcinogenesis. Thus, at this stage, our interest is focused on genetic pathways involving genes involved in 'family cancer syndromes'. An update of our current results will be presented. HER2-positive breast carcinomas have been shown to display an early peak of relapses in the first 4 years after surgery, especially in the node-positive subgroup. To explain this observation, growth factors released at the time of surgery were investigated. The level of growth factors of the EGF family, detected in postsurgical sera from breast carcinoma patients, was found to correlate with surgical invasiveness. Indeed, following radical mastectomy, higher levels of serum EGF-like factors were found than after conservative surgery. This implicates that the growth of tumors overexpressing HER2, activated by these growth factors, should be stimulated after invasive surgery. Two retrospective analyses of the HER2 status of primary tumors included in a randomized clinical trial addressing the issue of conservative versus invasive surgery and of radiotherapy were performed by immunohistochemistry using the standardized herceptest. Survival analysis according to surgery indicated no differences in HER2-negative cases but indicated a poorer survival for HER2-positive node-positive patients who had mastectomy in comparison with those who had conservative surgery. Furthermore, local relapses in patients who had conservative surgery without radiotherapy were found to be anticipated in the HER2-positive subset. This is a 'proof of principle' that surgery by inducing growth factor release may be detrimental for patients with HER2-positive tumors. To verify these findings, a prospective analysis of the follow-up of more than 2000 patients who have had mastectomy or conservative surgery is ongoing. Preliminary data indicate a significantly worst prognosis of HER2-positive tumors after invasive surgery, above all for tumors scoring 2+ by immunohistochemistry. Acknowledgement Partially supported by the AIRC. Nonetheless, many of these patients relapse with disease that was either intrinsically resistant to treatment or that has acquired resistance to the endocrine treatment. Laboratory studies have revealed that growth factor receptor pathways form an important route of growth signalling in both these circumstances, and there is now a series of agents available that target these pathways at different points. This provides the opportunity to utilise these agents in combination with endocrine treatment and the possibility that this may extend the effectiveness of the hormonal agents. The effective delivery of such combinations depends on a detailed knowledge of the degree to which the highly encouraging laboratory findings are translated into the clinical scenario. We have demonstrated that almost all breast ERpositive cancer shows some proliferative dependence on oestrogen, but that this is very variable. We have begun to identify in clinical samples the key genes whose expression both determines this variability and are themselves dependent on it. The development of novel models of drug development that allow the assessment of the expression of these genes, particularly within the presurgical setting, offers major opportunities to assess the potential of the various new targeted agents to be combined with endocrine therapy. Intense studies of one such inhibitor, R-Roscovitine (CYC202), including trials involving more than 100 patients, have established the potential of the class as nongenotoxic anti-cancer drugs. In some model systems the activity of this class of compound is best explained by their activity as inhibitors of transcriptional elongation, and a link between this mechanism and the induction of apoptosis has been established. The concept of cyclinspecific inhibitors as more sophisticated genetic models of target validation in this field will be discussed.

S.39
Genes, genomes, and cancer Methods We have carried out association studies to search for common variants (minor allele frequency >5-10%) that contribute to predisposition,. To date we have studied 400 SNPs in 110 genes using a two-stage study design, in which a first set of 2300 cases and controls is analysed and all SNPs with a significance value of P < 0.1 or better are then tested in a second, similar, case/control set.

Results
No individual SNP has, to date, given a P value for association (based on genotype distribution) lower than 10 -4 . A number of SNPs give P values between 10 -2 and 10 -4 , depending on the genetic model that is chosen for the analysis. Most of these are probably false positives, the consequence of multiple testing. However, comparison of the distribution of P values across the entire study set with that expected if there were no genetic effect suggests that some of these are probably true positive associations, representing low-level predisposing effects.
Conclusions A candidate gene approach is slow and relatively expensive, and has not so far yielded unequivocal positive results for any individual gene. The 'genetic architecture' of breast cancer -that is, the number and characteristics of predisposing genetic variants -is still not known. In an attempt to elucidate this and to hasten the process of gene discovery, we have initiated (with collaborators in the UK and at Perlegen Science Inc.) a genome-wide scan. Again we use a two-stage approach. In the first stage we will evaluate 266,000 SNPs in 400 breast cancer cases and 400 controls. The cases will be 'enriched' for genetic effects by choosing those with a family history, tested negative for BRCA1/2 mutation. In the second stage, ~5% of SNPs will be further evaluated in 4600 cases and controls. A final stage of evaluation for positives from the second stage, and from our earlier studies, will require analysis of a further, very large (~10,000), case/control set, which we hope to assemble through international collaboration.

S.41
Functional genomic approaches to breast cancer One of the major remaining deficits in our understanding of the human genome is that information regarding gene function is available for only one-quarter of the approximately 30,000 genes. Many of these hitherto anonymous genes are potential targets for the development of new anti-cancer drugs. It is therefore important to functionally annotate the tens of thousands of genes for which this information is currently lacking. My laboratory has developed functional genetic approaches to obtain information regarding gene function using high-throughput screens in mammalian cells. We have developed both gain-of-function genetic screens (using retroviral cDNA expression libraries) and loss-of-function genetic screens (using vectorbased RNA interference libraries) to carry out large-scale genetic screens in mammalian cells. We focus on the central growth-regulatory pathways that are most frequently deregulated in cancer.

Methods
We have designed a mammalian expression vector (pSUPER), which directs the synthesis of short hairpin transcripts (shRNAs) that are processed intracellularly into siRNA-like molecules. This vector mediates persistent inhibition of gene expression in a highly specific fashion. We have used this vector to stably suppress expression of individual members of several cancer-relevant gene families.

Results
We used a retroviral derivative of the pSUPER siRNA vector to generate a large collection of siRNA vectors that each target a single gene for suppression. In total, we constructed a set of 23,742 siRNA vectors that together target 7914 human genes for suppression by RNA interference. Furthermore, we developed a very efficient way to identify biologically active shRNA vectors in a large population of vectors, a technology that we named 'siRNA bar code screening'. We will present two applications of this technology to study major questions in breast cancer. First, we have used the RNAi library to identify genes whose suppression causes resistance to anti-hormonal therapy (tamoxifen resistance and pathologists in an integrated, mission-oriented, discovery-driven translational research environment. The unifying concept behind our experimental strategy is the use of multiple experimental paradigms for the prospective analysis of clinically relevant samples obtained from the same patient, along with the systematic integration of the biological and clinical data. Here I will describe our efforts to apply proteomics approaches to search for markers for early detection of breast cancer using the newly characterized interstitial fluids recovered from fresh tissue biopsies of both normal (NIF) and tumour (TIF) origin. The protein composition of the fluids is strikingly different to that of serum and cyst fluids, although they share some of their major components. The TIF is highly enriched in proteins that are either secreted via the classic endoplasmic reticulum/Golgi pathway, shed by membrane vesicles (membrane blebbing), or externalized by plasma membrane transporter. Hundreds of primary translation products, as well as post-translational modifications, have so far been identified using a combination of procedures that include mass spectrometry, two-dimensional gel immunoblotting, and cytokine and signalling pathway-specific antibody arrays. The workflow to biomarker discovery as well as recent developments will be discussed. In silico screening In order to make use of the exponential increase of published data on gene expression arrays, we have launched a project to acquire and make use of these data as a discovery resource. We currently have data on 5700 samples analyzed on the Affymetrix gene expression platform stored in our relational database. These samples include, for example, 64 normal tissues/cell types, 43 tumor types, many other diseases as well as functional experiments; altogether 84 million data points. We have developed methods to mine these data to identify tissue-specific and disease-specific expression patterns of transcripts, to identify gene coexpression profiles, to explore networks of gene regulation as well as methods to interpret new microarray experiments. In silico transcriptomic screening makes it possible to generate dozens of testable hypotheses for laboratory analysis based on datasets that are much larger and more extensive than any single academic laboratory can afford to generate. Analysis of gene expression profiles across hundreds of tissue and tumor types, diseases and experimental manipulations generates novel, often unexpected, insights of gene function as well as of the underlying biology and medicine. Molecular screening Large cohorts of clinical samples are now being investigated not only at the RNA level by gene expression profiling, but also at the DNA-level using comparative genomic hybridization (CGH) arrays for analysis of somatic genetic alterations or SNP arrays for studies of allelic gains and losses. There is also an emerging interest for large-scale proteomic and metabolic profiling. It will be increasingly important to integrate multiple levels of molecular profiling data to gain new insights and comprehensive views on mechanisms of cancer development. We are applying single-gene resolution oligo-CGH arrays and integrating these data with gene expression information on the same samples. The increased CGH resolution has highlighted several microdeletions as well as small amplifications, whose impact on gene expression can be substantial and highly specific. This has led to an opportunity for rapid identification of genes that may be targets of genetic alterations in cancer. As demonstrated by several recently approved drugs for cancer, such mutated genes represent attractive targets for the development of effective cancer-specific therapeutics. Functional screening using RNA interference The molecular profiling of DNA expression patterns, RNA expression patterns or protein expression patterns in patient samples is not sufficient for implicating these molecules or molecular mechanisms as therapeutic targets. It is also necessary to generate functional information on such genes and pathways. Towards this aim, we have developed a high-throughput screening system that is composed of a robotic, automated platform for the analysis of up to 20,000 functional experiments with living cells at a time using the 384-well microplate format. Cells are dispensed into culture wells, exposed to siRNAs or small molecule compounds, incubated for 1-3 days, washed, and stained with phenotype-specific markers for cell growth, cell cycle distribution or induction of apoptosis. The results are read by plate readers or cell cytometers. Functional studies with RNAi libraries (e.g. 1000-10,000 siRNAs) have implicated genes whose targeting by RNAi is lethal to specific cancer types, such as breast cancer. Integration of such functional RNAi data with gene expression and aCGH data has enabled us to identify genes that are targets of genetic alterations and whose expression is required for the maintenance of the malignant phenotype. In order to further increase the throughput of molecular analyses, we are developing methods to print tissue lysates from frozen samples in an array format. Tissue protein lysate arrays will help to automate and expand antibody-based validation of molecular targets in large cohorts of tissue specimens. Conclusion In summary, high-throughput technologies can be applied in the 'in silico', molecular, cellular and clinical screening studies, thereby substantially increasing the information and insights that can be obtained on a single platform. Integration of data from the various highthroughput platforms will facilitate a deeper, mechanistic interpretation and understanding of cancer biology, which is an essential requirement for the development of next-generation targeted therapies.  [1][2][3]. A recent pooled analysis of 53 epidemiologic studies found that the positive relationship with smoking and breast cancer was present only among those consuming alcohol, an established cause of breast cancer [4]. The purpose of this report was to examine the effect of smoking on breast cancer risk among non-drinking women in a large population-based cohort of women, many of whom started smoking as teenagers.

S.44
Methods We followed 102,098 women, aged 30-50 years, completing a mailed questionnaire at recruitment to the Norwegian-Swedish Cohort Study in 1991/1992, through December 2000. Altogether, 1240 women were diagnosed with incident, invasive breast cancer. We used Cox proportional hazard regression models to estimate the relative risk of breast cancer associated with different measures of smoking initiation, duration, and intensity adjusting for confounding variables. We conducted analyses on the non-drinking study population, and especially among non-drinking women who had smoked for at least 20 years. Results We identified 214 breast cancer cases among the 19,288 non-drinking women. Compared with never smokers, women who smoked for at least 20 years, and initiated smoking prior to their first birth (relative risk = 1.75; 95% confidence interval = 1.02-2.99), before menarche (relative risk = 1.66; 95% confidence interval = 0.79-3.49) or before age 15 (relative risk = 2.20; 95% confidence interval = 0.92-5.30), had an increased risk. In contrast, women who had smoked for at least 20 years, but started after their first birth, did not experience an increased breast cancer risk (relative risk = 0.97; 95% confidence interval = 0.44-2.12). Conclusion Our results support the notion that non-drinking women who start smoking as teenagers and who continue to smoke for at least 20 years may increase their risk of breast cancer. The human BRCA1 gene is under transcriptional control of two alternative promoters, α and β, that drive the transcription of exon 1a or exon 1b, respectively. Additionally, the 5′-UTR region, encompassing both exons 1, contains multiple putative and functional regulatory sequences. At the RNA level each exon 1 is linked, by splicing, with exon 2 that contains the translation start site. The aim of this study was to search for sequence alterations within the BRCA1 promoter and 5′-UTR in patients with breast and/or ovarian cancer, and to assess whether these sequence variants influence activity of the BRCA1 promoter region.

References
Methods The 5′ region of the BRCA1 gene, containing promoters α and β as well as exons 1a and 1b and the fragment of intron 1, was sequenced in 87 breast/ovarian cancer cases. All patients had a strong family history of breast and ovarian cancer, but were found mutationnegative in our previous search for founder mutations in BRCA1 (185delAG, 300T/G, 4153delA, 5382insC) and BRCA2 (6174delT, 9631delC). The frequency of the 2223delAAAAA deletion was assessed using allele-specific PCR amplification (ASA) in a larger group of breast/ovarian cancer patients fulfilling the aforementioned criteria. The functional significance of sequence variants within the 5′-UTR of BRCA1 was analyzed by luciferase assay. A 1.5 kb DNA region encompassing minimal BRCA1 promoter and 5′-UTR, both wild type and variant sequence, was cloned into the pGL3 vector containing the luciferase reporter gene. The luciferase activity reflected the influence of the sequence alterations on the transcriptional activity of the BRCA1 promoter and other gene regulatory regions.

Results
We found several sequence variants within the examined noncoding region of BRCA1. The frequency of the largest sequence alteration found, deletion 2223delAAAAA (according to the Acc. U37574) within exon 1b, was determined in a group of 150 patients. Three families have been identified bearing the said deletion. We also found two linked nucleotide substitutions (2642A>T, 2743T>C) in BRCA1 intron 1. The functional impact of the most frequent sequence alterations was examined in lung cancer cell line NCI-H1299 and breast cancer cell line MCF7. In the MCF7 cell line all tested variants of BRCA1 promoter/5′-UTR showed lower activity than the control wildtype sequence, while in NCI-H1299 cells the variant promoter/5′-UTR activity was higher than the control. However, observed differences of luciferase activity were not statistically significant. Conclusion Our luciferase assay showed that sequence variants detected in our study within the BRCA1 promoter/5′-UTR do not change the functional activity of the BRCA1 promoter in the experimental system that we have used, and may not be associated with an increased risk of breast and ovarian cancer. The detailed results of the analyses will be presented. From this it emerges that a number of mutations have been found repeatedly, and specific mutations are common in defined populations. This can be referred to a founder effect that in human genetics refers to the presence of genetic disorders that are either endemic to an isolated population or are very rare elsewhere. In a previous populationbased study we described the first example of a founder BRCA1 mutation in Italy, with a strong recurrence of the 5083del19 in high-risk patients all of Calabrian origin selected for family history of the disease [1]. The same mutation accounts in Canada for a significant percentage of women of Italian ancestry with breast-ovarian cancer [2]. The aim of this study was to investigate the extent to which the 5083del19 mutation contributed to breast-ovarian cancer incidence in Calabria, and to perform some genotype-phenotype correlations.

P1.04
Methods We tested 70 paraffin-embedded tissue specimens from a consecutive series of breast or ovarian cancer cases. It is estimated that our tissue archive collects more than 90% of incident cancers in the Catanzaro area. Archival tissue samples were genotyped only for the 5083del19 founder mutation. Immunohistochemical staining of several markers (i.e. ER, PR, Ki67, p53, HER2, CK5/6) as well as some pathological features (i.e. histology, grade) were evaluated. Results Seven out of the 70 samples screened for this BRCA mutation had mutations. Haplotype analysis revealed a common ancestor. All cases aged in premenopausal years. The phenotype of the 5083del19 BRCA1-associated breast tumours appears characterized by a lack of expression of hormone receptors and is generally associated with high proliferation markers and poorly differentiated aspects. All cases were ductal invasive carcinomas and showed a 'basal-like' portrait according to expression profiling studies [3]. The median age at diagnosis of the ovarian cancer carriers was 49 years and the histological type was serous adenocarcinoma. Conclusion A predominant BRCA1 gene founder mutation associated with a high risk of early onset breast cancer and ovarian cancer and unfavourable immunophenotype features has been identified and found to occur in a restricted geographical area, thereby allowing timely and cost-effective mutation screening using blood samples or archival histological material. mostly aimed at identifying changes in the coding sequences and in the donor-acceptor splice sites. Hence, mutations in the promoter and the untranslated regions, and large rearrangements, are not detected by these methods. To assess the importance of BRCA1 and BRCA2 alterations that are neglected by standard screening methods, we monitored germline rearrangements in these genes using 'multiplex ligation-dependent probe amplification' technology [1]. One hundred and seventy-nine Norwegian breast and ovarian cancer families were screened for rearrangements in BRCA1 while 97 families were tested for aberrations in BRCA2. Whereas no rearrangements were detected in BRCA2, four distinct deletions were found in BRCA1. Those deletions originating by Alu-mediated homologous recombination include: exons 1-13, exons 3-16, exons 8-13 and exon 23, respectively. The large 23.8kb deletion excluding exons 8-13 in BRCA1 has been found both in the French and British breast cancer population [2][3][4]. The deletions of exons 1-13, exons 3-16 and exon 23 have not been previously reported. Background Germline mutations in the highly penetrant cancer susceptibility genes BRCA1 and BRCA2 cause genetic predisposition to breast and ovarian cancers. Molecular genetic testing of pathogenic mutations in these two genes is an effective method for breast cancer risk prediction. Genetic counselling and testing has been provided to high-risk women in our institute since 1997. Until now 589 probands (580 women and nine men) with breast/ovarian cancer have been tested for BRCA1/2 germline mutations. Methods Genetic counselling was performed by a medical geneticist in our institute or in other genetic centres of the Czech Republic. Informed consent was signed in all tested individuals. For genetic testing the non-radioactive protein truncation test of exon 11 of BRCA1 and exons 10 and 11 of BRCA2 were used, followed by heteroduplex analysis of the remaining exons with their splice sites and by sequencing. The frequency of unknown variants was tested in a control group of healthy women older than 60 years without a positive family history of breast/ovarian cancer.

Results
The pathogenic mutation was found in 179 of 589 tested probands (30%), 106 in the BRCA1 gene and 73 in the BRCA2 gene.  [1,2] and the other six mutations have not yet been published. A spectrum of 28 variants (13 in BRCA1 and 15 in BRCA2) with unknown clinical significance was found. The most frequent variant was c.5075G>A, p.M1652I, in the BRCA1 gene (in 22 cases), but the frequency of this variant in a control population was 6.3%. The missense mutation in the BRCA1 gene c.172T>A, p.M18K, was detected three times in HBOC and HBC families, and is very probably pathogenic (localised in a highly conserved structure of the Ring finger domain); it was not detected in a control group. Another two variants were detected in a control population with frequency 1.6%: IVS24/c.9485-16T>C and c.10323delCins11 in the BRCA2 gene. By predictive testing, positive carrier status was disclosed in 163 individuals of 395 tested. The clinical follow-up was offered to all individuals at high risk. The high-risk clinic at MMCI is following 62 healthy carriers, 139 healthy people at risk, 47 patients-carriers and 29 high-risk patients with negative testing result. From these, 14 patients and three healthy carriers underwent prophylactic mastectomy, 26 patients and four healthy carriers underwent prophylactic adnexectomy. According to the questionnaire for diagnoses of depression CES-D instituted before testing, 51% of women did not report any signs of depression, 27% reported very mild, 14% middle and 8% severe signs of depression. A psychologist in our institute follows 27 women. No severe psychological complications of testing have so far been reported. Conclusion Overall, pathogenic mutation was disclosed in 30% of all our tested individuals with familial or sporadic breast (or ovarian) cancer. At least one ovarian cancer in a high-risk family increases twice the likelihood of mutation detection. Fourteen novel mutations and a spectrum of unknown variants were detected. Prophylactic mastectomy is not so frequently accepted by mutation carriers as in Western Europe, but there is an increasing tendency especially in young women. The quality of preventive care in other institutions and the psychosocial influence of genetic testing are now being investigated.  [1]. To date, several small studies have suggested a worse outcome of survival in BRCA1 and BRCA2 carriers. However, the evidence is inconsistent and most studies were subject to different types of bias. An increased risk for contralateral breast cancer, especially in interaction with radiotherapy, for CHEK2*1100delC carriers has been shown [2]. Only one study has so far evaluated the impact of CHEK2*1100delC on survival, showing a worse disease-free survival compared with control breast cancer patients [3]. Our aim is to evaluate breast cancer survival and to determine risk estimations for the development of contralateral breast or ovarian cancer (as a second primary), as well as to evaluate tumour characteristics, in BRCA1/2 and CHEK2*1100delC carriers in an unselected, non-family based, retrospective cohort of breast cancer patients diagnosed under age 50. The cohort to be evaluated will include approximately 5000 patients, treated in several Dutch hospitals between 1973 and 1995. Tissue blocks from these patients are being obtained and, after coding, about 70 BRCA1/2 founder and recurrent mutations, representing approximately 72% of the Dutch BRCA1/2 mutations, and the CHEK*1100delC mutation are being determined. Data for 1700 patients from the Netherlands Cancer Institute and the Leiden University Medical Center are being completed. We have so far found, in 1255 samples, 4.1% BRCA1/2 carriers (41 BRCA1 and 11 BRCA2 mutations) and 3.8% CHEK2*1100delC, with no overlap among these groups. An interim analysis showed that BRCA1 tumours seem to have less favourable prognostic characteristics while BRCA carriers have an OR of 3 for contralateral breast cancer compared with the non-BRCA carriers. Genetic determinants of tumour characteristics, risk for contralateral breast cancer and survival of CHEK2*1100delC carriers will be presented. Numerous studies have shown that exposure to ionizing radiation is a risk factor for BC. Because of the role of the BRCA proteins in DNA repair it is plausible that women who carry mutations in these genes might be more sensitive to ionizing radiation than women in the general population. We therefore determined the role of low-dose ionizing radiation in carriers of a BRCA1 or BRCA2 mutation. Methods A retrospective cohort study of 1601 female BRCA1/2 carriers, of whom 879 were affected with BC at the time of interview, was performed. Exposure data were analyzed using a weighted Cox proportional hazards model. We assessed the relative risk of BC as a function of exposure to chest X-rays as assessed by questionnaire.
Conclusions The observed patterns of risk are consistent with those found in previous studies of radiation and BC, but the extent of the risk increase in BRCA carriers appears to exceed several-fold that observed for other radiation-exposed cohorts. The results of this study have important implications for the use of X-ray imaging in BRCA1/2 carriers, particularly before age 20. We identified a cohort of 148 female BRCA1 or BRCA2 mutation carriers (115 and 33, respectively) who previously were treated for unilateral invasive breast cancer stage I-IIIa. Seventy-nine women underwent a CPM, while the other women remained under intensive surveillance. The mean follow-up was 3.5 years and started at the time of CPM or at the date of mutation testing, whichever came last (i.e. on average, 5 years after diagnosis of the first breast cancer). One woman developed an invasive contralateral primary breast cancer after CPM, whereas six were observed in the surveillance group (P < 0.001). CPM reduced the risk of contralateral breast cancer by 91%, independent of the effect of bilateral prophylactic oophorectomy (BPO). At 5-year follow-up, overall survival was 94% for the CPM group versus 77% for the surveillance group (P = 0.03). Unexpectedly, this difference in survival was mostly due to higher mortality related with the first breast cancer and ovarian cancer in the surveillance group. After adjustment for BPO in a multivariate Cox analysis, the CPM effect on overall survival was no longer significant. Our data show that CPM markedly reduces the risk of contralateral breast cancer among BRCA1 or BRCA2 mutation carriers with a history of breast cancer. Longer follow-up is needed to study the impact of CPM on contralateral breast cancer specific survival. The choice for CPM is highly correlated with that for BPO while only BPO so far leads to a significant improvement in overall survival. The marked reduction of the risk of breast cancer following a prophylactic oophorectomy illustrates that endogenous hormones play an important role in the etiology of breast cancer among BRCA1/2 mutation carriers, as they do in the general population. In the general population the use of oral contraceptives has been associated with a slightly increased risk. Little is so far known about the safety of oral contraceptives among BRCA1/2 carriers, who have much higher premenopausal background rates of breast cancer. Methods A retrospective cohort study was performed using an international cohort of 1601 BRCA1/2 mutation carriers. A timedependent proportional hazard Cox regression was used, stratified for birth cohort, gene, country of residence and relatedness. All analyses were adjusted for prophylactic oophorectomy and number of full-term pregnancies. To reduce possible testing bias, the analyses were weighted to achieve the rate of breast cancer within the cohort as a priori estimated for BRCA1/2 mutation carriers.

MA Rookus 1 , RM Brohet 1 , N Andrieu 2 , AC Antoniou 3 , J Chang-Claude 4 , DF Easton 3 , S Peock 3 , C Noguès 2 , FE van Leeuwen 1 , DE Goldgar 5 , on behalf of the IBCCS
Results We found a slightly increased risk of breast cancer for BRCA/12 mutation carriers who ever used oral contraceptives, with an adjusted hazard ratio of 1.47 (95% confidence interval = 1.16-1.87). The risk increase did not vary according to various aspects of oral contraceptive use, such as time since stopping, duration of use, age at start, and calendar year at start. In addition, the risk increase was similar for BRCA1 and BRCA2 mutation carriers. Conclusion Use of oral contraceptives seems to be associated with a slightly increased relative risk of breast cancer among BRCA1/2 mutation carriers, comparable with the general population. However, due to the high background rates of breast cancer among BRCA1/2 carriers, oral contraceptive use may result in a considerable absolute excess risk of breast cancer, if the association is causal.

P1.11
Genome-wide scanning for linkage in 56 Dutch breast cancer families selected for a minimal probability of being due to BRCA1 or BRCA2 the disease is associated with known risk factors including highpenetrance genes and lifestyle factors. Candidate low-penetrance genes are involved in a variety of pathways; for example, DNA damage by free radicals. The enzymes involved in this mechanism are N-acetyltransferase2 (NAT2) and manganese superoxide dismutase (MnSOD). NAT2 catalyzes acetylation of aromatic amines and hydrasines and forms of free radicals, and MnSOD catalyzes their dismutation. Genes TGFBR1 and RGS19IP1 encode proteins that take part in transforming growth factor beta signaling, mainly resulting in inhibition of cell proliferation. The first exon of TGFBR1 contains a polymorphic GCG repeat; (GCG) 6 was previously reported as the tumor susceptibility allele associated with some types of cancer. The 5′-untranslated region of RGS19IP1contains a CGG repeat, the polymorphism of which had not been studied before.  [1]. Given that no additional high-penetrance susceptibility genes have been found to be involved in breast cancer, it has been proposed that different genetic backgrounds due to the combination of low-penetrance genes (polygenic mechanism) could explain the remaining familial breast cancer risk [2]. Hence there is much interest in the search for lowpenetrance gene/variants for breast cancer, which exist with high prevalence in the general population. Single nucleotide polymorphisms (SNPs) have emerged as genetic markers of choice because of their high density and relatively even distribution in the human genomes [3,4], and are being using for fine mapping of disease loci and for candidate gene association studies. Approximately 10 million SNPs have been identified across the human genome and new technologies are available today for high-throughput genotyping.
In this study we used the SNPlex™ (Applied Biosystems, Foster City, CA, USA) high-throughput genotyping platform, which allows the study of up to 48 SNPs simultaneously, to study 984 SNPs of 92 cancerrelated genes, in a total of 480 female cases of breast cancer and 480 female controls. Gene selection was made on the basis of their involvement in different cancer pathways and genes: DNA reparation, cell cycle control, BRCA1-associated binding proteins, and so on. SNP selection was performed using an indirect approach (1 SNP/10 kb) and based on the individual allele frequency (FAM ≤10%) in the European population, using public and private SNP databases and bioinformatics tools (dbSNP, HapMap, Sequenom Real SNP, PUPASNPI Ensembl, and Celera, among others).
To date, 415 SNPs from 44 genes have been genotyped in nine SNPlex pools. A case-control analysis was conducted for the 318 remaining SNPs. Preliminary results showed association in 24 SNPs from 12 candidate genes (P < 0.05). We will present the analysis of the remaining 48 genes at the time of the congress.

S23
populations. To what extent does this knowledge of the LD domains affect previous findings from pharmacogenetic studies of single candidate SNPs? Here we select candidate genes as part of a given functional pathway, and report their chromosomal localization and extent of LD. Methods A total of 193 breast cancer patients have been genotyped for 725 SNPs in 206 genes selected through the candidate gene approach. Two hundred and fifty-three of the SNPs have also been genotyped in a cohort of 109 healthy Norwegian women. SNPs that had a discovery rate lower than 75% were excluded. Hardy-Weinberg equilibrium was calculated prior to further statistical analysis of LD. LD estimations were made using PHASE, a program that implements methods for calculating haplotypes from population genotype data [1].

Results
The 725 SNPs were divided between 206 different genes with 1-20 SNPs per gene distributed on all chromosomes. Initially, SNPs were grouped in clusters containing a minimum of three SNPs with no more than 100 kb between neighbouring SNPs. Based on the PHASE output, D′ and the P value of the Fisher's exact test were calculated. We observed strong LD in 74 genes, and 10 genes were split into more than one LD domain. Furthermore, neighbouring clusters of genes were studied for common LD. Genotype frequencies and the extent of LD were compared in a case-control study when possible. Glutathione-S-transferase P1 (GSTP1) is involved in thiol-mediated detoxification and breakdown of reactive oxygen species created by anticancer drug exposure. GSTP1 is also an inhibitor of c-Jun Nterminal kinase 1, a kinase involved in stress response, apoptosis and cellular proliferation. Hypermethylation of the GSTP1 promoter has been associated with gene silencing in prostate cancer, kidney cancer, and breast cancer, among others. Although frequently described, the mechanism underlying promoter hypermethylation of the GSTP1 gene is poorly understood. It has been reported that an ATAAA repeat of the GSTP1 promoter separates methylated from unmethylated CpGs in normal prostate tissue [1]. These separate methylation domains are lost in prostate cancer, and methylation extends throughout the whole promoter region. It has been proposed that hypermethylation of GSTP1 requires a combination of gene silencing and random seeds of methylation in prostate cancer cells, and that these combinatorial effects lead to histone deacetylation and subsequent chromatin remodeling [2]. To further elucidate the mechanisms underlying the hypermethylation of the GSTP1 promoter, we genotyped the (ATAAA) repeat and the linked SNPs in positions -354, -288, -287 and -282 in the GSTP1 promoter and we performed methylation analysis using mass spectrometry in tumor DNA from 82 breast cancer patients. The role of the different allelic variants on methylation status of the GSTP1 promoter and expression levels was assessed. We quantitatively determined the methylation status of six CpGs spanning the transcription start site of the GSTP1 promoter: -22, +8, +14, +38, +47 and +55. The average percentage methylation for each individual CpG for the 82 tumor samples analyzed was 16.9%, 30.3%, 18.2%, 21.2%, 18.6% and 8.1%, respectively. The average percentage methylation for all CpGs in all tumor samples was 19%. There was a correlation between the degree of methylation of the individual CpGs and their neighboring CpGs (P < 0.001). When correlating the extent of methylation to the mRNA levels previously assessed by whole genome gene-expression profiling of the same tumors, a significant inverse correlation was observed (P < 0.01). The methylation status of the three CpGs closest to the transcriptional start site was more highly associated with the level of GSTP1 mRNA expression than the CpGs further downstream of the +1 site. Furthermore, we observed differences in the degree of GSTP1 promoter methylation between the different tumor subclasses defined by whole-genome microarray analysis [3]. The methylation of the GSTP1 promoter was significantly lower in the basal subtype compared with the luminal subtype, which corresponded to elevated GSTP1 mRNA levels in the basal subtypes [4]. We further analyzed the impact of the most frequent haplotype structure of the GSTP1 promoter in relation to the extent of methylation, and a correlation was observed (P = 0.003) suggesting that haplotype structures can affect de novo methylation of adjacent sequences.   [1]. One member of this pathway is CRK (v-crk sarcoma virus CT10 oncogene homolog), which is tyrosine phosphorylated upon IGF-1 stimulation. CRK has been found to be overexpressed in various human tumor tissues and cell lines [2]. However, little is known about alterations in the genomic sequence of this gene.

P1.19 Alpha-1 antitrypsin genotypes in breast cancer patients Z El-Akawi, M Nusier, DH Sawalha
Methods We sequenced the promoter and the coding region of the CRK gene in a small sample set of 23 breast cancer samples. We confirmed a C to A polymorphism at nucleotide position 49 with a synonymous amino acid change at Arg17, which was in nearly 100% linkage to the promoter polymorphism C-289A, and we identified a novel polymorphic duplication of 22 bp in the promoter region. In the further analyses we used a TaqMan allelic discrimination assay for the Arg17 polymorphism and a fluorescent fragment analysis to detect the duplication in a sample set of 352 Polish familial breast cancer cases and 485 matched controls. We determined the genotype and haplotype frequencies and calculated the odds ratios with 95% confidence intervals.
Results We did not observe any differences in the allele or genotype frequencies between the cases and controls for the duplication polymorphism. For the Arg17 polymorphism, the allele frequency of the A allele was slightly decreased among the cases compared with the controls (52.3% vs 56.0%, respectively), but the difference was not statistically significant. In the haplotype analysis, we observed a protective effect for the carriers of the Arg17 A and the duplication alleles (odds ratio = 0.17, 95% confidence interval = 0.03-0.76, P = 0.007).
Conclusions CRK is a member of the GH1/IGF-1 pathway, whose members are often found to be overexpressed in human tumors. The, to our knowledge, novel 22 bp duplication in the promoter region results in multiplication of various putative transcription factor binding sites. This may lead to an altered expression of the CRK gene. In combination with the Arg17 A allele it showed a protective effect. The Arg17 polymorphism was in nearly 100% linkage with a polymorphism in the promoter, which also might have an effect on transcription. However, a functional analysis is needed to investigate the effect of these polymorphisms on the expression.  [2]. ZBTB7 is therefore a good candidate for a breast cancer low-penetrance gene. We aim to determine whether common polymorphisms (frequency ≥10%) in the ZBTB7 gene are associated with breast cancer risk in Spanish patients. A set of 22 validated binary SNP polymorphisms were selected from public databases (e.g. HapMap) and private databases (e.g. Celera and RealSNP) according to criteria of coverage (average 1 SNP/1.7 kb). These SNPs cover the flanking regions (10,000 bp), including the promoter region, introns, and coding nonsynonymous SNPs. About 550 cases and 550 controls have been genotyped. Genotyping was performed using the MassARRAY SNP genotyping system (Sequenom Inc., San Diego, CA, USA). To our knowledge, this is the first time that the ZBTB7 gene has been analyzed in breast cancer patients. The results will be presented at the time of the congress and will be interpreted in the light of the worldwide population study shown in another presentation from the same authors. The present high-density SNP mapping study will facilitate a map of specific population variation and patterns of linkage disequilibrium at the ZBTB7 region in different human populations, and will facilitate the adequate selection of a highly efficient set of tag SNPs that will capture the bulk of the (potentially pathogenic) variation. We have found that patterns of LD and haplotype diversity at the ZBTB7 gene vary considerably among different populations. Thus, sub-Saharan African populations showed higher levels of haplotype diversity and shorter blocks, while non-Africans showed a higher level of LD and lower haplotype diversity, as expected according to population history. According to these patterns, we will discuss the efficiency of these LD patterns and tagSNPs to capture candidate SNPs at the ZBTB7 gene in tumor association studies. Carcinogenesis is a chronic disease process underlying the clonal evolution of cells progressing to the point of uncontrolled growth, metastatic potential, and molecular heterogeneity. By convention, chemoprevention drugs are developed from a molecular perspective with the goal of interrupting carcinogenesis before the occurrence of invasive lesions or extreme heterogeneity. The most successful demonstration of cancer chemoprevention to date has been an overall 49% reduction of invasive breast cancer, with a similar reduction in premalignant lesions by tamoxifen in the Breast Cancer Prevention Trial (BCPT) [1]. It is noteworthy that the main effect of tamoxifen is likely to be mediated through the estrogen receptor (ER), as reflected by a 70% reduction of lesions that are ER-positive in contrast to little or no effect on the incidence of ER-negative lesions. For interventions that are related to ER as a target, a number of drug development issues remain to be addressed including dose, schedule, and the comparative net clinical benefit of various selective ER modulators versus pure antiestrogens, aromatase inhibitors, and combinations. In the BCPT there were 13,388 participants, and there were more than 22,000 in the STAR trial. The large sample size that may be needed for a randomized clinical trial to observe a prevention effect severely limits the opportunity to explore a multiplicity of important questions in clinical chemoprevention. Molecular studies have been helpful in classifying breast cancers according to categories of response to intervention. For instance, cytogenetic studies combined with molecular profiling suggest that ERfocused interventions are likely to address a particular subset of tumors arising from the luminal cell population [2]. As tumor subsets become better characterized, the need for additional prevention studies can be anticipated to address larger subsets (e.g. a combination of drugs for overlap) versus smaller subsets of individuals at risk. In order to reduce the sample size of future prevention trials, new molecular approaches are needed. One strategy would be to use non-invasive molecular tests to identify individuals at increased breast cancer risk so that populations for prevention trials could be further enriched according to that risk. Of the approaches currently under investigation, proteomic studies theoretically offer an opportunity to improve risk identification. Investigators who are performing proteomic studies for early detection are encouraged to expand their investigations to see whether it is possible to delineate according to ER status and between non-invasive conditions such as hyperplasia and DCIS versus invasive cancer. Another strategy for increasing the efficiency of breast cancer prevention trials is the validation of intermediate endpoint biomarkers [3] to secure validated intermediate endpoint biomarkers (VIEBs). If molecular entities in serum could be identified on the basis that they are predictably correlated with the future development of breast cancer, then a reduction in the VIEB level could serve as evidence of a preventive effect. Early work in this area suggests that nucleic acids in serum might be used to identify individuals with premalignant lesions [4]. Clinical correlation is needed for VIEBs and other molecular indicators of risk so that targets in addition to the ER can also be more efficiently studied. Targets of interest for breast cancer prevention include the EGFR family, RAR/RXR and mediators of inflammation or oxidative damage.

P2.01
Gene expression profiling in whole-blood samples from postmenopausal women exposed to hormone replacement therapy Results Genes identified by t test with P < 0.03 (n = 253) were used to build a classifier using the nearest shrunken centroids method. Results did not reveal any distinct gene list that predicts accurately HRT exposure (error rate = 0.42). We performed a new analysis including, among HRT users, only women who were using continuous combined treatment (ethinylestradiol and norethisterone acetate). The performance of the classifier (i.e. 98 genes) improved (error rate = 0.25). The specificity (78.7%) was slightly better than the sensitivity (68.0%). We then tested the significant changes in a single gene by different methods like the t test, significance analysis of microarrays controlling for the false discovery rate and Bayesian ANOVA analysis, which balance the false discovery rate and the true positive rate. Only few expression alterations of minor magnitude caused by HRT could be detected in whole-blood total RNA. The weak signals of exposureinduced changes in whole blood made it very difficult, or even impossible, to identify single genes statistically significant with the background of thousands of individuals genes tested simultaneously. Instead, we focused on identifying significant pattern changes of biologic process in genes identified from the t test, using the annotations defined by the Gene Ontology Consortium.
Conclusions Mixed cell types in whole blood made it more difficult to observe differences in gene expression profiles. According to the little amplitude of expression alterations observed in whole blood, large sample sizes are needed to conduct global expression profiling. Although one gene change may be small and difficult to detect accurately in a significant test, significant enrichments in the biologic process of genes with small changes after HRT use have been assessable.

P2.02
Effects of oestrogen on gene expression in the epithelium and stroma of the normal human breast Previous expression profiles of the response to E2 were performed on tumour cell lines, in the absence of stroma [1][2][3].
Methods We investigated gene expression in normal human breast tissue (removed from benign lesions in premenopausal women) transplanted into 9-week-old to 10-week-old athymic nude (Balb/c/nu/nu) mice. Transplantation was performed in order to obviate the potential effects of the phase of the menstrual cycle. After 2 weeks, when proliferation and progesterone receptor (PR) expression are minimal, the mice were treated with E2 using a 2 mg slow-release pellet for 1 week, which we have previously shown to be optimal for inducing proliferation and PR [4]. At completion of treatment, three tissue samples were generated from each of the six original normal breast tissue samples -two of which were untreated and one of which had been treated with oestrogen. RNA was isolated from each of these samples, then labelled and hybridised to Affymetrix HG-U133A (human) chips on which 22,283 genes are represented. RMA and MAS5 normalisation methods were utilised with bioconductor analysis software.
Results Oestrogen treatment was found to be the major source of variation in gene expression. Our study shows that known E2responsive genes such as trefoil factor 1 (pS2) and amphiregulin are also differentially expressed due to oestrogen treatment of normal breast tissue. In addition, many of the genes that showed the greatest responses to E2 have previously been suggested as independent breast cancer prognostic or diagnostic markers (including mammaglobin, prolactin-inducing peptide and keratin 19). Conclusion We report the first global gene expression study to look at the effects of oestrogen on the epithelium and stroma of normal human breast tissue, which may give clues to the paracrine action of oestrogen in proliferation. These data form the basis for efforts towards the detection of early gene expression changes leading to breast cancer development. While the DNA repair function of BRCA1 clearly contributes to overall genetic stability, loss of BRCA1 function in DNA repair, a process that is universally important to all cell types in both genders, cannot easily explain why BRCA1 mutations predominantly affect breast and ovary, two major estrogen-responsive tissues in women. We hypothesize that BRCA1 confers tissue-specific tumor suppression by directly modulating expression of a set of breast-cancer-related genes.

Methods and results
Aromatase, which catalyzes conversion from androgen to estrogen, is a rate-limiting enzyme in estrogen biosynthesis and is a key player in breast cancer development.
Aromatase is specifically expressed in ovarian granulosa cells and several peripheral tissues via the actions of multiple tissue-specific promoters. In premenopausal women, expression of aromatase in ovarian granulosa cells dictates levels of circulating estrogen. In many postmenopausal breast cancer and endometrial malignancy tissues, the in situ aromatase production is significantly elevated. Here, we show that the cyclic AMP-dependent expression of aromatase is inversely correlated with the protein level of BRCA1 in human ovarian granulosa cells. Importantly, transient knockdown of BRCA1 enhances the expression of aromatase in both ovarian granulosa cells and cells from breast tissue. Chromatin immunoprecipitation indicates that endogenous BRCA1 is preferentially associated with a tissue-specific transcriptional promoter of the aromatase gene. Thus, BRCA1 negatively controls estrogen biosynthesis by repressing aromatase expression, providing a novel and logical molecular explanation for the tissue-specificity conundrum in BRCA1 biology.

Conclusion and significance
The notion that BRCA1 confers tissuespecific tumor suppression in an endocrine fashion is also consistent with results from recent clinical and animal studies. In particular, prophylactic oophorectomy reduces the risk of both breast cancer and ovarian cancer in carriers of BRCA1 mutations [1,2], suggesting that circulating estrogen in premenopausal women may contribute to the development of BRCA1-associated tumors. Moreover, tissue-specific knockout of Brca1 in mouse ovarian granulosa cells induces tumors in the ovaries and uterine horns. Importantly, these tumors still carry wildtype alleles of Brca1 [3]. Given that aromatase inhibitors are one of the most efficacious endocrine agents in breast cancer treatment, the link between BRCA1 and aromatase expression may lead to novel therapies for both prevention and treatment of breast cancer.  [1]. The ability to accurately predict tamoxifen treatment outcome would therefore significantly advance the management of breast cancer. The aim of the current project is to identify a gene expression profile associated with tamoxifen resistance using microarray analysis. Patients and methods To identify gene expression patterns that might predict response to tamoxifen, 90 breast cancer patients were selected for whom fresh frozen tissue was available. All these patients had received surgery with or without radiotherapy for primary breast cancer, while none had received adjuvant systemic treatment. All patients developed metastatic disease and were subsequently treated with tamoxifen. Response was mainly determined radiographically [2]. Approximately 40% of these patients had CR/PR or clinical benefit during less than 6 months and are defined as tamoxifen resistant, whereas the remaining 60% are defined as tamoxifen sensitive (CR/PR or clinical benefit for more than 6 months). Of the latter group, 17 out of 54 even showed no progression for at least 2 years. Gene expression profiling was performed using 44K oligo microarrays. Data analysis is ongoing and results will be presented at the MBBC symposium. Future directions A validation set will be analyzed to confirm our initial findings. Furthermore, we will test whether this profile can also be used in the adjuvant setting. In addition, we will evaluate the combination of microarray analysis and targets identified by RNAi screens in vitro in determining diagnostic tools for prediction of therapy outcome.  phosphorylation. However, a combination of the two inhibitors resulted in almost complete loss of phosphorylation. These data were confirmed using siRNA technology to suppress MAPK and AKT expression. Taken together these data suggest that, in this setting, the ER functions via a non-genomic mechanism, associating with ERBB2 and PI3K at the cell membrane leading to activation of both p90 RSK and AKT. This in turn leads to phosphorylation of ERα ser 167 , ultimately regulating cell growth via genomic mechanisms. Although several of these complexes have previously been postulated, to our knowledge this is the first demonstration of this phenomenon in a tamoxifenresistant cell line. Convergence between these pathways may lead to a synergistic feedforward circuit, resulting in a stronger or more sustained proliferative response in breast cancer cells [1]. Another signaling pathway involved in the modulation of breast cancer cell proliferation is mediated by the receptor for sex hormone-binding globulin (SHBG-R). SHBG has been found to function as part of a novel steroid signaling system that is independent of the 'classical pathway' for intracellular steroid receptors. This 'alternative pathway' (Fig. 1) involves the activation of membrane SHBG-Rs by SHBG and E2. It has been demonstrated that SHBG, through cAMP and protein kinase A (PKA), can inhibit the proliferative effect of E2 on breast cancer cells [2][3][4]. Its role in E2dependent cancer cell proliferation may in the longer term be exploited for therapeutic strategies. The pathways involved have not yet been elucidated. Indeed, the receptor has not yet been identified, although it is known to exist from binding kinetics studies. Methods We are applying proteomics techniques to identify the plasma membrane SHBG-R, and to elucidate the key signaling proteins involved in pathway(s) mediated by SHBG/SHBG-R binding.  Model for SHBG-mediated signaling. The plasma membrane binding site for SHBG is preferentially expressed in ER+ cells with a reduced proliferative index [2][3][4]. SHBG-R only binds steroid-free or unliganded SHBG. Following binding of unliganded SHBG to SHBG-R on the cell membrane, E2 binds to the SHBG-SHBG-R complex, thereby activating it. Activation induces the synthesis of cAMP, which, in turn, triggers downstream signaling via PKA [2][3][4]. The biological outcome of this signaling pathway in cells of the human breast carcinoma cell line MCF-7 is decreased E2-mediated cell growth.

Figure 2
Identification of key signaling components of SHBG-mediated pathway(s). A flow-chart of the experimental design and method (involving comparative proteomic analyses) for identification of key regulated signaling components in SHBG-mediated pathways in breast cancer cells. Each cell type will be exposed to SHBG (vs not exposed to SHBG), and to human estradiol (E2) (vs not exposed to E2). Differential analysis of proteomes for the different cell lines/treatments will then be carried out. Differences in expression/post-translational modifications of specific proteins will determine what proteins are of interest. End effects will be examined as well as the results of time-dependent experiments. The protective effect of an early first full-term pregnancy in relation to breast cancer risk is well established, but the molecular and cell-specific changes in the human mammary gland remain unclear.

P3.01 Gene expression profiling to identify parity-induced changes in the human mammary gland I Verlinden, N Güngör, J Janssens, L Michiels
The search for these early predictive markers for breast cancer risk is of high priority because these markers could identify women at increased risk and could monitor preventive strategies. Methods To identify the molecular changes associated with pregnancy-induced differentiation, we analysed the global gene expression profiles of normal mammary tissues from both a parous and a nulliparous woman, using serial analysis of gene expression (SAGE) [1]. This approach allowed us to identify sets of genes, known and unknown, that are differentially expressed in parous versus agematched nulliparous mammary gland tissues. Pair-wise comparisons of each tag were carried out in both SAGE libraries.

Results
The normal mammary gland of a multiparous woman is characterised by several known differentiation markers such as casein κ, casein β, keratin 14, CCAAT/enhancer binding protein β and δ, and adipsin. Candidate genes involved in cyto-architectural remodeling and growth inhibition with a potential role in pregnancy-induced protection against breast cancer were also observed. Several of the genes that are highly expressed in the nulliparous mammary gland and that are lost after pregnancy encode for growthpromoting, cytoskeletal and extracellular matrix proteins. One of these genes, the small breast epithelial mucin, is almost completely downregulated upon first full-term pregnancy but is known to be expressed in more than 90% of invasive ductal carcinomas [2]. A subset of these differentially expressed genes was selected for realtime PCR validation experiments. Small-scale real-time PCR experiments revealed certain trends showing the upregulation of the genes BPAG1, VIM, DCN, RARRES2, EGR1 and ADH1B and the downregulation of SPARC and SBEM in the mammary gland after pregnancy, but nevertheless individual differences in expression levels were observed. These differences could be due to fluctuations in hormone levels at the time of surgical resection. However, these individual differences also may reflect the differential risk for breast cancer for each of the patients (nulliparous and parous) involved. Conclusions This study demonstrates that the SAGE technique is useful to detect alterations in gene expression, induced by an early first full-term pregnancy, in the human mammary gland. Our results show that pregnancy induces permanent gene expression modifications in the breast, which can be potentially valuable in the search for predictive markers for breast cancer. Ongoing additional research on these genes in a larger patient population will possibly lead to the development of early markers for breast cancer.  [1]. These cells contain hypermethylated and silenced p16(INK4a) (p16) promoters and overexpress COX-2. We found evidence that cells with these characteristics exist in disease-free women in morphologically normal tissue [2]. Furthermore, these distinguishing characteristics have also been found in DCIS lesions, indicating that these cells are very relevant to the carcinogenic process. In order to investigate the molecular mechanisms required for these cells to progress to a malignant phenotype, we examined the effect of oncogenic stress on the transformation of vHMEC by introducing constitutively active Ha-Ras V12 into these cells. Consistent with the idea that vHMEC are already engaged in the transformation process, upon exposure to oncogenic stress vHMEC failed to undergo a proliferative arrest as seen in normal fibroblasts or normal epithelium. We have used this model system to examine the early events that control expression of tumorigenic phenotypes in these cells. We find that critical interactions between stromal cells and initiated epithelial cells are necessary for the manifestation of specific tumorigenic phenotypes such as epithelial-to-mesenchymal transition.  Background Human breast luminal epithelium and the stromal compartment are separated by a layer of myoepithelial (ME) cells and basement membrane, whose disruption is required for tumor invasion. Basement membrane degradation has been ascribed largely to an overproduction of proteases by the tumor and stromal cells. However, the causes and phenotypes of ME cell layer disruption and its contribution to the initiation of tumor invasion remain to be further explored.

P3.05 Ex vivo isolation of adult stem cells from normal and tumour mouse mammary parenchyma G Bonizzi, A Cicalese, B Giulini, S Pesce, A Gobbi, S Pece
Methods Human ductal carcinoma in situ samples with ME disruption and microinvasion were sectioned and immunohistochemically stained with Ki-67, estrogen receptor (ER), leukocyte common antigen, and other biochemical markers.
Results A subset of pre-invasive and micro-invasive tumors contained focal disruptions in their ME cell layer. ER-negative cell clusters overlying a focally disrupted ME cell layer and the basement membrane showed a significantly higher proliferation rate than adjacent cells within the same duct. These disruptions were associated with histochemical and genetic alterations in the overlying tumor cells, including the loss of ER expression, a higher frequency of loss of heterozygosity, and a higher expression of cell cycle, angiogenesis, and invasion-related genes. Focal ME layer disruptions were, in general, associated with a higher rate of epithelial proliferation and leukocyte infiltration; however, a small fraction of these ER-negative cells lacked proliferation and differentiation markers resembling dormant cancer stem cells.
Conclusions We propose a novel hypothesis that a localized death of ME cells and immunoreactions that accompany an external environmental insult or internal genetic alterations are triggering factors for ME layer disruptions, basement membrane degradation, and subsequent tumor progression and invasion. Inflammation may contribute to the death of focal ME cells. Putative dormant cancer stem cells may be partially responsible for tumor drug resistance and recurrence. The aim in breast cancer treatment is to provide the correct diagnosis with minimal delay that makes it possible to immediately plan further treatment with the patient. In addition, the imprint method can be used to make a diagnosis on material that will be snap-frozen for future research purposes. Method Imprints are made by gently pushing the core biopsy to a coated glass slide, and then air-drying and staining with Diff-Quick. The diagnosis is usually made within 5 min.

P3.07 Imprint as a reliable diagnostic tool in breast cancer
Results Of 51 imprints, 15 were diagnosed as carcinoma. Histology confirmed carcinoma in 13. The two apparent false positives turned out to be cancer on further investigation. Two of the imprints were given a benign diagnosis; both turned out to be invasive lobular carcinoma. The rest of the imprints that were given a benign diagnosis were all confirmed as benign on histology.
Conclusions There was no true false positive diagnosis, but there were two false negatives, both invasive lobular carcinoma. Imprint of core biopsies is a reliable cytological method for diagnosing invasive ductal carcinoma in breast. The diagnosis is reached within minutes and treatment can be planned without delay, which ensures optimal patient care. In addition, this method can be used to establish a diagnosis on material that will be snap-frozen for future research purposes. ) that distinguished hTERTtransduced and non-transduced cultures. We suggest that hTERT immortalization reflects a physiologic process since expression patterns of hTERT-transduced and spontaneously immortalized primary cultures along IAS genes were correlated. To identify whether these genes relate to in vivo breast cancer, we used the IAS to supervise the analysis of 295 breast cancer patients. The signature strongly predicted distant metastasis-free (P = 0) and overall survival (P < 0.001) in ER-positive tumors but not in ER-negative tumors. Notably, hTERT gene expression level itself did not predict clinical outcome in primary tumors. Conclusion To our knowledge, this is the first demonstration of an immortalization signature in human breast cancer. This signature is strongly associated with specific cellular pathways in human tumors that may underlie the immortalization process and influence tumor behavior. The association between the IAP signature and outcome for ER-positive tumors but not ER-negative tumors suggests that different pathways are involved in the outcome of ER-negative tumors. The selective expression of keratin genes is highly tissue-specific, and antibodies to keratins have long been used as markers of differentiation in cell and developmental biology and in pathology applications. Antibodies to keratins can mark the progress of normal versus abnormal differentiation, can detect early apoptotic changes and may even identify stem cell-enriched tissue populations. A better understanding of the function of keratins has come from identifying links between keratin mutations and a wide range of tissue fragility disorders, which have shown that keratin intermediate filament proteins contribute physical resilience to epithelial tissues. The tissue specificity of keratins may thus reflect different requirements for stiffer or more plastic cells in particular organ sites. We re-examined the early development of the mammary gland to ask whether differences in the plasticity of cell compartments, proposed to result from expression of different keratins, may contribute to morphogenesis in development. If so, such physical differences might also be important in cancer metastasis. During early development, mouse mammary glands, like other epidermal appendages, first appear as thickenings of the ectoderm, which then expand to form solid buds of epithelial cells protruding into the mesenchyme. These buds then undergo substantial growth and remodeling. Many well-established signaling molecules are now known to play a role in epidermal appendage development, yet the physical changes that take place within the cells at these early stages, and which ultimately enable them to develop their functional tissue morphology, are still poorly understood. Using immunohistochemistry on mouse mammary gland precursors between days 12 and 19 of gestation, we mapped the sequential changes in expression of keratins and some other key structural proteins. Compartmentalization of keratin expression was observed to divide the mammary gland rudiment into a number of different zones. The subsequent fate of these zones suggests that the specific expression of keratins and other structural proteins may indeed predict the functional capacity of cell populations in tissues.  Cytotoxicity.
Conclusion Taken together, these results show that OXO may exert anti-tumor and anti-angiogenic activity against MCF-7 cells via regulation of HIF-1α and VEGF.

Figure 4
Effect of OXO on VEGF and HIF-1α in MCF-7 cells exposed to hypoxia.

Figure 5
Effect of OXO on the content of hemoglobin in bFGF-treated Matrigel.

Figure 2
Effect of OXO on the proliferation of bFGF-treated HUVECs.

Figure 3
Effect of OXO on the tube formation of bFGF-treated HUVECs.
Tumor Endocrinology we have established several human breast cancer cell lines with acquired anti-estrogen resistance through longterm treatment with different anti-estrogens. These cell lines have been used for our studies of the signaling pathways, which may be activated in cells with acquired anti-estrogen resistance. Analysis of the expression of genes known to be important for human breast cancer has revealed that the majority of the anti-estrogenresistant breast cancer cell lines have decreased estrogen receptor expression and signaling. However, increased expression of phosphorylated PKB/Akt (p-Akt) and Akt kinase activity was observed in several anti-estrogen-resistant cell lines. The PI3 kinase is an upstream signaling molecule for Akt, and inhibition of PI3-kinase activity with wortmannin or LY294002 deceases the level of p-Akt. Both PI3kinase inhibitors inhibited growth of the resistant cells. However, wortmannin displayed a more profound growth inhibitory effect on antiestrogen-resistant cell lines than on parental MCF-7 cells. Treatment with the novel Akt inhibitor SH-6 resulted in a very strong growth inhibition of three resistant cell lines overexpressing p-Akt, whereas the parental MCF-7 cells were significantly less growth inhibited. It was investigated whether the increased level of p-Akt in the resistant cells was due to signaling from IGF-IR and IRS-1, or whether it resulted from decreased PTEN activity. Both involvement of IGF-IR and PTEN could be excluded. At present, our working hypothesis is that anti-estrogen-resistant human breast cancer cell lines with an increased p-Akt level require signaling via activated Akt to survive and maintain growth in the presence of the anti-estrogen. Studies on clinical material will be important to evaluate whether anti-estrogen-resistant tumors overexpress p-Akt and whether Akt may be a target for treatment of anti-estrogen-resistant breast cancer.  HIN-1) is a small, secreted protein that was initially identified as a protein the expression of which is lost in the vast majority of breast cancers. The silencing of HIN-1 expression is due to methylation of its promoter, which in addition to breast cancer also occurs in a significant fraction of many other types of solid tumors including prostate cancer, lung cancer, pancreas cancer, and retinoblastoma, suggesting a potential tumor suppressor function. Consistent with this hypothesis, in non-small-cell lung cancer, downregulation of HIN-1 expression was found to be the most significant independent predictor of poor clinical outcome in stage I disease, suggesting loss of HIN-1 expression is a functionally important event. The receptor of HIN-1 is unknown, but ligand-binding studies indicate the presence of high-affinity cell surface HIN-1 binding sites on the same epithelial cells that express HIN-1, suggesting that HIN-1 functions as an autocrine factor. Methods To further explore the function and mechanism of action of HIN-1, we examined the effect of HIN-1 on immortalized mammary epithelial cells and a panel of breast cancer cell lines in vitro. HIN-1 was delivered via multiple expression systems, including a tetracyclineregulated expression system, recombinant adenovirus, and recombinant HIN-1 fusion protein.

P4.04 HIN-1, an inhibitor of cell growth, invasion, and AKT1 activation
Results We find that HIN-1 is a potent inhibitor of anchoragedependent and anchorage-independent cell growth. Expression of HIN-1 in MDA-MB-435 cells causes reversion of their invasive morphology when these cells are grown in a three-dimensional culture system. As further evidence that HIN-1 can influence invasive behavior, expression of HIN-1 markedly inhibited the invasion and migration of breast cancer cell lines in Boyden chamber assays. Expression of HIN-1 in synchronized cells inhibits cell cycle re-entry and the phosphorylation of the retinoblastoma protein (Rb), while in exponentially growing cells HIN-1 induces apoptosis without apparent cell cycle arrest or an effect on Rb phosphorylation. To begin to dissect the mechanism by which HIN-1 suppresses growth, we analyzed the activation status of various signal transduction pathways involved in cell proliferation and survival using activation state specific antibodies. This investigation revealed that mitogen-induced phosphorylation of Akt (Ser 473) is inhibited in HIN-1-expressing cells. Expression of HIN-1 also inhibits Akt-mediated retention of p27 in the cytoplasm. Further supporting the role of Akt in HIN-1-mediated growth inhibition, expression of constitutively activated Akt abrogates HIN-1-mediated growth arrest. Conclusion Taken together, these studies provide further evidence that HIN-1 possesses tumor suppressor functions and suggest that these activities may be mediated through the Akt signaling pathway. The signal transducer and activator of transcription 1 (STAT1) in human primary mammary carcinoma was found to be a predictor of good prognosis for the outcome of disease [1]. This is in accordance with its documented role in growth arrest and in proapoptotic signaling.

P4.05 Expression of STAT1 target genes and interferon gamma in human mammary carcinoma tissue
Methods In order to define signaling pathways employed by STAT1 to exert its effect on the tumor and to define the role of interferon gamma (IFN-γ) in its activation, we have investigated the expression of known STAT1 target genes and of IFN-γ in the primary tumor by quantitative RT-PCR. The study was performed with a total of 111 different primary tumor samples.

Results
The expression of the two tumor suppressor genes IRF-1 and suppressor of cytokine signaling 1 (SOCS1) were found to be correlated with the activation status of STAT1, as determined by measuring tyrosine phosphorylation of STAT1 by western blotting, DNA binding by electromobility shift assays and nuclear localization by immunohistochemistry. IFN-γ expression was correlated to the expression of some, but not all, STAT1 target genes. However, it did not correlate with constitutive STAT1 activation. Survival analysis revealed that, in contrast to STAT1 activation, IFN-γ expression was not a predictor of a longer overall or relapse-free survival.

Conclusions
Our results indicate that, in the majority of primary mammary carcinomas investigated, the constitutive activation of STAT1 does not depend on increased IFN-γ secretion (e.g. as a result of an inflammatory reaction in the tumor). This suggests a prominent role for IFN-γ-independent mechanisms leading to the constitutive activation of STAT1 in primary mammary carcinomas. The frequent induction of the tumor suppressor genes SOCS1 and IRF-1 in carcinoma tissue with activated STAT1 implies a potential role of these genes in mediating the good prognostic effects of STAT1 activation. cancers. An assay, based on a non-random elimination of human chromosome segments in mouse-human microcell hybrids during tumor growth in SCID mice, has been developed. It is called the Elimination test (Et), and is designed for the identification of chromosomal regions containing putative TSGs. A commonly lost segment, termed chromosome 3 common eliminated region 1 (C3CER1) (also referred to as CER1), was identified using the Et. This region spans ~2.4 Mb at 3p21.3 and contains 33 active genes, including the putative TSGs: LF, LIMD1 and RIS1. It was recently reported that the LIMD1 gene product functions in connection with pRB to suppress cell proliferation. Here, we addressed the question of whether the C3CER1 region at 3p21.3 is preferentially lost in actual human breast tumors, and whether the fragile nature of FRA3B induces terminal deletions leading to 3p14.2-pter losses or whether the eventual 3p21.3 losses are interstitial. We also analysed whether the LIMD1 gene was mutated in human breast tumors. Methods To evaluate the loss of C3CER1 in human tumor tissues we performed loss of heterozygosity (LOH) analysis of 159 breast tumors. We compared the deletion frequency of the C3CER1 area with two other regions on 3p; that is, the FHIT/FRA3B region at 3p14.2 and VHL at 3p25.3. The tumor material was screened for mutations with the SSCP method, and samples with abnormal mobility in SSCP gels were sequenced in an ABI 3100 genetic analyser. We are conducting multipoint FISH analysis to confirm microdeletions in the breast tumor specimens. Results LOH was detected in the C3CER1 region in 84% of informative tumors. Thirty-nine percent of LOH-positive tumors showed LOH at all informative C3CER1 markers. The other 61% had a discontinuous LOH pattern suggesting interstitial deletions or breakpoints. In the VHL and FHIT regions, deletions were observed in 69% and 30% of tumors, respectively. We found polymorphism in the first three exons of LIMD1, but sequencing revealed no mutations leading to changes in the protein product. Conclusion Of the three 3p regions analysed, the highest deletion frequency was observed at the C3CER1 region. We demonstrate that the interstitial deletions including C3CER1 prevail over 3p14.2-pter losses. The LIMD1 gene is not frequently mutated in breast cancer biopsies. We have previously demonstrated that overexpression of cyclin A is associated with poor prognosis in breast cancer patients [2]. In this study we wanted to evaluate the impact of other cyclins, both at the gene level and at the protein level. We wanted to evaluate whether the overexpression of cyclins is a result of gene amplification, as well as to evaluate the prognostic value of gene amplification of different cyclins for breast cancer patients. The impact of TP53 gene mutations on gene amplification of cyclins was also evaluated. Real-time quantitative PCR was used to detect gene amplification of cyclin A, cyclin B 1 , cyclin C, cyclin D 1 , cyclin D 3 , cyclin E and cyclin H in tumour tissue from 80 patients operated for invasive breast carcinomas, while immunohistochemistry was applied to detect protein expression of the same cyclins. Among the 80 breast cancer tumour samples examined, 26.7% was defined to have ccnA 2 gene amplification, 37.2% had ccnB 1 gene amplification, 82.6% of the samples harboured amplification of ccnC, 74.4% had ccnD 1 gene amplification, 41.9% had ccnD3 gene amplification, 29.1% of the patients had ccnE gene amplification and 9.3% of the samples showed amplification of the ccnH gene. When correlation between gene amplification and protein expression was evaluated, we observed a statistical significant correlation between gene amplification and protein expression of cyclin A (correlation coefficient = 0.287, P = 0.009) and cyclin D 3 (correlation coefficient = 0.906, P = 4.9 × 10 -33 ). Protein expression as well as gene amplification of cyclin A was also correlated with gene amplification of other cyclins. When the impact of gene amplification of different cyclins on the patient survival was analysed, only gene amplification of cyclin A was associated with patient survival. We found a significant interaction between amplification of cyclin A and cyclin E (Cox regression, P = 0.02). These two cyclins are sequentially time related in the cell cycle. The effect of amplification of cyclin A was therefore tested in a stratified analysis both when the cyclin E gene was not amplified and when the cyclin E gene was amplified. When the cyclin E gene was not amplified, the statistical strength of the cyclin A amplification increased with a HR of 5.5 (95% confidence interval: 2.2-14.3, P < 0.0001). When cyclin E was amplified, amplification of cyclin A had no significant impact on survival (P = 0.45). In summary, we have analysed gene amplification and protein expression of both primary and secondary cyclins in invasive breast carcinomas. Overexpression and gene amplification of cyclin A is correlated with gene amplification of other cyclins. Only gene amplification and overexpression of cyclin A was associated with poor prognosis, and amplification of cyclin A is the strongest prognostic factor in patients that have a normal amplicon of cyclin E.  [2]. IGF-independent effects of IGFBP5 have also been described and it has been postulated that these are at least partially mediated via IGFBP5 localized in the nucleus [3,4].

References
Methods The cellular localization of IGFBP5 was analyzed by confocal microscopy after either applying exogenous fluorescent-labeled recombinant protein or applying immunostaining of cells ectopically expressing IGFBP5. HC11, MCF10A mammary epithelial and T47D mammary carcinoma cell lines were used in this study. Results Nuclear localization of IGFBP5 was observed under two conditions: fluorescent-labeled IGFBP5 added to cells with selectively permeabilized plasma but not nuclear membrane; and cells transfected with IGFBP5 expression vectors lacking the coding region for the signal peptide. By contrast, non-permeabilized cells could be stimulated to take up IGFBP5 only into intracellular vesicles outside the nucleus and this was enhanced by adding a conjugate of polylysine and transferrin, indicating an endocytotic uptake route. In addition, cells transfected with IGFBP5 containing the signal peptide secreted IGFBP5 into the medium but did not show any detectable nuclear staining.
Conclusions Nuclear localization of IGFBP5 in mammary epithelial cells required the crossing of the plasma membrane, which does not appear to occur under normal cell culture conditions. Exit of IGFBP5 from endosomal vesicles into the cytosol followed by nuclear uptake was never observed. Our results indicate a minor role or no role of nuclear IGFBP5 in mediating its IGF-independent effect in the mammary epithelium and in breast cancer. Oligonucleotide microarray analysis identified 78 probe sets differentially expressed in the hybrids as compared with CAL51 and the rerevertant cells. The majority of these genes is involved in signal transduction, developmental processes, angiogenesis, cadherin signaling, Wnt signaling or inflammation. It is of particular interest that the 78-gene signature is also reflected in a panel of breast tumors, lymph node and distant metastases, and is correlated with several prognostic markers including tumor size, grading, metastatic behavior and estrogen receptor status. As opposed to the corresponding non-tumorigenic phenotypes demonstrated for the MDA-MB-231-derived and CAL51-derived microcell hybrids, the respective differentially expressed genes strongly differ from each other. However, it was of special interest that the majority of genes of both gene sets could be integrated into a similar spectrum of biological processes and pathways. Our findings provide an experimental system to identify and evaluate genes but, more importantly, signatures of biological processes and pathways involved in the development and/or progression of breast cancer. S40 supplied. In order to identify key genes and elucidate the regulatory pathways that are involved in the development and progression of BC, we combined array-based expression profiling with a powerful functional approach, the microcell-mediated chromosome transfer.

Methods and results
An intact copy of chromosome 6 was transferred into the 6q deleted and highly invasive BC cell line MDA-MB-231 using the microcell-mediated chromosome transfer. It was demonstrated by microsatellite allelotyping, CGH and FISH that the hybrid clones contain two fragments of chromosome 6, spanning 6p22-q14 and 6q15-6q27. The transfer of parts of chromosome 6 into the MDA-MB-231 cells resulted in reduced anchorage-dependent growth, reduced in vitro invasion and a strongly reduced tumorigenic potential. In order to identify genes responsible for the observed reversion of the tumorigenic phenotype in MDA-MB-231 cells, differential gene expression between the parental cell line and the hybrid clones was analysed using oligonucleotide micro-arrays (HG-U133; Affymetrix, Santa Clara, CA, USA). The identified set of differentially expressed genes will be presented. Conclusions Our results provide functional evidence that the suppression of the tumorigenic phenotype of the BC cell line MDA-MB-231 is mediated by a specific set of genes regulated by one or more genes on chromosome 6. We have recently reported that TGF-β regulates tropomyosin-based actin microfilament fibers [1], which are essential for cell proliferation, morphology and motility [2]. Smads and p38 MAPK mediate induction of tropomyosin and formation of stable actin microfilament fibers (stress fibers), thereby reducing cell motility. Tropomyosin (TM) is a dimeric coil-coiled protein that binds along actin microfilaments forming a head-to-tail polymer. TM stabilizes microfilaments and protects them from the depolymerizing action of gelsolin and cofilin. Importantly Background Promoter hypermethylation is a common inactivation mechanism in the development and progression of neoplastic transformation. For mammary carcinoma numerous genes have been described to be silenced due to aberrant methylation, some of them already in intraductal carcinomas. Much less is known about the association of hypermethylation events with the different histological subtypes of breast cancer. Methods Genomic DNA was isolated from fresh-frozen and formalinfixed paraffin-embedded biopsies and was treated with bisulfite for subsequent methylation analysis. Altogether 40 lobular breast cancer, 89 ductal breast cancer, and 20 normal breast tissue samples were analyzed. For this purpose, real-time PCR-based quantitative methylation assays were developed for the following genes: p16INK4a, cyclin D 2 , RASSF1A, GSTp1, RIZ-1, HIN-1, APC, DAP kinase, Twist, and SOCS-1.

P4.15 Epigenetic silencing of tropomyosin alters transforming growth factor beta control of cell invasion and metastasis A Bakin, A Varga, Q Zheng, A Safina
Results A stringent threshold for scoring a sample as 'methylated' (mean of the methylation level in the control group plus twice the standard deviation) was established for every gene analyzed. Differences in DNA methylation between ductal and lobular breast cancer concerning frequency, intensity, age dependence and concurrence of hypermethylation were uncovered. The most frequently hypermethylated genes in the whole series of 129 breast cancer specimens were cyclin D 2 (75.2%), RASSF1A (71.3%), and HIN-1 (66.7%). The p16INK4a and the RIZ-1 genes were only rarely methylated. A quantitative analysis of the methylation levels using the Mann-Whitney test revealed a statistical significant association of the methylation of the genes DAP kinase and cyclin D 2 with the lobular subtype. By contrast, a mere qualitative scoring of methylation data did not reveal any significant differences.

Conclusions
The results presented in this study demonstrate that subtype-specific patterns of aberrant gene methylation exist in breast cancer, which will help to elucidate the underlying biological differences. These subtype-specific patterns could only be revealed by using stringent real-time PCR-based quantitative methylation assays. Five specimens (section material) of the normal mammary gland tissue and peripheral blood lymphocytes of 30 healthy subjects were also investigated.

P4.17 Methylation profiling of carcinogenesis-associated genes in sporadic breast cancer
Results By methylation-sensitive PCR with specific primers we detected no methylation of any investigated genes in control peripheral blood lymphocytes and in five normal breast tissues. High frequencies of promoter methylation were observed for the major TSG involved in controlling the cell cycle through the Cdk-Rb-E2F signaling pathway: RB1, 17%; p16, 56%. Methylation of both genes was revealed in 12% of tumors. p15 was methylated in 2% only. No methylation was observed for the CpG island of p14. The methylation frequency was rather high in the case of the CDH1 promoter (37%), maximal in the case of the HIC1 promoter (79%), and relatively low in the case of MGMT (8%) and N33 (9%). Methylation was shown to be 32.5% for LAMC3. Laminins contain motifs providing cell growth and differentiation via selective activation of signaling pathways dependent on protein kinase C and, probably, involved in metastases spreading. We have detected abnormal methylation of the TGFBR1 CpG island in 35% samples. TGFBR1 takes part in transforming growth factor beta signaling, mainly resulting in inhibition of cell proliferation. None of tumors showed methylation of all genes. No gene was methylated in 11% of tumors, and at least one gene in 89% of tumors.
Conclusion We have shown that CpG methylation in the promoter regions of TSG is common to sporadic BC. Genes with a higher methylation frequency may be included in the BC methylotype. Identification of the genes with a high methylation frequency is a necessary step in characterizing a particular tumor. Along with other molecular genetic markers, the methylation profile may be employed in early diagnostics and prognostication. Background Wnt signaling, initially identified in early embryogenesis of Drosophila, is involved in a large set of cellular processes, including proliferation, differentiation, migration, and apoptosis. Canonical Wnt signaling is involved in cell fate choices, stem cell renewal and differentiation, whereas non-canonical signaling deals with morphological changes and tissue organization. Since the discovery of Wnt-1 as a virally-induced oncogene in mouse mammary tumors, Wnt signaling has become a center of interest in human breast carcinogenesis. Surprisingly, genetic aberrations of Wnt signaling have been revealed in different malignancies including colon cancer, liver cancer and hematological cancer, but not in breast cancer. Thus, epigenetic changes in Wnt signaling, rather than mutations, may be more relevant to breast cancers in humans. However, this aspect of Wnt signaling in breast cancer is not fully understood. We aimed to study the expression of key components of Wnt signaling in breast cancer. Here, we report the preliminary results obtained from our expression profiling studies for all known Wnt ligand, frizzled receptor, co-receptor and Lef/TCF transcription factor genes, using a set of cell lines and primary tumors. Methods We studied the expression profile of 34 Wnt pathway genes by the RT-PCR technique. RNAs were extracted from a telomeraseimmortalized human mammalian epithelial cell line (HMEC), six breast cancer cell lines and 15 breast tumors. Tumor samples were selected following pathological analysis of fresh-frozen tissue slices. Corresponding cDNAs were synthesized and subjected to PCR amplification using specific pairs of primers. Equal loading of total cDNAs was checked by PCR analysis of the housekeeping gene GAPDH.

P4.18 Expression profiling of Wnt pathway genes in breast cancer
Results HMEC cells, used as a normal control, expressed many Wnt signaling genes, including 9/19 (48%) Wnt ligands, 7/9 (78%) frizzled receptors, LRP-5, LRP-6, as well as four LEF/TCF transcription factors. Expression patterns of frizzled receptors, LRP-5, LRP-6, and LEF/TCF transcription factors did not show major changes in breast cancer cell lines. The major change in Wnt signaling genes was observed at the level of ligand expression. The expression of Wnt-3a, Wnt-4, Wnt-6, Wnt-8b and Wnt-9a were upregulated in 50% or more breast cancer cell lines. Conversely, the expression of Wnt-5a, Wnt-9b and Wnt-16 was downregulated. Our ongoing studies with breast tumors indicate that Frizzled-1, Frizzled-2 and Frizzled-6 expression is also maintained in breast tumors. Moreover, upregulation of Wnt-4 and Wnt-9a, as well as downregulation of Wnt-5a expression, were observed in 79-100% of tumors. Conclusion These observations provide evidence for redundant expression of major genes involved in Wnt signaling in both normal and malignant breast cells. The expression of at least nine Wnt genes in HMEC strongly suggests that some Wnt ligands may provide autocrine or paracrine signaling to normal breast epithelial cells. Six Wnt genes were commonly expressed in both HMEC and breast cancer cell lines, suggesting that some Wnt ligands may not be significantly involved in malignant transformation of mammary epithelial cells. On the other hand, malignant cells have upregulated the expression of Wnt-3a, Wnt-4, Wnt-6, Wnt-8b and Wnt-9a genes that may play a positive role in malignancy. Wnt-3a and Wnt-4 are known to display transforming activity in mammary epithelial cells. The function of Wnt-8b in mammalian cells is not well known, but its Xenopus homolog displays strong axis-duplication activity, suggesting that it may also be a transforming Wnt. On the other hand, the expression of Wnt-5a, Wnt-9b and Wnt-16 was switched-off in malignant breast cells. Although the functions of Wnt-9b and Wnt-16 are not well known, Wnt-5a has been identified as a tumor suppressor in hematological malignancies, and acts as an antagonist of canonical Wnt signaling. Taken together, these results indicate that there is a switch in Wnt ligand expression pattern in breast cancer cells, and that this may provoke a functional switch in Wnt signaling from non-canonical to canonical pathways. Background BRCA1 possesses a number of features common to transcriptional regulatory proteins, suggesting that it may regulate the expression of one or more downstream genes. It is important to determine which genes are transcriptionally influenced by BRCA1 in vivo to explain its role in tumor suppression and in cancer development. In our previous study, a BRCA1 overexpression system enabled us to define the genes whose expression levels were induced in MCF-7 breast cancer cells by using the PCR-dependent suppression subtractive hybridization technique [1]. Herein, we report the preliminary results obtained from our real-time expression profiling of normal-matched primary breast tumors for six genes, three of which were previously reported [1]. The association between the gene expression profiles and histopathological states of these tumors will contribute to the definition of possible diagnostic markers.

P4.19 Real-time PCR-based expression profiling of BRCA1-induced genes in primary breast tumors
Methods Breast tumors were selected following pathological analysis of fresh-frozen tissue sections. RNAs were extracted from 31 normalmatched breast tumor tissues. Synthesized cDNA samples were subjected to real-time PCR using the QuantiTect SYBR green PCR Master Mix with gene-specific primers. GAPDH is used as a housekeeping gene for normalization. The gene expression levels were quantified using the delta-delta Ct method after normalizing each tumor with its normal counterpart.
Results The real-time expression level of BRCA1 was highly correlated with ERBIN and SMG1 (Pearson correlation, Minitab; n = 31; r = 0.765 and r = 0.673, respectively; P < 0.0001). The pairwise correlations of BRCA1 expression with those of RENT2 and OVCA1, but not with OVCA2, were at moderate levels (r = 0.41 and r = 0.46, respectively; P < 0.05). Furthermore, primary breast tumors were hierarchically clustered into two major groups based on their real-time gene expression profiles using the CLUSTER program and were visualized by TRIVIEW [2]. Cluster I tumors were characterized by a high-level expression in BRCA1 target genes (n = 20; 1.52 ± 0.6, log 2 ) and were low grade on average (37.5% I, 50% II, 12.5% III; n = 16).
On the other hand, Cluster II included higher grade tumors (45% II, 55% III; n = 11) expressing BRCA1 target genes at a lower level (n = 11; -0.79 ±0.7, log 2 ). Based on the Mann-Whitney U test, Cluster I and Cluster II were significantly different in terms of their tumor grades (W = 206; P = 0.0059).
Conclusion This study demonstrated that real-time RT-PCR studies provide highly accurate quantitative profiling for marker gene association with tumor subtypes. The mRNA expression of ERBIN, ERBB2/HER2 binding protein, was found to be tightly correlated with that of BRCA1 in primary breast tumors, as found in MCF7 cells ectopically expressing BRCA1 [1]. The OVCA1 tumor suppressor gene (17p13.3) that displays frequent LOH in both ovarian cancer and breast cancer also showed correlation with BRCA1 in primary breast tumors used in our study. A certain degree of expression variability, part of which could be attributable to the variation in tumor grade, exists for the genes used in this study, including BRCA1. Our findings support the view that association of the patients' clinical and pathological parameters with the gene expression profiles of breast tumor samples carries great importance in the classification of tumor subtypes.

P4.20
Development of a rapid screening approach for candidate gene sets in cancer for certain cancer phenotypes. A synergy between these advances and the development of screening tools for a rapid and reliable screening of marker gene expression represents an important step towards an improved treatment of cancer. Methods For the semiquantitative expression screening of 11 candidate genes for drug resistance in melanoma, we combined multiplex RT-PCR (mRT-PCR) with subsequent microfluidic fragment analysis.
Results The functionality of this approach was demonstrated by low interexperimental variations of amplicon quantities after endpoint analysis. Applied to RNA samples derived from drug-sensitive and drug-resistant melanoma cell lines, mRT-PCR delivered results qualitatively concordant with data obtained from northern blot analyses and array analyses. A preliminary screen of four additional melanoma cell lines points to IL1B, APOD, and CYR61 as interesting candidates for drug-resistance associated genes. First tests using an automated on-chip electrophoresis platform indicate the applicability of this approach for high-throughput measurements.
Conclusion mRT-PCR combined with on-chip electrophoresis reveals a rapid and easy-to-handle method for candidate gene set evaluation from limited amounts of mRNA. Using gene sets indicative for different tumor phenotypes, this procedure may represent an alternative for future cancer diagnostics. Background Among breast cancer patients receiving ionizing radiation (IR) treatment, a subgroup shows adverse long-term effects in the normal tissue. Radiation-induced fibrosis (RIF) is one of the most serious complications, and risk of RIF is a dose-limiting factor in the treatment of breast cancer patients with IR. The mechanisms whereby IR induces RIF are not fully understood. However, several observations indicate that the variation in normal tissue sensitivity and the consequent risk of developing late morbidity may be genetically determined. The aim of this study was to obtain a comprehensive overview of the changes in gene expression after IR and to identify genes that can be used to predict risk of RIF, using microarray analyses. Materials and methods Normal fibroblasts were achieved from 41 patients treated with postmastectomy radiotherapy in Aarhus, Denmark, from 1978 to 1982 and subsequently evaluated in detail with regard to development of RIF. The fibroblasts were grown to early confluency before they received radiation. Total RNA was isolated both before and after radiation, labelled and hybridized to cDNA microarrays consisting of 15,000 cDNAs and ESTs [1]. Expression profiles were identified using hierarchical cluster analyses [2]. Statistically significant changes in gene expression were identified using significance analysis of microarrays (SAM), and predictive genes were identified using prediction analysis for microarrays (PAM) [3]. Results and conclusion Microarray data were first analyzed in order to identify radiation-responsive genes. While several genes were involved in known IR response pathways such as cell cycling, proliferation and stress, a substantial fraction of the genes were involved in processes not previously associated with IR response. Of particular interest are genes involved in extracellular matrix composition. SAM analyses were also applied to identify genes in which the expression level correlated with the level of fibrosis. PAM analyses identified a limited set of predictive genes that may provide a basis for a diagnostic tool in the identification of patients with adverse responses to radiation, and to improve and optimize radiotherapy at the individual level. Background Repair of damaged DNA is a highly regulated process in normal tissue. Several human genetic diseases are known to be or suspected to be due to defects in DNA repair or cell cycle control. Some of these patients are radiation sensitive and/or predisposed for cancer as a cause of mutations in genes involved in these cellular pathways. A well-known group of radiation-sensitive patients is the ataxia-telangiectasia (A-T) patients. This disease is caused by mutations in the ATM (A-T mutated) gene, whose gene product is involved in detecting double-strand breaks. In this study we are trying to reveal the cause of radiation sensitivity in a group of radiationsensitive patients having an A-T phenotype without mutations in ATM.

P4.21 Application of microarray analyses to identify genes involved in radiation-induced fibrosis OK Rødningen 1 , J Alsner 2 , T Hastie 3 , J Overgaard 2 , A-L Børresen
Methods Gene expression studies were conducted using 15k cDNA (NRH) microarrays on lymphoblastoid cell lines obtained from four control individuals ('normal'), four A-T patients and 10 radiationsensitive patients, before and after radiation. Cells were harvested prior to radiation (0 hours) and at 2 hours, 8 hours and 24 hours, respectively, after exposure to ionizing radiation. The cell lines were irradiated with a dose of 2.0 Gy. To be able to study possible similarities and differences in the expression patterns between the three groups of cell lines, we used cluster analyses.

Preliminary results
The preliminary results suggest that the radiationsensitive patients constitute a heterogeneous group, and that the cause of their radiation sensitivity may be diverse. Conversely, several samples showed consistency in their gene expression patterns, which might reveal relevant genes and unknown pathways. To understand the biological context we need a broader base of comparison. Ongoing experiments include more samples in this study and will hopefully enable us to reveal the cause of the radiation sensitivity in these patients and bring us a step closer to the understanding of early malignancy development. Background Women who received mantle-field irradiation following Hodgkin's lymphoma (HL) have an age-dependent increased risk of developing breast cancer. It has been estimated that approximately 90% of the breast carcinomas in these patients is a result of their radiation treatment, which makes this series extremely appropriate to determine a potentially radiation-induced genomic profile.

P4.23 Breast tumors induced by high-dose radiation display similar genetic profiles
Methods In this study we have used array-comparative genomic hybridization (array-CGH) and gene expression profiling (GEP) technology to assess the genomic and gene expression changes in radiation-induced breast tumors. For genomic profiling we used DNA isolated from paraffin-embedded primary breast tumors of breast following HL (BfHL) patients (n = 29). These DNAs were hybridized to a small custom-designed BAC array containing 180 clones specifically selected on their function in the DNA-damage repair pathway or breast cancer susceptibility. For gene expression profiling, RNA was isolated from fresh-frozen tissue samples of 13 BfHL patients and hybridized on 35K human oligo-array as well as from sporadic breast tumors that were included as controls, matched for age at diagnosis and no exposure to radiation.
Results Hierarchical clustering of all the array-CGH data divided the samples into two groups. One cluster consisted of the tumors that had developed in the unprotected area of the breast that received highdose radiation (20-40 Gy) during treatment. These tumors showed a significantly higher frequency of amplifications and deletions than those that had developed in the low-dose radiation (1-10 Gy) breast area and the control breast tumors. We performed supervised classification on the two groups in order to construct a predictor for identification of tumors that occurred in high-dose irradiation or low-dose irradiation fields. A classifier employing 15 BAC clones achieved the best crossvalidation performance, and had an associated validation performance of 91.7%. Hierarchical clustering of the GEP data was performed using 6111 significantly differently expressed oligomers, which resulted in a clustering of the so-called radiation-induced tumors separate from the sporadic tumors (software program Genesis). A supervised classification strategy of these two groups revealed 50 significant genes that could differentiate low-dose and high-dose radiation BfHL versus sporadic tumors. Importantly, the 'radiation profile' found with the array-CGH could be confirmed by the GEP data; for example, deleted genomic regions showed downregulation of expression. Conclusion Our preliminary results indicate that low-dose radiation as well as the high-dose radiation-induced tumors can be distinguished from sporadic tumors on the basis of their genomic and expression profile. Results Hierarchical clustering sharply discriminated two groups of BCC. To determine whether a BCC was of 'luminal' or 'basal-like' subtype, hierarchical clustering was done with the subset of genes selected by Sørlie and colleagues to discriminate molecular subtypes of tumors: eight BCC were luminal-like, 12 were basal-like, and 11 were not clearly affected to one subtype. Supervised analysis selected the most discriminant genes between basal and luminal subtypes, and corresponding antibodies were tested on CMA when available. The immunohistochemical profile was in accordance with the transcriptional profile.

P4.24 Molecular characterization of breast cell lines: a tool for breast cancer studies
Conclusion According to the molecular classification of breast tumors, we have determined two different subtypes of cell lines, luminal and basal-like, and isolated a subset of genes that discriminates both subtypes. The DNA array data were confirmed at the protein level. This characterization of breast cell lines provides a powerful tool to study specifically each molecular subtype of breast cancers. The master genes isolated from each group may be new targets for breast cancer management. Immunohistochemical and mRNA expression profiling studies of large breast cancer cohorts have reproducibly identified a subset of tumors (~15%) that express markers characteristic of the basal layer of the mammary gland. This is in contrast to the many human breast cancers that uniformly express luminal markers such as the simple cytokeratins (K8/K18) and appear to originate from transformed luminal epithelial cells. A logical next step is to determine the dominant signaling pathways and genetic defects that drive tumor initiation and progression, and to understand how they are related to cell lineage in each breast cancer subtype. A large number of breast cancer cell lines have been isolated and individually characterized over the past few decades. We and others have begun to comprehensively align these cell lines with primary tumors based on gene expression profiles and other parameters in order to improve the relevance of data obtained from these experimental models for understanding human disease. Methods We have characterized a panel of 51 breast cell lines for a large number of properties including in vitro and in vivo growth rates, morphology on plastic and in three-dimensional matrices, and sensitivity to estrogens/anti-estrogens. Baseline microarray profiles were generated using Agilent 60-mer oligonucleotide arrays for each cell line. In addition, we have generated 'signature profiles' for selected cell lines in response to exogenous stimulation such as estrogen and TGF-β. A constraint-based analysis of microarray profiles generated from primary tumors and breast cancer cell lines in combination with response signatures was used to identify candidate genes and pathways that may play dominant roles in the breast cancer subgroups. transduction pathways with cell-type origin, and further resolves the biological and clinical significance of the well-established markers ER and HER-2. The proposed stratification is likely to help explain the wellknown diversity in response of breast cancers to standard therapeutic regimens and, more importantly, may identify appropriate breast cancer subtypes amenable to targeted therapeutics. Methods In 295 stage I and stage II breast carcinomas treated at the Netherlands Cancer Institute, we have obtained gene expression data of 25,000 genes using micro-array analysis. We have used the previously described Intrinsic Gene Set [3] to define basal type, luminal A and luminal B, ErbB2 and normal epithelium-like type tumors (431 of 487 unique genes matched). We have used two different methods to classify the tumors: two-dimensional hierarchical cluster analysis and nearest centroid classification. We have compared the reproducibility by both methods and we have analyzed clinical outcome (distant metastasis-free probability and overall survival) of these 295 patients based for the different classes. The median follow-up is 6.7 years for all patients and 7.8 years for patients alive.

P4.26
Results Based on hierarchical clustering, the basal subgroup can be easily recognized; the ErbB2 group is reasonably well defined and the luminal A and luminal B groups form a large cluster, with subclusters that have more luminal A or luminal B patients. For the nearest centroid classification we used a correlation threshold of 0.1 to classify patients. One hundred and nine (37%) patients did not have a correlation of more than 0.1 to one of the five centroids ('unclassifiable'). Forty-five (15.25%) patients were assigned to the basal group, 39 (13.2%) ErbB2, 47 (16%) luminal A, 45 (15.25%) luminal B and 10 (3.3%) normal-breast like. The relatively large group of patients that could not be assigned to one of the classes was further analyzed. These tumors appear to represent a relatively homogeneous group that differs from those that can be classified. The ER receptor is positive in 106/109 (120/188 classifiable patients: two-sided Fisher's exact P < 1 × 10 -9 ) and 80% of the tumors are histological grade I or grade II (47% for classifiable patients; P < 1 × 10 -6 ). Not surprisingly, the 10-year overall survival is higher in these patients as well (80% vs 64%; log-rank: 0.0005). Using predicting analysis of micro-arrays [4], the unclassifiable 'class' could be predicted using 200 genes with an accuracy of 90% (cross-validation results). The 10-year metastasis-free probability and overall survival for the subgroups are: basal, 54% and 46%; erbB2, 55% and 56%; luminal A, 70% and 83%; luminal B, 56% and 63%; and normal-breast like, 67% and 90% (overall P value: metastasis-free probability, 0.15 and overall survival, 0.001).
Conclusion In this series of consecutively treated breast cancer patients, the molecular portraits identify patients that differ with respect to prognosis. The relatively high proportion of unclassifiable patients can possibly be explained by both the cross-platform matching, the difference in clinical stage (locally advanced in the original series versus early stage in our patients), and the fact that the original classification was derived from a relatively small series of tumors. The  [1]. These two subtypes of breast tumors have also been associated with a significant difference in clinical outcome: the Luminal A subtype patients have been correlated with a significantly longer overall survival or they lived considerably longer before experiencing relapse or metastatic disease, whereas patients with Basal subtype tumors showed the shortest overall survival time or experienced much shorter disease-free time intervals [2,3]. To further substantiate the prognostic value of such expression-based phenotypes in diagnosis/prognosis of breast cancer, we report here an extended study on identification and molecular characterization of clinically relevant subtypes in early stage breast carcinomas.
Methods In this study, we profiled 20 biopsy tissues from early stage breast carcinomas using the Applied Biosystems Human Genome Survey Microarrays, which is a relatively new array platform containing 31,700 60-mer oligonucleotide probes representing a set of 27,868 individual human genes, with single-color chemiluminescence detection. To identify the subtypes in these tumors, we first carried out a centroid correlation analysis coupled with an unsupervised hierarchical clustering analysis. We utilized the 'intrinsic' gene list consisting of 534 genes that have been used to define the five subtypes of breast tumors and their core expression centroids in 122 previously published breast tumors samples [3]. Using the mapped 526 intrinsic genes, we computed the Pearson's correlation coefficient of each sample from this study to each of the five centroids and assigned each sample to the subtype to which it showed the highest correlation. As a second approach, we applied a supervised analysis using the 'Nearest Shrunken Centroid classifier' and the PAM software [4]. We took the previously published 122 Norway/Stanford tumor samples and the mapped 526 intrinsic genes as the training set to identify the predictor genes for the five subtypes. We then used this classifier to predict the subtypes of each of the 20 early stage carcinomas analyzed in this study. The same analyses were applied on parallel datasets generated from Stanford cDNA Arrays and Agilent Human Whole Genome Arrays. Welch-ANOVA analysis coupled with Benjamini and Hochberg False Discovery Rate multiple testing corrections were performed to identify the 'signature' genes that are most differentially expressed between the subtypes. PANTHER™ protein classification analysis (Applied Biosystems, Foster City, CA, USA) [5,6] and PathArt™ (Jubilant Biosys Ltd) pathway analysis were carried out to identify molecular mechanisms underlying these 'signature genes'. A minimal set of genes that best discriminated the two identified subtypes were determined using PAM analysis on the combined datasets generated on the three different array platforms.
Results Both unsupervised and supervised analysis identified the two main clinically relevant subtypes of breast cancer, Luminal A (correlated with a relatively good outcome) and Basal-like (correlated with the poorest outcome). The identification of the Luminal A and Basal subtypes in these early stage breast carcinomas was further validated by parallel data generated from Stanford cDNA Arrays and Agilent Human Whole Genome Arrays. Statistical analysis identified 1210 genes as signature genes characterizing the two subtypes of breast cancer. Protein function and biological pathway analysis on these signature genes revealed different molecular mechanisms descriptive of the two expression-based subtypes: signature genes of the Luminal A subtype were over-represented by genes involved in fatty acid metabolism and steroid hormone-mediated signaling pathways, in particular estrogen receptor-signaling, while signature genes of the Basal-like subtype were over-represented by genes involved in cell proliferation and differentiation, the p21-mediated pathway, and the G 1 -S checkpoint of cell cycle signaling pathways. Finally, we identified a minimal set of 59 predictor genes to best discriminate and characterize the Luminal A and Basal subtypes using PAM analysis on the combined data from the three array platforms. These predictor genes were further verified by TaqMan ® expression assays.
Conclusions We have identified and validated the two previously defined clinically relevant subtypes, Luminal A and Basal, in early stage breast carcinomas. This finding further substantiates the prognostic value of such expression-defined phenotypes in breast cancer at an earlier stage. Signature genes characterizing these two subtypes also revealed that distinct molecular mechanisms have been preprogrammed at an early stage in the different subtypes of the disease. Our results provide further evidence that these breast tumor subtypes represent biologically distinct disease entities and may require different therapeutic strategies. Finally, validated by multiple gene expression platforms, the set of 59 predictor genes identified in this study define potential prognostic molecular markers for breast cancer.
mechanisms that control lymph node metastasis, however, remain poorly understood. The aim of our study was to define patterns of genes or gene regulatory pathways that drive breast cancer lymph node metastasis.
Methods We compared the gene expression profiles of 15 primary breast carcinomas and their matching lymph node metastases using microarrays. Furthermore, we analyzed the expression profiles of two primary breast tumors and a metastasis obtained from the same patient.

Results
The gene expression profile of a primary breast carcinoma is more similar to its affiliated metastasis than the second primary tumor of the same patient. In general, primary breast carcinomas and lymph node metastases do not differ at the transcriptional level by a common subset of genes. However, subtle differences in the expression of genes involved in extracellular matrix organization and growth factor signaling are detected in individual pairs of matching primary and metastatic tumors. Surprisingly, however, different sets of these genes are either upregulated or downregulated in lymph node metastases.
Conclusions The overall gene expression profiles of primary breast carcinomas are maintained in their lymph node metastases. This similarity in gene expression can be attributed to tumor-intrinsic factors rather than to patient-specific factors. No common denominator for breast cancer lymph node metastasis could be identified, suggesting that breast carcinomas do not use a shared gene set to accomplish lymph node metastasis. Genome-wide expression profiles of currently 68 breast carcinomas, eight lymph node metastases, and eight normal breast samples were determined, and were correlated with detailed clinical outcome data to identify new and improved diagnostic marker genes of breast cancer. These marker genes were validated by real-time PCR as well as by IHC and/or FISH on tissue arrays. We identified a set of genes by which these patients could be classified into those with a short overall survival and those with a good prognosis with high accuracy (P < 0.0006). The prognostic markers thus identified included the estrogen receptor (ESR1), a known key prognostic marker gene of breast cancer. Defined sets of <60 genes each were identified that discriminated significantly (P < 0.01) between control patients and patients who develop lymph node, bone, or lung metastases, or other distant metastases. In contrast, >1000 genes were differentially expressed in primary tumors progressing to liver metastases, thus identifying a novel breast cancer subclass comprising ~15% of our cases. Downregulation of p14A and p14B, two homologous proteins of unknown function, was found to be associated with lung metastases. Tumors expressing high levels of p14A and p14B do not metastasize to the lung, tend to have a low tumor grade, are predominantly hormone receptor-positive, and are associated with a longer overall survival. This metastasis suppressor activity of p14A and p14B appears to be lung-specific, as no effect on liver metastasis and only a slight delay on bone metastasis formation were observed in tumors expressing high levels of p14A or p14B. We also identified ~75 genes coregulated with HER2, an RTK oncogene highly relevant to breast cancer therapy. Sixteen out of the 20 genes most closely co-expressed (P < 10 -5 ) are located in the same chromosomal region as HER2, suggesting that this entire region is co-amplified in ~20-30% of breast carcinomas. Furthermore, a potential key role in breast cancer progression of the PI3K/mTOR and the WNT signaling pathways was strongly suggested by our expression profiles. We found that stimulation of protein synthesis and cell growth via PI3K, mTOR, and eIF4E is the primary function of IGF signaling, and that activation of the WNT pathway in breast tumors significantly correlated with metastases and poor prognosis. Acknowledgement This work was supported by funds of the Austrian Ministry of Education, Science, and the Arts (Austrian Genome Research Program GEN-AU). Background Until now, microarray studies exploited the differences between cancer and corresponding normal tissues or the molecular differences between tumor histotypes originating from one tissue. However, a sound understanding of neoplastic transformation and progression will benefit from comparison of tumors originating from diverse tissues, especially if they share some biological or clinical properties. Such analysis may aid to seek novel therapeutic targets, which are tumor-specific rather than tissue-specific. The aim of our study was to compare the expression profile in breast cancer (BC) and ovarian cancer (OC), two female adenocarcinomas with similar genetic background and comparable chemosensitivity and radiosensitivity.

P4.30
Methods We compared expression profiles of 21 breast carcinomas and 17 serous ovarian carcinomas. We used the GeneChip U133 2.0 Plus microarray and a standard amplification procedure. We applied two methods of data preprocessing, RMA and MAS5 algorithm, and compared the results (only RMA data are shown).
Results Both preprocessing approaches resulted in a huge difference between BC and OC (4427 genes, False Discovery Rate lower than 0.1%). To base the comparison on well-described transcripts, we used the signature of neoplastic transformation proposed by Rhodes and colleagues [1] in a large meta-analysis of 40 cancer datasets. From 168 probe sets found on the U133 2.0 array that were corresponding to Rhodes genes, 30 were differentially expressed between BC and OC (the strongest differences were within KDELR2, PLK1, PPP2R5C, ACLY, G3BP, MMP9, TRA1, HSPD1) and the remaining 138 probe sets did not show differences in expression. The results confirm the Rhodes signature in BC and OC; however, these genes were not able to ensure the full subdivision of tumors into breast and ovarian (hierarchical clustering). Furthermore, we analyzed the tissue-specific expression of genes that were either uniformly or differentially expressed in BC versus OC, by comparison with normal tissues (data from GeneAtlas 2.0 [2]).
In the next step, we performed the unsupervised analysis of BC and OC expression profiles. By Singular Value Decomposition we revealed that the samples were divided into three large clusters, which corresponded to two groups of breast carcinomas (BC1 and BC2) and a separate group of ovarian cancers (OC). These groups were properly separated by expression of estrogen receptor probe set ESR1, which was low in BC1, showed variable and moderate expression in OC and showed very high expression in BC2. Methods Six datasets totalling about 700 primary breast cancers, mostly publicly available data, were used in the analysis. Gene expression profiles (GEP) from Affymetrix U133A GeneChips were contrasted between HG 1 (low grade) and HG 3 (high grade) tumours on a training set of 64 estrogen-receptor-positive breast cancer samples. A set of genes positively and negatively correlated with grade was identified on this training set and chosen as grade reporting genes. A scoring system called the 'gene-expression grade index' (GGI), which essentially summarizes the grade reporting genes by their average expression level, was introduced. The GGI was applied to patients not used in the gene selection to test its prognostic value. Results Using 33 HG 1 and 31 HG 3 ER-positive breast carcinomas, 112 Affymetrix probe sets were significantly upregulated in grade 3 and 16 in grade 1, at a stringent and objective cut-off P value of 0.05 for a false discovery count >0. These 128 probe sets represent 97 different reporter genes. Quantifying the level of expression of these reporter genes with the GGI, many tumors in the HG 2 (intermediate grade) populations assume values typical for the HG 1 and HG 3 groups in the same study. The HG 2 tumors can therefore be naturally split into a 'HG 1 like' group and a 'HG 3 like' group, to which we attribute a gene expression grade (GG) of 1 and 3, respectively. Their survival curves follow the GGI and are similar to those of the HG 1 and HG 3 groups, respectively, splitting HG 2 into a good prognosis group and a poor prognosis group (Fig. 1).
Similar observations were made in the different datasets analysed, in untreated as well as in systemically treated patients, and on the three different main types of microarray platforms, with substantial variability in the number of reporter genes available. Almost all known clinicopathological variables were significantly associated with clinical outcome in univariate analysis, while in a multivariate model only the GG, tumour size and nodal status were significant factors. Replacing the HG with the GG significantly improved the prognostic two-group classification obtained with the Nottingham Prognostic Index. Conclusion Gene-expression-based grading has the potential to significantly improve current grading systems by rendering them more objectively measurable and improving their prognostic value. The superior performance of the two-grade GG system challenges the purpose of classifying tumors as of intermediate grade. Reproduction of these findings in four independent datasets, and across different platforms and with a simple computational system, gives hope that the approach will prove robust and reliable. , each with a distinct molecular portrait [1]. Whole genome geneexpression analyses of independent sets of breast tumors have revealed repeatedly the robustness of this classification [2]. These patterns have clinical implications in terms of disease-free survival time and are always determined by the same set of genes in all datasets [3]. A list of 552 genes, whose expression in terms of mRNA varied considerably among the different tumors but little between two samples of the same tumor, has been nominated to be sufficient to separate these tumor subgroups. Why exactly these genes? What is the mechanism of their abnormal regulation? Genes are regulated by multiple transcription binding sites that interact with a specific S49 combination of transcription factors. Here we report the promoter composition of the genes that strongly predict the patient subgroups. Using a random expectation value (re-value) to generate a background model, we analyzed a total of 277 cis-elements (Genomatix software).

P4.34 Promoter composition predicts gene classes in microarray expression analyses of breast cancer
The gene classes showed a clear separation when based solely on their promoter composition. This finding suggests that studying those transcription factors associated with the observed expression pattern in breast cancers could identify novel and important biological pathways, including the NF-κB and Ets transcription factor families. Ovarian hormones and signaling cascades involving them play an important role in mammary tumor formation and progression in the female dog, as early ovariectomy may reduce the incidence of mammary carcinoma from 30% to less than 2%. The phenotypic expression of canine mammary tumors may vary, even within one affected animal. In the past, growth factor-independent canine mammary cell lines have been isolated [1]. Molecular characterization of these cell lines and correlation of their molecular signatures to biological behavior may lead to better understanding of signaling pathways involved in mammary cancer and classification of spontaneous mammary tumors. Methods Three canine mammary tumor cell lines (CMT) originating from primary mammary spindle cell tumor (CMTU309), primary mammary osteosarcoma (CMTU335) and primary mammary anaplastic carcinoma (P114) were compared directly with each other in this study. Cell lines were tested for in vitro invasion using a Transwell assay. Total RNA was isolated from cells grown to near confluence. In vitro transcription followed by labeling and hybridization to a cDNA microarray was carried out according to published protocols [2]. In a loop design of hybridization, labeled cRNA from cell lines were hybridized against each other on a dog-specific cDNA microarray containing 20,160 independent genes, which was developed and spotted in our laboratory. Statistical analysis of microarray data was carried out using significance analysis of microarrays [3]. Further analysis of microarray data was done using GeneSpring.  [1]. Also, BRCA2 show some specific alterations, but are more similar to sporadic breast tumors [2]. These results illustrate that breast tumors from different genetic backgrounds (BRCA1 and BRCA2) develop different genomic instabilities, and therefore genomic profiles. We hypothesize that this may also be true for BRCAx (BRCA3, BRCA 4, etc.) tumors. We therefore applied CGH to familial breast cancer cases from families without BRCA1/2 mutations. Aims To produce high-resolution profiles for various types of familial breast cancer, including BRCA1, BRCA2 and BRCAx. To build classifiers based on aCGH profiles. We further aim to optimize class discovery by parallel data analysis of continuous and discrete data as obtained by 'amplicon-finding' algorithms [3]. We also compare BRCA1/2 murine breast tumors with human tumors in an attempt to extract maximal biological meaning from the ploody changes observed in both species [4]. Methods Array-CGH was performed on genomic DNA isolated exclusively from formalin-fixed paraffin-embedded archival breast cancer specimens. Prior to hybridization, multiplex PCR was performed to assess DNA quality. constitute key genetic alterations in the cellular transformation of many tumors [1]. Microarray-based comparative genomic hybridization (array-CGH) allows the construction of high-resolution genome-wide maps of copy number alterations, and statistical software packages are available for exploring and analysing array-CGH data (see, for example, [2,3]), facilitating the delineation of the boundaries of CNAs in individual tumors and thereby localizing and identifying potential oncogenes and tumor suppressor genes. Although CNAs vary widely with respect to size and location, some genomic regions are known to have much higher prevalence of alteration than others. Mapping the location of these CNA hotspots facilitates location of genes of potential importance to tumor development as well as identification of alterations forming key steps in tumor development. There is, however, a need for consistent ways of combining array-CGH results for different arrays.

Results
Here, we present a statistical modelling-based approach for this. Methods Suppose we have available for each gene (clone) on an array a binary (0/1) variable indicating whether the gene is amplified or not. Such data may be constructed from array-CGH data using one of the aforementioned software packages. Each tumor may then be represented by an m-dimensional binary vector, where m is the number of genes on the array. For an experiment involving n tumors we thus have a set of m-dimensional vectors z 1 , …, z n and we consider the latter to be realizations from a multivariate distribution P(z). We consider three models for P(z) of increasing sophistication. The first assumes complete independence between genes, the second assumes a Markov-chain dependence structure and the third assumes a Markov Random Field dependence structure [4]. We demonstrate how P(z) can be estimated in each case and show that, by suitable constrained maximization of P(z), we may determine genomic intervals corresponding to probable occurring intervals of copy number alteration.

Results
The method is demonstrated (for all three models) on simulated binary copy number status data for varying number of genes and tumors. We also demonstrate the use on real array-CGH data that have been processed by CGH-Explorer [2] in order to obtain a binary copy number status vectors for each tumor.
Conclusion We have proposed a novel statistical method for the derivation of probable intervals of CNA, based on copy number status data from a sample of tumors. The method is based on a probabilistic model for the copy number status in a tumor, and we have discussed three models of increasing sophistication. The most basic of the three models corresponds to simply reporting all genes that are amplified in at least k% of the tumors. The other two models take into consideration the important fact that neighboring genes are not, in general, altered independently of each other. Utilizing this property of copy number data allows derivation of probable intervals of CNA that are less prone to noise degradation than alternative methods. In addition, results are derived in the context of a well-defined probabilistic framework and are therefore more easily interpretable.

P4.40
Comparative genomic hybridization using oligonucleotide arrays and total genomic DNA The broadest utility of aCGH is obtained by enabling flexible and highresolution probing of regions of interest while preserving the greatest possible complexity of targets derived from whole genome samples. We therefore developed probe design criteria, assay conditions, and analysis methods that enable 60-mer oligonucleotide arrays to be used for CGH measurements using total genomic DNA [1]. We designed a 60-mer oligonucleotide array with 40K probes specifically designed for CGH representing sequences throughout the human genome with a bias for known and predicted gene loci. We tested the performance of this array for reproducibly measuring and mapping losses, and amplification events of varying levels and sizes using both unamplified and phi29 (Qiagen, Valenica, CA, USA) amplified total genomic DNA from a series of model systems. The mean slope of experimental versus theoretical log-ratios for chromosome X probes on this genome-wide human CGH array in XY versus XX hybridizations typically exceeds 0.9, with probe by probe error rates of less than 10% in the separation of their log-ratio distributions. Additionally, we used this platform to examine well-characterized cell lines, including diploid cells with partial deletions in chromosome 18q, and diploid and aneuploid tumor cell lines with known amplification and deletion events. We show that the highly processive DNA polymerase phi29 can be used to prepare aCGH templates from as little as 10 ng starting material that yield highquality aCGH measurements throughout the genome. While phi29 provides a simplified isothermal method for amplifying limiting material, non-specific DNA fragments of high MW are generated in the absence of sufficient input template. Although these products do not hybridize to the array, the presence of these amplification products obscures the accurate quantification of DNA template specific to the input genomic DNA prior to the labeling reaction. To ensure reproducible and robust aCGH assay quality, we developed methods and protocols using the Agilent BioAnalyzer (Agilent Technologies, Palo Alto, CA, USA) to enable accurate quality control for key prehybridization steps, including: phi29 amplification of genomic samples, restriction digestion of templates and target labeling. We have also developed visualization tools and statistically robust computational tools that take into account the estimated errors on the measured log ratios in mapping aberration boundaries, and for identifying common aberrations across multiple samples. We tested the reproducibility of our platform using tumor cell line samples including the colon adenocarcinoma cell line HT29 in hybridizations performed in different laboratories ( [1]. Nguyen and Rocke [2] proposed first applying PLS in order to derive a small set of covariates, and then performing proportional hazards regression on the reduced set of covariates. In their approach, however, PLS is applied to survival times without taking into consideration the fact that the latter may be censored. A further problem with their approach is that the PLS step of their procedure is based on the assumption of a Gaussian (normal) likelihood. Methods Here, we propose a novel method for combining Cox proportional hazards regression and PLS. This method is a direct generalization of PLS to arbitrary likelihoods, whereas the original PLS method (including that used by Nguyen and Rocke) is designed for Gaussian likelihoods only. Furthermore, in our method PLS is directly integrated with the optimization of the Cox partial likelihood. Specifically, we propose to utilize the equivalence between PLS and a modification of the well-known numerical optimization method called the conjugate gradients (CG) algorithm: applying the modified CG algorithm to a Gaussian likelihood yields PLS. We propose instead to apply the modified CG algorithm to the Cox partial likelihood, hence directly generalizing the PLS algorithm to the Cox likelihood. Our method will take into account the censoring of the outputs, as only the original data will be used during the estimation. Our method also easily generalizes to other likelihoods than the Cox proportional hazards likelihood.

Results
We present results from the use of these methods for a dataset containing gene expression data and survival outcome from patients with breast cancer published by Sørlie and colleagues [3]. data were analyzed using 42,000 clone cDNA microarrays. Seventeen of these samples that showed 2+ or 3+ protein expression were evaluated by fluorescence in situ hybridization (FISH). Using microarray data from this cohort, we determined an RNA expression threshold associated with IHC/FISH positivity. Twenty-nine IDCs from Norway with TP53 mutation data and whose RNA expression of HER2 exceeded this threshold were added to the subsequent analysis, giving a total of 47 samples defined as HER2-positive. ANOVA correction was used to address differences in methodology and the data were analyzed with hierarchical clustering and disease-specific genomic analysis. Results At least three molecular subtypes of HER2-positive breast carcinomas were identified by hierarchical clustering of the 18 HER2positive samples based on the 42K array data. One subgroup contained tumors that overexpressed estrogen receptor (ER)-associated genes, another subgroup expressed cell-to-cell and cell-to-stroma signaling genes as well as varying amounts of the ER-associated genes, whereas a third subgroup showed no expression of the ER-associated gene cluster. These subgroups were confirmed when analyzing the 47 samples. TP53 mutation data were available for 32 of the 47 samples and 21 of these had a somatic TP53 mutation. We found less mutations in the ER group, with five mutations and seven wild type, than in the other two groups (four wild type and 16 mutated). This difference was significant (P = 0.034). Discussion The TP53 mutations seem to be differentially distributed among the molecular subtypes of HER2-positive tumors. Further studies are required to shed light on the implications of this finding. biological gain of function. In particular, mutant p53 exerts antiapoptotic effects. Likewise, NF-κB is a potent inhibitor of apoptosis, whose extended activation can promote cancer. We discovered that mutant p53 is in complex with the p65 NF-κB subunit in tumor cells treated with TNF, a potent inducer of NF-κB. In addition, we demonstrated that mutant p53 enhances the transcriptional activity of NF-κB and its anti-apoptotic efficacy. Moreover, we were able to show that mutant p53 and NF-κB are recruited together with the p300 acetyltransferase to anti-apoptotic target gene promoters. Interestingly, mutant p53 ablation attenuates the activity of NF-κB and renders cancer cells susceptible to killing by TNF. Finally, we observed a close correlation between the high frequency of p53 mutations and the elevated expression of NF-κB target genes in breast tumors. Therefore, our findings support an important role of NF-κB in mediating the oncogenic activities of mutant p53 in tumor cells.  [1,2]. In addition, LOH with retention of the Arg allele has been associated with reduced disease-free survival and overall survival [3].

Patients and methods
In the period January 1990-1994 a consecutive cohort of 204 Danish women were diagnosed with primary breast cancer. TP53 mutations were assessed by denaturing gradient gel electrophoresis analysis and DNA sequencing. The Arg72Pro polymorphism was measured in DNA extracted from blood lymphocytes and LOH was measured in DNA extracted from invasive breast carcinomas by a method including PCR, primer extension reactions and denaturing high-performance liquid chromatography analyses [4].  [2]. Genomic DNA is amplified by PCR, and dUTP is incorporated. The amplification products are then concentrated and purified with spincolumns. Amplification products are fragmented by Uracil N-glycosylase, and unincorporated dNTPs are inactivated by shrimp alkaline phosphatase. The fragmented PCR products are mixed with thermosequenase and four fluorescencelabelled ddNTPs. The sample mixture is transferred to a chip that contains sequence-specific oligonucleotides. So far, exons 2-9 are included on the array. Genorama™ QuattroImager is used for scanning. The Genorama imaging system and genotyping software are used for imaging and semiautomatic sequence analysis. DNA samples from 48 primary breast carcinomas, 11 ovarian carcinomas and 34 cell lines were used for evaluation. Results from a titration experiment with different ratios of the Arg/Arg and Pro/Pro alleles on codon 72 in the TP53 gene showed that mutations could be detected even if the mutated cells were present in less than 5%. We have experienced that homozygous and hemizygous mutations occasionally are missed by the TTGE technique, but that they all were easily detected by APEX [3]. Detection of deletions and insertions, however, is not yet optimal using the APEX technology and they are frequently missed. For the tumour samples the resequencing efficiency using APEX was 92% for both DNA strands and 99.5% for sense and/or antisense strands. The strength of using the APEX technology is that both strands are simultaneously analyzed, and that no further sequencing is needed. It is rapid and sensitive. Cost-effectiveness is still under evaluation.

P5.01
Detection of circulating cancer cells in peripheral blood as a prognostic factor in early breast cancer and adjuvant therapies. This fact has been attributed to early tumor seeding via the bloodstream. Therefore, identification of breast cancer cells in blood could enable early detection of micrometastases and could potentially be of prognostic significance. The aim of our study was to evaluate the correlation between circulating breast cancer cells and classical prognostic factors. Methods From February 1999 until April 2004 blood samples from 81 patients, aged 36-72, stages I and II, with or without metastases to regional lymph nodes, were collected every 3-6 months. All patients underwent therapeutic surgery and subsequent adjuvant therapies. Blood samples were screened for mRNA encoding hMAM, EGFR, CK-19 and β-hCG by nested RT-PCR. The result of the test was treated as positive when the expression of one or more mRNA markers was observed in at least two samples taken from the same patient. Clinical data, such as histological grade, pT-grade and pN-grade, menopausal status, estrogen receptor and progesterone receptor status, malignancy grade and others were available for analysis.

Results
We found that breast cancer cells were present in peripheral blood of patients even in very early stages of the disease. Any of the two-marker tests used in our experiments, hMAM/β-hCG, hMAM/ EGFR or EGFR/β-hCG, allowed comparable detection of breast cancer cells: in 68-74% of lymph node-positive (N1) and 49-59% of lymph node-negative (N0) patients. Addition of a third marker did not significantly increase detection sensitivity. While analyzing the presence of cancer cells in blood samples collected before mastectomy, we found cancer cells in the blood of only 13% of N0 patients, but in 46% of N1 patients. Thirteen of the 81 patients (16%) experienced recurrence of the disease within a 4-year follow-up period. We did not find any correlation between the presence of cancer cells and recurrence or any other clinical prognostic factors except one: patients with premenopausal status had cancer cells in the blood in 37% of cases, in comparison with those with postmenopausal status who had cancer cells in 63% of cases. Conclusion We have shown that a two-marker RT-PCR assay for hMAM/β-hCG, hMAM/EGFR or EGFR/β-hCG may be used for detection of occult breast cancer cells in peripheral blood. Except for menopausal status, we did not find any correlation between the presence of cancer cells in the blood and classical prognostic factors. Our data may suggest a different mechanism of disease dissemination in premenopausal and postmenopausal women. We investigated the prognostic significance of circulating breast cancer cells in peripheral blood detected by quantitative RT-PCR of marker genes in patients with advanced breast cancer. Blood samples from 94 breast cancer patients with metastatic disease (M1) were examined for circulating tumour cells by studying the mRNA expression of CK19, p1B, PS2, EGP2, mammaglobin and SBEM by real-time PCR. Using a score function, developed for predicting circulating tumour cells by quadratic discriminant analysis (QDA), four expression levels were combined into a single discriminant value. Tumour cells were present in 24 out of 94 (31%) of the patients, as compared with 0 out of 104 controls. The patients with a positive QDA value did have a progression-free survival at 1 year of 3% and overall survival at 2 years of 17%, against 22% and 36% for patients with a negative QDA value (P = 0.015 and P = 0.0053, respectively). Breast cancer patients with metastatic disease have a significantly worse progression-free survival and overall survival when circulating tumour cells can be detected in their peripheral blood [1]. This method was used but was not sensitive enough to predict survival on the basis of a positive score in peripheral stem cell preparations (PBSC) of patients treated in a high-dose chemotherapy trial. The number of tumour cells in these preparations is too low. We therefore sought a method that would specifically select breast tumour cells. We now show that a combination of magnetic beads cell separations and one-step cDNA synthesis of mRNA increases the sensitivity by at least 10-fold. Ten MCF7 cells mixed into 10 ml peripheral blood of a healthy control can be easily identified.
To validate this new method we will analyze blood samples of 200 stage I or stage II breast cancer patients and 50 PBSC of breast cancer patients (and 20 PBSC of patients with lymphoid malignancy as controls). So far, 30 stage I or stage II patients have been evaluated with the marker panel of six genes. A combination of four markers employed in the discriminant score revealed a sensitivity of 10% in detecting epithelial marker genes in peripheral blood. The cell separation technique allows at least a 10-fold increase in sensitivity to detect tumour cell specific mRNA in mixed cell suspensions. Further analysis of the 50 PBSC patients and the prospective 200-patient stage I and stage II diagnostic study will reveal specificity and sensitivity. Interim results indicate an increase in sensitivity of at least 10%.  Methods To identify a protein signature correlating with metastatic relapse, we performed surface-enhanced laser desorption/ionizationtime of flight mass spectrometry profiling of early postoperative serum from 81 high-risk EBC patients. Denatured serum samples were fractionated and the resulting fractions were incubated with ProteinChip arrays (Ciphergen Biosystems, Fremont, CA, USA).
Results Several protein peaks were differentially expressed according to clinical outcome (long-term metastasis-free survival versus metastatic relapse). By combining partial least squares and logistic regression methods, we built a multiprotein model that correctly predicted outcome in 83% of patients. Consistency and robustness of the model were verified using leave-one-out cross-validation. Five-year metastasis-free survival in 'good prognosis' and 'poor prognosis' patients as defined using the multiprotein index were strikingly different (83% vs 22%, P < 0.0001, log-rank test). In a multivariate Cox regression including conventional pathological factors and multiprotein index, only the latter retained independent prognosis significance for metastatic relapse. Major components of the multiprotein index were identified and included haptoglobin, C3a complement fraction, transferrin, apolipoprotein C 1 and apolipoprotein A 1 .
Conclusions Postoperative serum protein pattern may have an important prognostic value in high-risk EBC. In addition, it may reveal new insights on the metastatic process while providing new targets for future therapeutics.

P5.04
Monitoring of minimal residual cancer in bone marrow in high-risk breast cancer patients treated with high-dose chemotherapy  [1] were randomised to nine cycles of dose-escalated FEC (5-flurouracil, epirubicin, cyclophosphamide) or three cycles of standard FEC followed by high-dose chemotherapy. Bone marrow (BM) samples at diagnosis and 6 months after completion of chemotherapy were assessed for the presence of cytokeratin-positive (CK+) cells. CK+ cells in BM were evaluated as a prognostic and predictive marker and were compared with other defined prognostic factors of the primary tumour.

Results
Monitoring BM changes at time of diagnosis and at 6 months post-treatment is an independent predictive factor for breast cancerspecific survival (P = 0.001, univariate analysis). Those who have consistent CK-negative BM findings constitute a group of patients with good prognosis. Conclusion Monitoring of CK+ cells in BM before and after high-dose chemotherapy with or without stem cell support can be used clinically as a surrogate maker to predict outcome in breast cancer patients. . Hypoxic cells are highly aggressive and metastatic, although the underlying processes remain unclear. We have found lysyl oxidase (LOX) expression to be increased by hypoxia in a variety of human cancer cell types. Lysyl oxidase (LOX) plays an essential role in the formation and maintenance of the extracellular matrix, and has previously been linked to increased in vitro invasion of breast cancer cells [1]. Methods Human breast and cervical cancer cells were deprived of oxygen for 18 hours and the expression levels of LOX were examined by RT-PCR (including quantitative), northern blotting and western blotting. LOX activity was inhibited by antisense or siRNA treatment or by addition of a chemical inhibitor, and the effect on in vitro invasion was examined using Boyden chambers. The in vivo metastatic potential of these cells was also examined in air and hypoxia via tail-vein injection of mice subsequently housed for 4 weeks under 20% or 10% oxygen and assessment of lung micro metastases. Previously published microarray datasets were examined for correlation between LOX expression and metastasis in human breast cancer patients [2,3]. Results Incubation of human breast and cervical cancer cells in oxygen-deprived conditions resulted in elevated levels of LOX due to a hypoxia inducible factor 1-dependent increase in mRNA levels.

P5.05 Hypoxia promotes invasion and metastasis of breast cancer cells by increasing lysyl oxidase expression
Oxygen-deprived cells demonstrated enhanced in vitro invasion that could be blocked by transfection with LOX antisense oligonucleotides or LOX-specific siRNA, or by treatment with an inhibitor of LOX activity. Cells stably expressing LOX siRNA grew slightly faster in air but demonstrated non-invasive and non-metastatic phenotypes in threedimensional culture, and formed dramatically fewer lung micro metastases in vivo when injected into mouse tail veins, particularly those housed in hypoxic conditions. Analysis of expression data from breast cancer patients revealed a good correlation between LOX and lymph node status (Pearson correlation value of 0.78). Conclusion Our data reveal that hypoxia-induced LOX plays a key role in invasion and metastasis in human breast (and cervical) cancer, and that inhibition of LOX blocks these processes and may enhance effectiveness of therapy. These novel findings suggest that LOX may represent a novel marker of patient prognosis, particularly as an indicator of lymph node status in breast cancer.

P5.09
Brn-3b transcription factor in breast tumourigenesis: regulation of genes associated with growth and migration of cancer cells SA  The Brn-3b transcription factor is elevated in many breast cancers compared with levels found in normal breast epithelial cells. High levels of Brn-3b increase growth and proliferation of cancer cells, both in vitro and in vivo, but also alter migration and confer resistance to growth inhibitory stimulus [1]. Conversely, low levels of Brn-3b slow the growth of these cells. As a transcription factor, Brn3b changes the growth and behaviour of breast cancer cells by modifying the expression of target genes, either directly or indirectly upon association with the proliferation-associated estrogen receptor (ER) [2]. A number of Brn-3b target genes have been identified that alter the growth and behaviour of these cancer cells. For instance, Brn-3b transactivates the promoter of the cyclin-dependent kinase, CDK4 [3], that is required for cell cycle progression and hence proliferation. Brn-3b also represses the promoter of the tumour suppressor gene, BRCA1, and inversely correlates with BRCA1 protein in tumour biopsies. We have recently demonstrated that the small heat shock protein, HSP27, is also regulated by Brn-3b [4]. High expression of HSP27 in breast cancers is associated with increased anchorage-independent growth, increased invasiveness and resistance to chemotherapeutic drugs and poor prognosis. Thus, in cancers expressing high levels of Brn-3b the downstream target genes regulated by this transcription factor can alter the growth and behaviour of these cells.
Methods Western blot analysis of tumour samples was used to correlate Brn-3b and HSP27 proteins. Transient co-transfection and reporter assays were used to look at the effects of Brn-3b and/or ER on the HSP27 promoter. Short hairpin RNA interference was used to target Brn-3b protein and to test its requirement for transactivation of the HSP27 promoter. EMSA was used to demonstrate direct binding of Brn-3b to a specific site in the HSP27 promoter. Chromatin immunoprecipitation (ChIP) was performed to show that Brn-3b was associated with the HSP27 promoter in intact cells.
Results Brn-3b protein levels correlated strongly with HSP27 levels in a significant number of breast cancer biopsies (R = 0.87) as well as in the breast cancer cell line, MCF7. Overexpression of Brn-3b in MCF7 cells resulted in increased HSP27 protein levels while reducing Brn-3b proteins using antisense correlated with decreased HSP27 compared with controls. Co-transfection analysis using a HSP27 reporter construct showed that Brn-3b could directly transactivate HSP27 promoter but cooperated with the ER for maximal expression. Decreasing Brn-3b using targeted RNA interference prevented activation of the HSP27 promoter by Brn-3b alone but also attenuated the response by ER. The Brn-3b site in the HSP27 promoter is flanked by two sequences that constitute half estrogen receptor elements. Site-directed mutagenesis demonstrated that this DNA sequence was required for maximal transactivation while the ChIP assay showed that Brn-3b protein binds to the HSP27 promoter in vivo.
Conclusion The ability of Brn-3b to increase expression of proteins such as HSP27 in cancer cells may help to understand the altered growth and migration of tumour cells with elevated Brn-3b proteins. Therefore factors such as Brn-3b, which increase the expression of HSP27 in breast cancers, are likely to affect the progression of diseases, prognosis and outcome of treatment.  [1]. These subtypes are believed to develop from different epithelial cell types and show different overall survival outcomes. Of particular interest is the estrogen receptor (ER)-negative Basal-like subtype, which accounts for 10-15% of all breast tumors and shows poor outcomes. In the breast cancer clinic, there are currently two biologically directed therapies that target either the ER or HER2 proteins. The Basal-like tumors lack both of these proteins [2], and hence the only treatment options for these patients are cytotoxic chemotherapies. A goal of ours was therefore to use primary breast tumor gene expression data and cell line models to identify and validate candidate biologically-based therapies for Basal-like tumors.

P6.01 Identification of drug targets for the treatment of
To identify potential targets, the gene expression data for approximately 1500 drug targets were examined across a breast tumor data set of 150 samples. Squalene epoxidase (SQLE) was expressed in most Basal-like tumors, as well as in the Basal-like tumor-derived cell lines SUM102 and SUM149. SQLE is an attractive target because it is highly expressed, it is a rate-limiting step in the cholesterol biosynthetic pathway, and there is an available inhibitor (NB598) [3]. Recent studies using inhibitors of HMGCoA reductase (the first rate-limiting step) in epithelial cell lines suggest that inhibition of this pathway may be a potential target for therapeutic intervention [4].
Using the SUM102 and SUM149 cell lines and two more widely used luminal/ER+ lines (MCF-7 and ZR-75-1), we treated cells with NB598 and separately with lovastatin (an HMGCoA reductase inhibitor) and determined their sensitivity by identifying their 72-hour IC 50 dose. Sensitivity was similar across three of the four cell lines for NB598, with the exception of SUM102, which was approximately 300 times more sensitive. Conversely, sensitivity to lovastatin was similar across three of the four cell lines except MCF-7, which was approximately five times more resistant. Since many drugs are rarely used as single agents, we also looked at the interactions between these two inhibitors and commonly used chemotherapeutics. Drug-combination sensitivities again varied across the four cell lines; however, it appears that combinations of NB598 and 5-fluorouracil were typically synergistic, while combinations with carboplatin or paclitaxel were typically antagonistic. Similar analyses are being performed for lovastatin/ chemotherapy combinations. Gene expression responses of these cell lines were also assayed using DNA microarrays. The effect on the cholesterol pathway showed that, for MCF-7 and SUM102, adding either inhibitor greatly induced most genes in the cholesterol biosynthetic pathway, while SUM149 treated with lovastatin showed induction of the pathway but treatment with NB598 did not. ZR-75-1 treated with either drug showed a slight reduction in expression of the pathway. These in vitro data suggest that inhibition of SQLE activity can reduce cell line proliferation rates and, in some instances, was synergistic with chemotherapy. These data also suggest that inhibition of the cholesterol pathway by addition of HMGCoA reductase inhibitors is different from inhibition of the pathway with SQLE inhibitors.

P6.02
Determining the factors affecting breast cancer infectivity by oncolytic adenovirus Genetically modified adenoviruses -specifically, adenovirus serotype 5 (Ad5) -are commonly used to generate oncolytic viruses. These adenoviruses are replication-selective, meaning that they have been engineered to replicate only in cancer cells bearing certain mutations. For example, ONYX-015 is a mutant adenovirus designed to exploit the loss of functional p53, a loss common to many cancer cells, in order to selectively destroy malignant cells [1]. Although the use of oncolytic viruses holds great promise for cancer therapy, the success of this strategy depends on the ability of adenovirus to infect cancer cells. We are using a panel of 50 breast cancer cell lines to study Ad5 infectivity. Affymetrix array data and CGH data have been collected for all of these cell lines. We have found that the ability of Ad5 to infect these cell lines is highly variable. CAR and αv integrins are known to be required for Ad5 entry. However, the infectivity of the breast cancer cell lines does not correlate with CAR levels or αv integrin levels. The treatment plan for a breast cancer patient is based on tumor size and grade, lymph node involvement and steroid hormone receptors. Lymph node status is the strongest prognostic factor for breast cancer patients. About 25% of node-negative patients experience recurrence or metastasis [1]. Additional methods might be important for better treatment strategies. Malignant cells have an altered metabolism, and metabolic mapping might become a tool in cancer diagnostics. High-resolution magic angle spinning (HR-MAS) magnetic resonance (MR) spectroscopy of tissue biopsies provides detailed information on their metabolic composition [2]. The aim of this study was to compare MR spectroscopic findings from breast cancer tissue with histological grading of tumor and patient lymph node status. Methods Breast cancer and non-involved adjacent tissue were excised from patients with palpable breast cancer diagnosed as invasive ductal carcinoma (IDC). Tissue specimens were analyzed in D 2 O-PBS in a 50 µl MAS rotor (4 mm o.d.). HR-MAS MR spectra were recorded on a BRUKER AVANCE DRX600 spectrometer at 4°C. The samples were spun at 5 kHz. Proton MR spin echo spectra were acquired with a total echo time of 285 ms and presaturation of the water peak. A pathologist scored the relative areas of normal and neoplastic elements visually after MR analysis. Samples with less than 5% tumor content were excluded. This resulted in a final database consisting of 115 samples:

Results
The 2D score plot of PC2 and PC3 from the PCA of all samples is shown in Fig. 1. All samples from non-involved tissue are clearly separated from tumor samples. Tumor samples intersperse with no possibility to differentiate among the three types of grading. The PNN of spectra from tumor samples resulted in true classification of 56 of the 69 samples with respect to grading, whereas two samples were not classified. The specificity and sensitivity of classification exceeded 80% for all groups.
A PCA score plot of PC2 and PC3 for tumor samples is shown in Fig.  2. A trend of clustering with respect to lymph node status can be seen. Classification results of node-positive and node-negative samples using PNN is presented in Table 1. Samples from patients with spread of cancer cells to lymph nodes can be predicted with a specificity of 97% and a sensitivity of 92%.
Conclusion PCA led to a complete separation of the non-involved and cancerous samples. The metabolism of cancerous tissue is clearly different from non-involved tissue. Samples from lymph node-positive Breast Cancer Research Vol 7 Suppl 2 Third International Symposium on the Molecular Biology of Breast Cancer

Figure 1
Principal component analysis score plot of all samples (n = 115). Labeling of samples from patient diagnosis.

Figure 2
Principal component analysis score plot of tumor samples (n = 69). Labeling of samples from patient lymph node status.

S61
and lymph node-negative patients could not be separated by PCA, while PNN led to classification of the two groups with misclassification of only four samples. Metabolic patterns in breast tumors from patients with lymphatic spread differ from those without lymphatic spread. These findings show that HR-MAS of breast cancer biopsies has the potential of becoming a diagnostic tool.