Positive Breast Cancer Cells by the Expression Profile of an Intrinsic Set

JOURNAL OF CELLULAR PHYSIOLOGY 200:440–450 (2004)

Molecular Identiﬁcation of ERa-Positive Breast Cancer Cells by the Expression Proﬁle of an Intrinsic Set of Estrogen Regulated Genes

ALESSANDRO WEISZ,1* WALTER BASILE,1 CLAUDIO SCAFOGLIO,1 LUCIA ALTUCCI,1 FRANCESCO BRESCIANI,1 ANGELO FACCHIANO,2 PIERO SISMONDI,3,4 LUIGI CICATIELLO,1 3,5 AND MICHELE DE BORTOLI 1Dipartimento di Patologia Generale, Seconda Universita` Degli Studi di Napoli, Vico L. De Crecchio 7, Napoli, Italy 2Istituto di Scienze dell’Alimentazione del Consiglio Nazionale delle Ricerche, Avellino, Italy 3IRCC-Institute for Cancer Research & Treatment, University of Turin, Candiolo (TO), Italy 4Department of Gynecology & Obstetrics, University of Turin, Candiolo (TO), Italy 5Department of Oncological Sciences, University of Turin, Candiolo (TO), Italy

Estrogens exert a key biological role in mammary gland epithelial cells and promote breast carcinogenesis and tumor progression. We recently identified a new large set of estrogen responsive genes from breast cancer (BC) cells by DNA microarray analysis of the gene expression profiles induced by 17b-estradiol in ZR-75.1 and MCF-7 cells. The purpose of the present study was to test whether the expression pattern of hormone regulated genes from this set identifies estrogen receptor (ERa) positive, hormone responsive BC cells. To this aim, we carried out in silico metanalysis of ERa positive and ERa negative human BC cell line transcriptomes, focusing on two sets of 171 and 218 estrogen responsive genes, respectively. Results show that estrogen dependent gene activity in hormone responsive BC cells is significantly different from that of non-responsive cells and, alone, allows to discriminate these two cellular phenotypes. Indeed, we have identified 61 genes whose expression profile specifically marks ERa positive BC cells, suggesting that this gene set may be exploited for phenotypic characterization of breast tumors. This possibility was tested with data obtained by gene expression profiling of BC surgical samples, where the ERa positive phenotypes were highlighted by the expression profile of a subset of 27 such hormone responsive genes and four additional BC marker genes, not including ERs. These results provide direct evidence that the expression pattern of a limited number of estrogen responsive genes can be exploited to assess the estrogen signaling status of BC cells both in vitro and ex-vivo. J. Cell. Physiol. 200: 440–450, 2004. ß 2004 Wiley-Liss, Inc.

Breast cancer (BC) remains a serious threat to the lump. Since no marker is available today to distinguish patients, despite signiﬁcant improvements in its clinical cancers most likely to recur, there is today a tendency to management (Harris et al., 2000). The most informative prescribe adjuvant treatments to all patients. The kind prognostic factor for primary BC is the presence of of pharmacological treatment is based on clinico- lymph node metastases: around 70% of node-negative pathological criteria and on the expression of biological patients can be expected to survive BC without markers in tumor cells, since lymph-node metastasis per additional treatments beyond surgical resection of the se is not predictive of sensitivity or resistance of the

Contract grant sponsor: Italian Association for Cancer Research (to AW and PS); Contract grant numbers: IG 2002, IG 2003; Contract grant sponsor: Italian Ministry for Education, Univer- sity and Research (PRIN 2002-03 and FIRB Post-genoma Grant to AW and MDB); Contract grant numbers: 2002067514_002, RBNE0157EH; Contract grant sponsor: European Commission (contracts to AW and LA); Contract grant numbers: BMH4-CT98- *Correspondence to: Alessandro Weisz, Dipartimento di Patologia 3433, QLG1-CT-2000-01935, QLK3-CT-2002-02029; Contract Generale, Seconda Universita` degli Studi di Napoli, Vico L. De grant sponsor: Ministry of Health; Progetti Speciali 2000 and Crecchio, 7, 80138 Napoli, Italy. 2002 (to AW, MDB, and PS); Contract grant sponsor: Second E-mail: [email protected] University of Naples; Ricerca di Ateneo 2002 and 2003 (to AW, FB, and LA); Contract grant sponsor: Regione Piemonte, Ricerca Received 31 July 2003; Accepted 3 December 2003 Sanitaria Finalizzata 2002, 2003 (to PS). DOI: 10.1002/jcp.20039

ß 2004 WILEY-LISS, INC. MOLECULAR TYPING OF ERa POSITIVE BREAST CANCER CELLS 441 disease to any of the adjuvant treatments available 2002). It is not clear, however, whether such patterns (Lønning et al., 2001). are indeed related to the estrogen signaling status of the Adjuvant therapy is the most effective weapon tumor or, instead, if they reflect its differentiation available today against recurrence and systemic spread status. of BC. Pharmacological treatments, involving the use of To help address this problem, we speculated that the non-selective cytotoxic treatments and/or endocrine identification of a set of hormone responsive genes in therapy, reduce the recurrence rate and the risk of human BC cells could provide specific markers to distant metastases. Similar pharmacological treat- monitor the hormone signaling status of tumors. ments are also used in the advanced disease. Endocrine Expression profiling of a specific set of estrogen therapy, based on systemic estrogen ablation by drugs responsive genes, in fact, would allow focusing the affecting hormone synthesis, metabolisms or actions in analysis on the hormone dependent transcriptome of BC target cells, is very effective on selected patients and cells. We present here evidence that a subgroup of generally well tolerated, both in the adjuvant setting hormone-responsive genes is indeed sufficient to iden- and in the metastatic disease. Hormone responsive tify ER-positive BC cells, either in culture or from tumor breast tumors are currently identified by the presence of biopsies. This is proof of concept that hormone-depen- ERs (ERa and in some cases also ERb) and of the dent gene expression profiles can identify ER pathway estrogen regulated progesterone receptors (PgRs). Pre- activation in tumors and thus may be used to predict diction of BC responsiveness to hormones, based on hormone-responsiveness in BC. current assays for ERa expression in tumor cells can, however, be misled by the existence of different MATERIALS AND METHODS molecular situations in which the ER is present but Selection of the starting set of inactive (Ferguson et al., 1998; Hopp and Fuqua, 1998). estrogen-responsive genes Indeed, comprehensive clinical trials have shown that The starting set of estrogen responsive genes used for about one third of ER/PgR-positive tumors fail to all the analysis was derived from our microarray study respond to endocrine therapy, while about 10% of of the genomic response of human MCF-7 and ZR75.1 receptor-negative ones do respond (Harris et al., 2000). BC cell lines to estrogen treatment (Weisz et al., 1999; Receptor-positive tumors are thus heterogeneous with Cicatiello et al., 2001; Cicatiello L, Natoli G, Scafoglio C, respect to estrogen sensitivity and, for this reason, Altucci L, Cancemi M, Facchiano A, Calogero R, Iazzetti functional markers of hormone dependent signaling or G, De Bortoli M, Sfiligoi C, Sismondi P, Biglia N, ERs function in BC cells are being actively searched for Bresciani F, and Weisz A. The gene expression program (Biswas et al., 1998; Bouras et al., 2001; Finlin et al., activated by estrogen in hormone responsive human BC 2001). Identification of such markers is complicated, cells. Submitted for publication elsewhere). This set was however, by the fact that hormone receptors expression produced using cDNA microarrays containing probes for in BC cells more often is associated with cell differentia- 8,372 unique human genes/EST and allowed to identify tion and low proliferation rates. 344 genes whose expression is either increased or Gene expression profiling is a promising new way to decreased after estrogen stimulation. classify BC in clinically relevant subtypes according to molecular ‘signatures’ of tumor biopsies (Martin et al., Expression data extraction and evaluation 2000; Perou et al., 2000; Sørlie et al., 2001; West et al., The first set of data was derived from the NCI60 gene 2001; Bertucci et al., 2002; van’t Veer et al., 2002). To expression database for the molecular pharmacology of provide the logical framework required for interpreting cancer (Ross et al., 2000) that includes the data relative to the biological significance of the gene expression the expression of approximately 8,000 genes in 60 cell patterns observed in tumor samples, however, the lines of different tissue origin. Seven BC cell lines are identification of the transcriptional programs associated present in this database, 2 ERa positive and estrogen with BC cell responses to relevant physiological stimuli responsive (T-47D and MCF-7) and 5 ERa negative, is much sought after. Perou et al. (1999), for example, hormone unresponsive (BT-549, Hs578T, MDA-MB-231, found that clusters of coexpressed genes identified MDA-MB-435, and MDA-N, an ERBB2-transfected sub- through manipulations of mammary epithelial cells clone of MDA-MB-435 cells). For comparison, data in vitro, including stimulation with growth factors and relative to one normal breast tissue biopsy (normal cytokines, showed consistent patterns of variation in breast) and 16 human cell lines from brain (SF-268 and expression among BC specimens. Also, Ross et al. (2000), U251), colon (COLO205 and HCT-116), kidney (A498 by comparing the gene expression patterns of BC and 786-0), lung (A549 and NCI-H322), ovary (OVCAR-4 biopsies and breast and mesenchymal cell lines found and -8), and prostate (DU-145 and PC-3) cancer, specific gene expression features in tumors which melanoma (SKMEL-5and -28), chronic myeloid leukemia appeared to be related to what observed in vitro in cell (K-562), and multiple myeloma (RPMI-8226) were lines, where specific gene expression signatures could be considered. The microarray data relative to each of these linked to defined biological properties of the cells, such cell lines were downloaded from the NCI60 Cancer Micro- as doubling time, drug metabolism, or response to array Project web site (http://genome-www.stanford.edu/ interferon stimulation. Gene expression studies led in nci60/). Each dataset included the normalized fluores- fact to the identification of cluster of genes whose cence data obtained by co-hybridization of Cy5-labelled expression pattern appear to relate to the presence of cDNA reverse transcribed from mRNA isolated from the ERa in the tumor (Perou et al., 2000; Dressman et al., cell line under study and Cy3-labelled cDNA reverse 2001; Gruvberger et al., 2001; Sørlie et al., 2001; West transcribed from a reference mRNA sample, which et al., 2001; van de Vijver et al., 2002; van’t Veer et al., consisted of a mix of mRNA extracted from 12 cell lines 442 WEISZ ET AL. of different histotype, chosen to maximize diversity in round of testing, of the resulting correlation coefficients gene expression. Thanks to this common internal for cell line clustering. reference, relative variations in gene expression across all cell lines can be inferred directly from the changes of Representation of results normalized Cy3/Cy5 ratios (Ross et al., 2000). On the The visualization software Tree-View (Eisen et al., other hand, these fluorescence ratios do not represent the 1998) was used to graphically display. In the output, absolute mRNA expression value in a given cell line. expression data are colored using a green/red color code, Following elimination of low quality spots (FLAG > 0) as shown in Figure 1, were red colors represent genes and spots characterized by hybridization signals below a whose expression in the sample were lower than in the threshold, set to 1/10th the average fluorescence data reference sample (pool of 12 lines, see: Perou et al., (Cy3 Cy5) of the array, the expression data relative to þ 2000). 171 of the 344 estrogen-responsive genes mentioned above were found in the NCI60 database and used for further analysis. The cDNA probes for 28 such genes were RESULTS present in duplicate or triplicate and allowed control for Typing of human breast cancer cell lines according reproducibility of the measurements. An internal control to the activity of the estrogen-responsive was also provided by the presence of triplicate measure- transcriptome ments relative to two cell lines (MCF-7 and K-562) that We have recently characterized the hormone depen- could be compared. The second set of experimental data dent transcriptome of BC by cDNA microarray analysis on cell lines was obtained from the ‘Molecular Portraits of the gene expression changes induced by a physiolo- of Human Breast Tumors’ web site (http://genome- gical dose of 17b-estradiol in hormone responsive www.stanford.edu/breast_cancer/molecularportraits/). human BC cells (Weisz et al., 1999; Cicatiello et al., These data had been generated in a microarray study 2001; Cicatiello L, Natoli G, Scafoglio C, Altucci L, that focused mainly on analysis of surgical BC speci- Cancemi M, Facchiano A, Calogero R, Iazzetti G, De mens (Perou et al., 2000), but included also three Bortoli M, Sfiligoi C, Sismondi P, Biglia N, Bresciani F estrogen-responsive (BT-474, MCF-7, and T-47D) and and Weisz A. The gene expression program activated by four-unresponsive (BT-549, Hs578T, MDA-MB-231, and estrogen in hormone responsive human BC cells. SK-BR-3) BC cell lines, two normal mammary epithelial Submitted for publication elsewhere). This work led so cell lines (HMEC and 184), analyzed under different far to the identification of a set of 344 genes whose culture conditions, and five mesenchymal-like cell lines expression is either increased or decreased in ZR-75.1 (HUVEC, MOLT4, NB4, RPMI-8226, SW872). Data and/or MCF-7 cells upon estrogen stimulation. This concerning the expression of >8,000 human genes were gene set, that includes a large series of newly identified available in this database. Proceeding as above described, hormonal targets but does not include not ERa (ESR1), 218 informative genes were found among the 344 ERb (ESR2), or PgRs (PGR), provides the opportunity to estrogen-regulated gene set, including 129 in common test whether the expression pattern of hormone respon- with those of the NCI60 database and 52 present in sive genes represents a molecular signature specific for replicate. estrogen dependent BC cells, and thus may allow to The dataset relative to breast tumor biopsies were identify them independently from other molecular extracted from the same database (http://genome- markers used to date, including estrogen and PgRs www.stanford.edu/breast_cancer/molecularportraits/). themselves. Here, the data relative to a comprehensive gene To this aim, as a first step we explored the NCI60 gene expression analysis performed in 62 tumor specimens expression database for the molecular pharmacology of from 40 patients, comprising also 22 sample pairs from cancer (Ross et al., 2000), containing the expression data the same tumor, obtained either before and after relative to approximately 8,000 genes in 60 cell lines of neoadjuvant chemotherapy or from the primary lesion different tissue origin. The data relative to two ERa and its lymph node metastasis (Perou et al., 2000). positive and five ERa negative human BC cell lines, present in this data base and to one normal breast tissue Clustering analysis biopsy and 16 human cell lines of different origin, were The log-transformed expression data relative to extracted (see Materials and Methods). Following a estrogen-responsive genes from either data set were round of selection, based on the quality of the individual analyzed by unsupervised two-dimensional hierarchical spots, the expression data relative to 171 out of the 344 clustering (Eisen et al., 1998). This algorithm produced estrogen-responsive genes of our set were found in the a table of results, with the array elements—represent- NCI60 database and used for further analysis. cDNA ing specific genes—on the first dimension, grouped probes for 28 such genes were present in duplicate or together according to similarities in their pattern of triplicate and data relative to two cell lines (MCF-7 and expression in the different samples, while clustering K-562) were present from three independent analyses, together on the ‘second’ dimension the different samples offering an internal validation of the data. Unsuper- under analysis (cell lines and tumors) according to vised two-dimensional hierarchical clustering analysis similarities in their overall gene expression patterns. (Eisen et al., 1998) of log-transformed expression data, Search for the best clustering subsets was carried out by relative to these 171 estrogen-responsive genes, calculating the cell line clustering efficiency relative to produced the results that are illustrated in Figure 1, individual gene clusters identified during the initial as colored matrices. The data relative to each cell line clustering analysis, and combinations thereof, followed analyzed were displayed as a column and the data by iterative exclusion tests and calculation, at each relative to each gene disposed as a row. All columns were MOLECULAR TYPING OF ERa POSITIVE BREAST CANCER CELLS 443

Fig. 1. Cluster analysis of estrogen responsive gene expression in scales to the right of each dendrogram mark the correlation coefficient normal and cancer cell lines from breast and other tissues. Two-way represented by the length of the dendrogram branches connecting hierarchical clustering analysis was applied to characterize ERa pairs of nodes. Part a: Clustering analysis based on relative expression positive (red) and negative (purple) BC cell lines, normal mammary data of 171 estrogen responsive genes from the NCI60 set (top-right gland tissue and cell lines (brown), and non-BC cells (black) based on and left) or of 49 genes selected among them during this study the expression pattern of sets of genes selected on the basis of their (bottom-right). Part b: Clustering analysis based on relative expres- estrogen responsiveness in ZR-75.1 and MCF-7 cells. Each column of sion data of 218 estrogen responsive genes from the ‘Molecular the expression matrixes represents the cell line/tissue sample Portraits of Human Breast Tumors’ set (top-right and left) or of 19 indicated at its top and each row refer to a gene, colors of the matrix selected genes (bottom-right). In the smaller matrixes, the expression elements represent mRNA expression levels relative to a common patterns of cytokeratin 5 (KRT5), 8 (KRT8), and 17 (KRT17), integrin reference sample (green for sample/reference ratios <1, red for ratios b4 (ITGB4), laminin g1 (LAMC1), GATA binding protein 3 (GATA3), >1, black for ratios near 1, and gray for missing data). Dendrograms estrogen regulated LIV-1 protein (LIV-1) and/or ERa (ESR1) genes, representing hierarchical relationships between cell lines have the which where not part of the clustering analysis, are included for terminal branches colored to reflect the nature of the cell line; the reference. 444 WEISZ ET AL. thus ordered along the horizontal axis so that cell lines expression profile is almost the same in the two estrogen with the most similar patterns of expression across all responsive BC cells (CC MCF-7 vs. T-47D: 0.81) but genes analyzed are placed adjacent to each other. Along clearly differs among them and non-responsive cells the vertical axis, all rows are instead ordered according (CC: 0.75). For comparison, this part of the figure to similarities in the expression pattern of the corre- includes also a color matrix showing the relative sponding genes in all cell lines. Similarities between cell expression of several markers of breast epithelial cell lines and genes, respectively, are measured by standard lineage (see Discussion). correlation analysis, whose results are visualized above The possibility of identifying estrogen responsive BC and sidewise the color matrix by ‘correlation trees.’ In cell lines based exclusively on expression of hormone these trees, each node represents a calculated correla- regulated genes was controlled by repeating the cluster- tion coefficient value (CC), ranging from þ1.00 (max- ing analysis using a second group of 218 genes, among imum degree of correlation between samples/genes) to our estrogen-regulated gene set, whose expression data 1.00 (lack of correlation). The results obtained from the was available in the ‘Molecular Portraits of Human NCI60 data set are displayed in Figure 1a, where the BC Breast Tumors’ web site (Perou et al., 2000). This site cell lines responsive to estrogen are depicted in red, contains the cDNA microarray data obtained on three those unresponsive in purple, the Normal Breast sample ER-positive and four ER-negative BC cell lines, two in brown, and the remaining, non-BC cell lines in black. normal mammary epithelial, and five mesenchymal-like The correlation tree relative to the clustering analysis of cell lines, along with data relative to several breast all cell lines selected is reported at the top-right of the tumor biopsies (see above). Of these 218 informative figure. The highest correlation among cell lines (CC: genes, 129 were in common with those from the NCI60 0.92–0.95) was observed for MCF-7 and K-562 cell set and 52 had been analyzed using replicate probes. The replicates, respectively, and for the MDA-N clone and results of hierarchical clustering analysis of these data the parental cells MDA-MB-435, indicating good repro- are summarized in Figure 1b (top-right). Even in this ducibility of the mRNA profiling results as well as case, all estrogen responsive BC cell lines clustered reliability of the data analysis performed here. The cell together and with SK-BR-3 cells in a cluster that is lines analyzed are grouped in two main clusters, which clearly distinguishable (CC: 0.01) from a second cluster display the lowest degree of correlation observed (CC comprising all remaining cell lines. This clustering 0.07). The estrogen-responsive BC cell lines are in the pattern becomes even more evident upon exclusion from left cluster, where they are disposed sidewise to each the analysis of the data relative to the mesenchymal cell other and clearly separated from two non-BC lines lines (left part of Fig. 1b). Interestingly, in the right (COLO205 and OVCAR-4) and the Normal Breast cluster normal and cancerous mammary epithelial cells sample. The closest correlation observed in the left appears now clearly distinct, with the former lines cluster (CC: 0.55) was among the MCF-7 cell replicates showing more similarities among each other than the and T-47D cells (the other ERa positive BC cell line latter (CCs: 0.64 and 0.39, respectively). The search for a analyzed). All hormone unresponsive BC cell lines subset of genes to better distinguish hormone respon- belong, instead, to the right cluster, where they do not sive and non-responsive BC cells led in this case to the appear related to each other (with the notable exception identification of a subset of 19 genes (No. 35–53 in of the related MDA-MB-435 and MDA-N lines). The Table 1), of which 15 were in common with the intrinsic same clustering pattern was obtained for BC cell lines NCI60 subset described above (No. 35–49). Clustering alone, that evidentiate segregation of ERa positive and against the expression data of these genes (bottom-right negative BC cell lines in separate clusters (left part of Fig. 1b) indicates high similarity among all estrogen of Fig. 1a). These results provide evidence that the responsive lines (CC: 0.83), that were highly dissimilar expression profile of a set of estrogen-regulated genes from all normal breast epithelial cells and from BT-549, can detect the phenotype of BC cells. We therefore asked Hs578T, and MDA-MB-231 BC cell lines (CC: 0.79). whether we could identify a subset of these genes Interestingly, this analysis indicates that the ER-SK- displaying increased power for discriminating ER- BR-3 cell line is more related to ER-positive than to the positive from ER-negative cells and searched for the other ER-negative BC cell lines. subcluster of genes having the greatest difference among the different BC cell lines. By visual inspection, Analysis of estrogen-responsive genes in breast we detected about 12 subclusters of such differentially cancer biopsies expressed genes. To assess the weight of each subcluster The results obtained by analysis of the estrogen in discrimination of ER positive from negative BC cell responsive transcriptome in BC cell lines support the lines, we performed a series of hierarchical clustering possibility that the set of informative genes hereby tests where, in each case, genes belonging to one identified may be useful also to characterize hormone subcluster were excluded from the analysis set. As a responsiveness and signaling in BC in surgical tumor selection criteria, we considered changes in the correla- specimens. It is possible, in fact, that the expression tion coefficient measured at the first node of the cell line profile of genes belonging to the informative subset of dendrogram following exclusion from the analysis of the estrogen responsive genes characterizing BC cell lines data relative to the subcluster under study, and set may specify also ERa positive breast tumors, where apart those genes whose absence caused the major these genes could collectively represent a marker to increase of this coefficient, corresponding to a reduction assess estrogen signaling. This possibility was tested in of the overall measurable distance between ERa positive silico using the ‘Molecular Portraits of Human Breast and negative cell lines. This led us to identify 49 genes Tumors’ data relative to a comprehensive gene expres- (bottom-right of Fig. 1a and No. 1–49 in Table 1) whose sion analysis performed in 62 tumor specimens from 40 MOLECULAR TYPING OF ERa POSITIVE BREAST CANCER CELLS 445

TABLE 1. Panel of estrogen-responsive genes whose expression proﬁles discriminate ERa-positive and -negative breast cancer (BC) cell lines and tumor biopsies

Symbola UniGenea Descriptiona Functionb E-resp.c Referencesd

1 ABCG1 Hs.10237 ATP-binding cassette, sub-fam. G Transporters þ 1–2 (WHITE), member 1 2 ACTG1 Hs.14376 Actin, gamma 1 Cell communication, adhesion, or þ 1–2 motility 3 ACTN1 Hs.119000 Actinin, alpha 1 Cell communication, adhesion, or þ 1–2 motility 4 ADCY3 Hs.8402 Adenylate cyclase 3 Signal transduction þ 1–2 5 ANXA9 Hs.279928 Annexin A9 Cell communication, adhesion, or 1–2 motility 6 ATP5G3 Hs.429 ATP synthase, Hþ transporting, Electron transport and oxidative þ 1–2 mitochondrial F0 complex, subunit phosphorylation c (subunit 9) isoform 3 7 ATP9A Hs.70604 ATPase, Class II, type 9A Transporters 1–2 8 CCT4 Hs.79150 Chaperonin containing TCP1, subunit 4 Protein folding þ 1–2 (delta) 9 CCT5 Hs.1600 Chaperonin containing TCP1, subunit 5 Protein folding þ 1–2 (epsilon) 10 CKS2 Hs.83758 CDC28 protein kinase 2 Cell cycle þ 1–2 11 COX5A Hs.296585 Cytochrome c oxidase subunit Va Ribosome biogenesis þ 1–2 12 DKC1 Hs.4747 Dyskeratosis congenita 1, dyskerin Ribosome biogenesis þ 1–2 13 FKBP4 Hs.848 FK506 binding protein 4 (59 kDa) Protein folding þ 1–2 14 FLJ13187 Hs.29724 FLJ13187 Phafin 2 Unnamed 1–2 15 FLJ14299 Hs.288042 Hypothetical protein FLJ14299 Unnamed þ 1–2 16 HSPA8 Hs.180414 Heat shock 70 kDa protein 8 Protein folding þ 1–2 17 LGALS1 Hs.227751 Lectin, galactoside-binding, soluble, 1 Cell communication, adhesion, or þ 1–2 (galectin 1) motility 18 MAD2L1 Hs.79078 MAD2 mitotic arrest deficient-like 1 Cell cycle þ 1–2 (yeast) 19 MARS Hs.279946 Methionine-tRNA synthetase Protein biosynthesis þ 1–2 20 MCM2 Hs.57101 Minichromosome maintenance deficient DNA replication and repair þ 1–2 2, mitotin (S. cerevisiae) 21 MRPL3 Hs.79086 Mitochondrial ribosomal protein L3 Ribosome biogenesis þ 1–2 21 NOLC1 Hs.75337 Nucleolar and coiled-body phosphprotein Unknown þ 1–2 1 22 NP Hs.75514 Nucleoside phosphorylase Purine and pyrimidine metabolism þ 1–2 23 NRP1 Hs.69285 Neuropilin 1 Signal transduction þ 1–2 24 NTRK1 Hs.85844 Neurotrophic tyrosine kinase, receptor, Signal transduction þ 1–2 type 1 25 PMAIP1 Hs.96 Phorbol-12-myristate-13-acetate-induced Unknown þ 1–2 protein 1 26 PMSCL2 Hs.75584 Polymyositis/scleroderma autoantigen 2 Protein kinase þ 1–2 (100 kDa) 27 RXRA Hs.20084 Retinoid X receptor, alpha Transcription 1–2 28 SLC25A5 Hs.79172 Solute carrier family 25 (mitochondrial Transporters þ 1–2 carrier; adenine nucleotide transloca- tor), member 5 29 SNRPG Hs.77496 Small nuclear ribonucleoprotein poly- MRNA processing þ 1–2 peptide G 30 STXBP1 Hs.239356 Syntaxin binding protein 1 Exocytosis þ 1–2 31 TAF9 Hs.60679 TAF9 RNA polymerase II, TATA box Transcription þ 1–2 binding protein (TBP)-associated factor, 32 kDa 32 TPD52L1 Hs.16611 Tumor protein D52-like 1 Unknown þ 1–2 33 TUBB Hs.336780 Tubulin, beta polypeptide Cell communication, adhesion, or þ 1–2 motility 34 VARS2 Hs.159637 Valyl-tRNA synthetase 2 Protein biosynthesis þ 1–2 35 BLVRB Hs.76289 Biliverdin reductase B (flavin reductase Xenobiotic metabolism þ 1–2 (NADPH)) 36 CLDN4 Hs.5372 Claudin 4 Cell communication, adhesion, or 1–2 motility 37 ERBB3 Hs.199067 V-erb-b2 erythroblastic leukemia viral Signal transduction 1–2 oncogene homol. 3 38 ESTs Hs.36102 ESTs, Highly similar to SMHU1B Unnamed þ 1–2 R99207/ metallothionein 1B [H. sapiens] H72722 39 FEN1 Hs.4756 Flap structure-specific endonuclease 1 DNA replication and repair þ 1–2 40 FLJ11796 Hs.284186 Homo sapiens cDNA FLJ11796 fis, clone Unnamed þ 1–2 HEMBA1006158, highly similar to Homo sapiens transcription factor forkhead-like 7 (FKHL7) gene 41 IGFBP5 Hs.180324 Insulin-like growth factor binding Signal transduction 1–2 protein 5 42 LDHA Hs.2795 Lactate dehydrogenase A Glycid metabolism þ 1–2 446 WEISZ ET AL.

TABLE 1. (Continued)

Symbola UniGenea Descriptiona Functionb E-resp.c Referencesd

43 ME1 Hs.14732 Malic enzyme 1, NADP(þ)-dependent, Citrate metabolism þ 1–2 cytosolic 44 MT1L Hs.94360 Metallothionein 1L Heavy metal binding þ 1–2 45 MUC1 Hs.89603 Mucin 1, transmembrane Cell communication, adhesion, 1–2 or motility 46 PFKP Hs.99910 Phosphofructokinase, platelet Glycid metabolism þ 1–2 47 SDF1 Hs.237356 Stromal cell-derived factor 1 Signal transduction þ 1–2 48 SLC9A3R1 Hs.184276 Solute carrier family 9 (sodium/hydrogen Transporters þ 1–2 exchanger), isoform 3 regulatory factor 1 49 YARS Hs.239307 Tyrosyl-tRNA synthetase Protein biosynthesis þ 1–2 50 CD44 Hs.169610 CD44 antigen (homing function and Cell communication, adhesion, þ 1–2 Indian blood group system) or motility 51 JUN Hs.78465 V-jun sarcoma virus 17 oncogene Transcription þ 1–2 homolog (avian) 52 RERG Hs.21594 RAS-like, estrogen-regulated, Signal transduction þ 1–2 growth-inhibitor 53 TFF1 Hs.350470 Trefoil factor 1 (BC, estrogen-inducible Secreted protein þ 1–2 sequence expressed in) 54 ASAH Hs.75811 N-acylsphingosine amidohydrolase Fatty acid metabolism þ 3–4 (acid ceramidase) 55 CEACAM5 Hs.220529 Carcinoembryonic antigen-related cell Cell communication, adhesion, 3 adhesion molecule 5 or motility 56 ERBB2 Hs.323910 v-erb-b2 erythroblastic leukemia viral Signal transduction 1, 3, 5 oncogene homolog 2 57 LIV-1 Hs.79136 LIV-1 protein, estrogen regulated Unknown þ 1, 3, 7 58 MGP Hs.365706 Matrix Gla protein Cell communication, adhesion, þ 1, 3, 8 or motility 59 SCNN1A Hs.78853 Sodium channel, nonvoltage-gated Transporter 3 1 alpha 60 TFF3 Hs.352107 Trefoil factor 3 (intestinal) Secreted protein þ 4, 9 61 XBP1 Hs.149923 X-box binding protein 1 Transcription þ 1, 3, 10 CYP2B6 Hs.1360 Cytochrome P450, subfamily IIB Xenobiotic metabolism na (phenobarbital-inducib.) GATA3 Hs.169946 GATA-binding protein 3 Transcription na HNF3A Hs.299867 Hepatocyte nuclear factor 3, alpha Transcription na HPN Hs.823 Hepsin (transmembrane protease, Cell communication, adhesion, na serine 1) or motility aGene symbol, UniGene cluster number and description (http://www.ncbi.nlm.nih.gov/UniGene/). bFunction assigned according to Gene Ontology categories from LocusLink (http://www.ncbi.nlm.nih.gov/LocusLink/) and/or GeneCards (http://bioinfo.weizmann.ac.il/ cards/). cResponse to estrogen in human BC cells in vitro. d(1) Cicatiello et al., 2001 and Weisz et al., 1999; (2) Cicatiello Cicatiello L, Natoli G, Scafoglio C, Altucci L, Cancemi M, Facchiano A, Calogero R, Iazzetti G, De Bortoli M, Sfiligoi C, Sismondi P, Biglia N, Bresciani F and Weisz A. The gene expression program activated by estrogen in hormone-responsive human BC cells. 2003. Submitted; (3) Finlin et al., 2001; (4) Charpentier et al., 2000; (5) Perissi et al., 2000; (6) Hoch et al., 1999; (7) Seth et al., 2002; (8) Sheikh et al., 1993; (9) May and Westley 1997; (10) Bouras et al., 2002. na: not applicable. patients (Perou et al., 2000). These data were found Figure 1b (Table 1: ASAH, CEACAM5, ERBB2, LIV-1, particularly apt for this test, since they have been MGP, SCNN1A, TFF3, XBP1) and eight additional thoroughly controlled and can thus be considered very estrogen responsive genes (No. 54–61 in Table 1) which robust. Furthermore, the tumor samples included in have been proposed, or even demonstrated, to mark the that study were well characterized and their molecular estrogen responsive transcriptome of BC cells (see Refs. subtyping can be used as reference. We thus performed in Table 1). In addition, a survey of the literature led us the hierarchical clustering of the expression data, to identify 4 additional genes (Table 1: CYP2B6, GATA3, retrieved from the ‘Molecular Portraits of Human HNF3A, and HPN) that were consistently reported as Breast Tumors’ web site, relative to the same 218 expressed in ERa positive, but not in ER-negative, BC estrogen responsive genes used in the case of cell lines, cells, and tumors and for this reason proposed to play a to verify whether the expression profile could classify role in hormone signaling and/or in maintenance of the the tumors in a significant fashion. The results obtained estrogen responsive phenotype of the cell (Perou et al., by this direct approach did not give informative 2000; Dressman et al., 2001; Gruvberger et al., 2001; classification, since the differences between the two Sørlie et al., 2001; West et al., 2001; van’t Veer et al., main clusters obtained were not significant (CC: 0.92). 2002). When expression data relative to these 31 genes For this reason, we pursued our analysis by focusing on were included for the hierarchical clustering analysis, a the specific subset of estrogen responsive genes that significant classification was obtained. As depicted in allow us to discriminate hormone responsive from non- Figure 2, the samples separate now in two easily responsive BC cell lines (Fig. 1). Therefore, we included distinguishable main clusters (CC: 0.52). The smaller for further analysis the data relative to the 19 hormone of such clusters (positioned to the right in Fig. 2) responsive genes selected for the BC cell clustering in includes all ERa-negative tumor samples, characterized MOLECULAR TYPING OF ERa POSITIVE BREAST CANCER CELLS 447

Fig. 2. Cluster analysis of 62 surgical specimens of breast tumors Each column of the expression matrix represents the tissue sample and three normal mammary gland biopsies based on expression of a indicated at the top and each row refer to a gene, colors of the matrix subset of 27 estrogen responsive genes identified in BC cell lines and elements represent mRNA expression levels relative to a common four molecular markers of ESR1 (ERa) positive breast tumors. reference sample (green for sample/reference ratios <1, red for ratios Expression data were from the ‘Molecular Portraits of Human Breast >1, black for ratios near 1, and gray for missing data). The top Tumors’ study and sample denomination has been maintained the dendrogram, representing hierarchical relationships between sam- same as in Perou et al. (2000). Colors highlights the molecular typing ples, have the terminal branches colored to reflect the molecular of the BC samples, e.g., luminal epithelial/ERþ (LE, red), basal-like nature of each tumor or normal tissue sample and the scale at the top- (BL, dark blue), cErb-B2-overexpression (ERBB2, green), normal-like right marks the correlation coefficient represented by the length of the (NL, light brown), and undetermined (black). The three normal breast dendrogram branches connecting pairs of nodes. The bottom matrix, tissue samples are also marked in light brown. BC tissue samples from showing the expression patterns of cytokeratin 5 (KRT5), 8 (KRT8), the same patient, before (BE) and after (AF) neoadjuvant chemother- 17 (KRT17), and 18 (KRT18), integrin b4 (ITGB4), and ERa (ESR1) apy or from the primary tumor (P) and a lymph node metastasis (LN), genes, which where not part of the clustering analysis, is included for are connected by horizontal square brackets or open and closed circles. reference. 448 WEISZ ET AL. by a basal-like molecular phenotype, plus one biopsy of analysis on cell lines, which clustered very efficiently receptor negative normal breast tissue. All ERa-positive ER-positive and -negative cell lines and distinguished tumors partition, instead, to the second primary cluster them clearly from lines derived from other tissues. The (on the left), where they are clearly separated from ERa- ER phenotype appears thus clearly linked to other negative tumors overexpressing ERBB2 and from two phenotypic markers of the cells. For comparison, in receptor negative samples, characterized by a luminal- Figure 2 a color matrix showing the relative expression like signature. Sample pairs from the same tumor are of molecular markers of breast epithelial cell lineage always more similar to each other than to any of the was included, which distinguish basal—from luminal— other samples present in the set, with three exceptions, like BC cells (Perou et al., 2000), i.e., cytokeratin 5 including two ERBB2-positive tumors which resulted (KRT5), 8 (KRT8), and 17 (KRT17), integrin b4 (ITGB4) normal-like following successful chemotherapy (Perou and laminin g1 (LAMC1). The expression pattern of such et al., 2000) (marked by open and closed circles, markers confirms that ERa expressing cells belong to respectively, in Fig. 2), and one luminal-like tumor the luminal-like cell lineage. The GATA binding protein (NORWAY 7), that appeared to express very different 3 (GATA3) and estrogen regulated LIV-1 protein (LIV-1) amounts of ERa mRNA before and after chemotherapy. genes, also included in this matrix are previously These results indicate that ERa positive and negative characterized molecular markers of ERa positive cells BC phenotypes are indeed characterized by specific (Perou et al., 2000; Dressman et al., 2001; Gruvberger expression ‘signatures’ of at least a subset of estrogen et al., 2001; Sørlie et al., 2001; West et al., 2001; van’t responsive genes. Veer et al., 2002) whose expression pattern confirms the nature of the cell lines analyzed. It is noteworthy that in DISCUSSION our analysis, SK-BR-3 cells were found to segregate with The work presented here was performed to verify ER-positive cells, despite the fact that these cells do not whether the expression profile of a novel subset of genes, contain measurable ER and are unresponsive to estro- whose regulation by estrogens has been demonstrated gen stimulation (Cavailles et al., 2002). An intriguing in the experimental model system, can be used to possibility to account for the discrepancy is that distinguish the activity of estrogen-dependent path- the estrogen responsive transcriptome could mimic here ways in BC cells and tumors, independently of their the presence of hormone even in the absence of receptor status. This was first suggested to us by detectable ER. Indeed, SK-BR-3 cells that do not express quantitative real-time RT-PCR expression analysis of ERa,ERb, or GATA3, but show the same pattern of eight such estrogen-regulated genes in BC biopsies KRT8 and 18 expression that characterizes luminal-like (Sorbello et al., 2003). By exploiting a group of 344 estrogen responsive BC cells (Fig. 1b). Interesting, SK- estrogen-responsive genes discovered by expression BR-3 cells show a gene expression pattern characteristic profiling of BC cells in culture, we have tested this of the luminal-like phenotype also when considering the hypothesis by exploring the NCI60 (Ross et al., 2000) expression profile of 1,753 genes (Perou et al., 2000). SK- and ‘‘Molecular portrait of human BC cells’’ (Perou et al., BR-3 cells could thus derive from a luminal-like cell 2000) data sets, generated by profiling cell line and clone where steroid signaling was initially inducible but breast tumor biopsies, respectively. Since the data become constitutive during progression of the carcino- sets to be aligned were generated with microarray genic process, perhaps concomitant with ERRB2 ampli- platforms including different target cDNA sets, the fication and overexpression, a genetic lesion that analysis was limited to the 179 and 219 estrogen- characterizes this cell line. It has been shown, in fact, responsive genes, respectively. The first hierarchical that hyperactivity of the p185c-erbB2 oncoprotein may clustering analysis performed on the NCI60 dataset induce loss of ERa expression (Pietras et al., 1995; Oh grouped ER-positive and ER-negative cells in clearly et al., 2001) and contribute to acquisition of a hormone distinct clusters, directly confirming our starting independent growth phenotype in estrogen responsive hypothesis. It is important to note that neither ER nor BC cells (Liu et al., 1995; Tang et al., 1996). The PgR genes were included among those analyzed, to avoid possibility that SK-BR-3 cells share a gene expression introducing strong bias in the analysis. In ERa positive background with hormone responsive cells could cells, activity of the estrogen responsive genes analyzed explain also the responses to estrogens that can be appears considerably different than in receptor negative observed in this cell line under certain experimental cells. This was expected, since gene expression measure- conditions (Yoo et al., 1998; Filardo et al., 2002), which ments were carried out under conditions of cell stimula- may reflect here a residual activity of the estrogen tion with estrogen (the cultures used for mRNA dependent signal transduction pathway. extraction were maintained in the presence of estrogen It is worth noting that 9 out of 53 genes identified in rich whole serum and phenol red). It is conceivable, on this analysis were described as estrogen-responsive also the other side, that the expression of a fraction of in two published studies (Inoue et al., 2002; Frasor et al., estrogen-regulated genes may be sustained, in ER- 2003). These authors used different platforms, with negative cells, by activation of other regulatory path- different probe sets, to analyze response to estrogen in ways. This brought us to consider the possibility to MCF-7 cells, while our original set was identified in identify among the genes analyzed a specific subset that ZR75.1 cells (Weisz et al., 1999; Cicatiello et al., 2001; can allow to discriminate hormone responsive from non- Cicatiello L, Natoli G, Scafoglio C, Altucci L, Cancemi M, responsive BC cells even more effectively and this was Facchiano A, Calogero R, Iazzetti G, De Bortoli M, indeed verified, by identifying a subset of 49 genes, Sfiligoi C, Sismondi P, Biglia N, Bresciani F and Weisz among the 179 genes used for the first study and a subset A. The gene expression program activated by estrogen in of only 19 genes among the 219 genes used for the second hormone responsive human BC cells. Submitted for MOLECULAR TYPING OF ERa POSITIVE BREAST CANCER CELLS 449 publication elsewhere) and this explains clearly the which may be exploitable to improve the definition of the relatively low occurrence value observed. hormone responsive BC phenotype, providing a new Clustering analysis of expression data relative to 219 way to complement the informations currently provided estrogen-regulated genes of 62 breast tumor biopsies did by classical molecular markers, in particular the not result in a biologically relevant clustering of the presence of ERs and PgRs in tumor cells. tumors. One possible explanation for this finding resides on the fact that surgical tumor specimens are not homogeneous in their cellular composition and include ACKNOWLEDGMENTS many cell types, which are generally present in varying We thank Raffaele Calogero and Massimo Cancemi proportions in different tissue samples. As a consequ- for technical assistance and useful comments and ence, the gene expression profile of these tissue biopsies suggestions. may have predominantly reflected, under these conditions, their histological complexity. As it has been shown that variations in the activity of specific subsets of genes LITERATURE CITED may indeed uncover important similarities and differ- Bertucci F, Nasser V, Granjeaud S, Eisinger F, Adelaide J, Tagett R, ences among breast tumors (Perou et al., 2000; Loriod B, Giaconia A, Benziane A, Devilard E, Jacquemier J, Viens Gruvberger et al., 2001; Sørlie et al., 2001; West et al., P, Nguyen C, Birnbaum D, Houlgatte R. 2002. Gene expression profiles of poor-prognosis primary breast cancer correlate with 2001; van’t Veer et al., 2002), we searched for genes survival. Hum Mol Genet 11:863–872. which may differentiate ER-positive from ER-negative Biswas DK, Averboukh L, Sheng S, Martin K, Ewaniuk DS, Jawde TF, tumors among those which allowed to discriminate Wang F, Pardee AB. 1998. Classification of breast cancer cells on the among these two phenotypes in BC cell lines. In the basis of a functional assay for estrogen receptor. Mol Med 4:454– 467. process, we combined the 19 more informative ones with Bouras T, Southey MC, Chang AC, Reddel RR, Willhite D, Glynne R, 12 additional genes reported to be estrogen-responsive Henderson MA, Armes JE, Venter DJ. 2002. Stanniocalcin 2 is an or expressed consistently in ER-positive, but not ER- estrogen-responsive gene coexpressed with the estrogen receptor in negative, tumors. The clustering analysis of tumors, human breast cancer. Cancer Res 62:1298–1295. Cavailles V, Gompel A, Portois MC, Thenot S, Mabon N, Vignon F. carried out with such estrogen-regulated gene set, 2002. Comparative activity of pulsed or continuous estradiol confirmed, and in some instances improved, the classi- exposure on gene expression and proliferation of normal and fication originally proposed using 1,753 genes (Perou tumoral human breast cells. J Mol Endocrinol 28:165–175. et al., 2000). Interestingly, certain distinctive molecular Charpentier AH, Bednarek AK, Daniel RL, Hawkins KA, Laflin KJ, Gaddis S, MacLeod MC, Aldaz CM. 2000. Effects of estrogen on properties of the tumor samples were still reflected, global gene expression: identification of novel targets of estrogen under these conditions, by their clustering pattern as, action. Cancer Res 60:5977–5983. for example, most BCs previously classified as ‘normal- Cicatiello L, Facchiano A, Caligero R, De Bortoli M, Bresciani F, Weisz like’ resulted here more closely related to normal breast A. 2001. Gene expression monitoring in hormone-responsive human breast cancer cells during estrogen-induced cell cycle progression. tissue than to other BC specimens (Fig. 2). Proceedings of the AACR/Nature Genetics Joint Conference: Onco- Since ERa-positive and -negative BC phenotypes genomics: Dissecting Cancer through Genome Research. p 95. (http:// are characterized by specific estrogen responsive gene www.nature.com/cgi-taf/DynaPage.taf?file¼/ng/journal/v27/n4s/full/ expression ‘signatures,’ the gene set identified ng0401supp_95a.html&_UserReference¼C0A804EC4650B9B7E02- E44E479153B0D3D72). here appear to be exploitable to improve the definition Dressman MA, Walz TM, Lavedan C, Barnes L, Buchholtz S, Kwon I, of the hormone responsive BC phenotype, suggesting a Ellis MJ, Polymeropoulos MH. 2001. Genes that co-cluster with new way to complement the informations currently estrogen receptor alpha in microarray analysis of breast biopsies. provided by other molecular markers, in particular the Pharmacogenomics J 1:135–141. Eisen MB, Spellman PT, Brown PO, Botstein D. 1998. Cluster presence of ERs and PgRs in tumor cells. This also analysis and display of genome-wide expression patterns. Proc Natl provides direct evidence that gene regulation patterns Acad Sci USA 95:14863–14868. identified in experimental model systems can be Ferguson AT, Lapidus RG, Davidson NE. 1998. The regulation of exploited to identify molecular phenotypes of tumors estrogen receptor expression and function in human breast cancer. Cancer Treat Res 94:255–278. which relate to responsiveness to endocrine therapy. Filardo EJ, Quinn JA, Frackelton AR Jr, Bland KI. 2002. Estrogen This hypothesis can now be tested retrospectively on action via the G protein-coupled receptor, GPR30: stimulation of tumors whose response to hormone therapy has been adenylyl cyclase and cAMP-mediated attenuation of the epidermal assessed or in prospective studies of neoadjuvant growth factor receptor-to-MAPK signaling axis. Mol Endocrinol 16:70–84. treatments. A further result of this study worth, in our Finlin BS, Gau CL, Murphy GA, Shao H, Kimel T, Seitz RS, Chiu YF, opinion, to be highlighted is represented by the Botstein D, Brown PO, der CJ, Tamanoi F, Andres DA, Perou CM. possibility that a relatively small set of selected genes 2001. RERG is a novel ras-related, estrogen-regulated and growth- may reveal important features of the hormone respon- inhibitory gene in breast cancer. J Biol Chem 276:42259–42267. Frasor J, Danes JM, Komm B, Chang KCN, Lyttle CR, Katzenellenbo- sive phenotype of BC. If the activity of so few genes can gen BS. 2003. Profiling of estrogen up- and down-regulated gene indeed be used to effectively monitor ER signaling status expression in human breast cancer cells: insights into gene of tumors for diagnostic/predictive purposes, in fact, networks and pathways underlying estrogenic control of prolifera- cost-effective, focused transcriptome analysis of small tion and cell phenotype. Endocrinology 144:4562–4574. Gruvberger S, Ringner M, Chen Y, Panavally S, Saal LH, Borg A, tissue samples, like core biopsies or microdissected Ferno M, Peterson C, Meltzer PS. 2001. Estrogen receptor status in specimens, may become feasible. This hypothesis is now breast cancer is associated with remarkably distinct gene expres- worth being thoroughly tested in clinical settings. sion patterns. Cancer Res 61:5979–5984. In conclusion, we have shown how ERa positive and Harris JR, Lippman ME, Morrow M, Osborne CK. 2000. Diseases of the breast. Philadelphia: Lippincott, Williams & Wilkins. negative BC phenotypes are characterized by specific Hoch RV, Thompson DA, Baker RJ, Weigel RJ. 1999. GATA-3 is estrogen-dependent gene expression ‘signatures,’ and expressed in association with estrogen receptor in breast cancer. Int identified at the same time a relatively small set of genes J Cancer 84:122–128. 450 WEISZ ET AL.

Hopp TA, Fuqua SA. 1998. Estrogen receptor variants. J Mammary Brown PO. 2000. Systematic variation in gene expression patterns Gland Biol Neoplasia 3:73–83. in human cancer cell lines. Nat Genet 24:227–235. Ince TA, Weinberg RA. 2002. Functional genomics and the breast Seth P, Krop I, Porter D, Polyak K. 2002. Novel estrogen and cancer problem. Cancer Cell 1:15–17. tamoxifen induced genes identified by SAGE (Serial Analysis of Inoue A, Yoshida N, Omoto Y, Oguchi S, Yamori T, Kiyama R, Hayashi Gene Expression). Oncogene 21:836–843. S. 2002. Development of cDNA microarray for expression profiling of Sheikh MS, Shao ZM, Chen JC, Fontana JA. 1993. Differential estrogen-responsive genes. J Mol Endocrinol 29:175–192. regulation of matrix Gla protein (MGP) gene expression by retinoic Liu Y, el-Ashry D, Chen D, Ding IY, Kern FG. 1995. MCF-7 breast acid and estrogen in human breast carcinoma cells. Mol Cell cancer cells overexpressing transfected c-erbB-2 have an in vitro Endocrinol 92:153–160. growth advantage in estrogen-depleted conditions and reduced Sorbello V, Fuso L, Sfiligoi C, Scafoglio C, Ponzone R, Biglia N, Weisz estrogen-dependence and tamoxifen-sensitivity in vivo. Breast A, Sismondi P, De Bortoli M. 2003. Quantitative real-time RT-PCR Cancer Res Treat 34:97–117. analysis of eight novel estrogen-regulated genes in breast cancer. Lønning PE, Sørlie T, Perou CM, Brown PO, Botstein D, Børresen- Int J Biol Markers 18:123–129. Dale A-L. 2001. Microarrays in primary breast cancer–lessons from Sørlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie chemotherapy studies. Endocr Relat Cancer 8:259–263. T, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H, Matese Martin KJ, Kritzman BM, Price LM, Koh B, Kwan CP, Zhang X, JC, Brown PO, Botstein D, Eystein Lønning P, Børresen-Dale A-L. Mackay A, O’Hare MJ, Kaelin CM, Mutter GL, Pardee AB, Sager R. 2001. Gene expression patterns of breast carcinomas distinguish 2000. Linking gene expression patterns to therapeutic groups in tumor subclasses with clinical implications. Proc Natl Acad Sci USA breast cancer. Cancer Res 60:2232–2238. 98:10869–10874. May FE, Westley BR. 1997. Expression of human intestinal trefoil Tang CK, Perez C, Grunt T, Waibel C, Cho C, Lupu R. 1996. factor in malignant cells and its regulation by oestrogen in breast Involvement of heregulin-beta2 in the acquisition of the hormone- cancer cells. J Pathol 182:404–413. independent phenotype of breast cancer cells. Cancer Res 56:3350– Oh AS, Lorant LA, Holloway JN, Miller DL, Kern FG, El-Ashry D. 3358. 2001. Hyperactivation of MAPK induces loss of ERalpha expression van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil DW, in breast cancer cells. Mol Endocrinol 15:1344–1359. Schreiber GJ, Peterse JL, Roberts C, Marton MJ, Parrish M, Atsma Perissi V, Menini N, Cottone E, Capello D, Sacco M, D, Witteveen A, Glas A, Delahaye L, van der Velde T, Bartelink H, Montaldo F, De Bortoli M. 2000. AP-2 transcription factors in the Rodenhuis S, Rutgers ET, Friend SH, Bernards R. 2002. A gene- regulation of ERBB2 gene transcription by oestrogen. Oncogene expression signature as a predictor of survival in breast cancer. N 19:280–288. Engl J Med 347:1999–2009. Perou CM, Jeffrey SS, van de Rijn M, Rees CA, Eisen MB, Ross DT, van’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Pergamenschikov A, Williams CF, Zhu SX, Lee JC, Lashkari D, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, Schreiber Shalon D, Brown PO, Botstein D. 1999. Distinctive gene expression GJ, Kerkhoven RM, Roberts C, Linsley PS, Bernards R, Friend SH. patterns in human mammary epithelial cells and breast cancers. 2002. Gene expression profiling predicts clinical outcome of breast Proc Natl Acad Sci USA 96:9212–9217. cancer. Nature 415:530–536. Perou CM, Sørlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Weisz A, Cicatiello L, Bresciani F. 1999. Microarray-based gene Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Perga- expression monitoring of hormone-responsive human breast cancer menschikov A, Williams C, Zhu SX, Lønning PE, Børresen-Dale cells during estrogen-induced G1 progression. Proceedings of the A-L, Brown PO, Botstein D. 2000. Molecular portraits of human EMBO Workshop ‘Structure and Function of Nuclear Receptors’. breast tumours. Nature 406:747–752. p60. Pietras RJ, Arboleda J, Reese DM, Wongvipat N, Pegram MD, Ramos West M, Blanchette C, Dressman H, Huang E, Ishida S, Spang R, L, Gorman CM, Parker MG, Sliwkowski MX, Slamon DJ. 1995. Zuzan H, Olson JA Jr, Marks JR, Nevins JR. 2001. Predicting the HER-2 tyrosine kinase pathway targets estrogen receptor and clinical status of human breast cancer by using gene expression promotes hormone-independent growth in human breast cancer profiles. Proc Natl Acad Sci USA 98:11462–11467. cells. Oncogene 10:2435–2446. Yoo JY, Lessor T, Hamburger AW. 1998. Inhibition of cell proliferation Ross DT, Scherf U, Eisen MB, Perou CM, Rees C, Spellman P, Iyer V, by 17beta-estradiol and heregulin beta1 in estrogen receptor Jeffrey SS, Van de Rijn M, Waltham M, Pergamenschikov A, Lee JC, negative human breast carcinoma cell lines. Breast Cancer Res Lashkari D, Shalon D, Myers TG, Weinstein JN, Botstein D, Treat 51:71–81.