ONCOGENOMICS Standardize (Pauletti Et Al., 2000)

Identiﬁcation and validation of an ERBB2 gene expression signature in breast cancers

Franc¸ois Bertucci1,2,3,6, Nathalie Borie4,6, Christophe Ginestier1,6, Agne` s Groulet4, Emmanuelle Charafe-Jauffret1,3,5, Jose´ Ade´ laı¨ de1, Jeannine Geneix1, Loı¨ c Bachelart1, Pascal Finetti1, Alane Koki4, Fabienne Hermitte4, Jacques Hassoun3,5, Ste´ phane Debono4, Patrice Viens2,3, Vincent Fert4, Jocelyne Jacquemier1,5 and Daniel Birnbaum*,1

1De´partement d’Oncologie Mole´culaire, Institut Paoli-Calmettes and UMR119 Inserm, IFR57, Marseille, France; 2De´partement d’Oncologie Me´dicale, Institut Paoli-Calmettes, Marseille, France; 3Faculte´ de Me´decine, Universite´ de la Me´diterrane´e, Marseille, France; 4IPSOGEN S.A., Marseille, France; and 5De´partement de Biopathologie, Institut Paoli-Calmettes, Marseille, France

ERBB2 is a transmembrane tyrosine kinase receptor Introduction encoded by a gene located in chromosome region 17q12. Overexpression of ERBB2, generally by way of gene The ERBB2 oncogene, also called HER2 or NEU,is amplification, plays a role in mammary oncogenesis.This located in band q12 of chromosome 17 (http://www. alteration can be overcome by use of the humanized ensembl.org/). It codes for a 185-kDa transmembrane monoclonal antibody trastuzumab (Herceptint).Accurate tyrosine kinase related to members of the ERBB/EGFR determination of ERBB2 status is required for appro- family. ERBB2 is amplified and overexpressed in 15– priate use of this targeted therapy and is currently 30% of breast cancers (Slamon et al., 1987). Although analysed by immunohistochemistry (IHC) on tissue its exact role in mammary oncogenesis remains unclear sections and/or fluorescence in situ hybridisation (FISH) (Holbro et al., 2003, for review), the receptor is a on interphase chromosomes.We have studied the gene clinically relevant target for the treatment of disease for expression profiles of a series of 213 breast tumours and two reasons. First, ERBB2 gene amplification and 16 breast cancer cell lines with known ERBB2 status, overexpression of ERRB2 gene products have been using Ipsogen’s DiscoveryChip microarrays with B9000 associated with prognosis or response to anticancer cDNAs.We have identified 36 genes and expressed therapies (for a review, see Hayes and Thor, 2002). sequence tags that were differentially expressed in Second, therapy based on a humanized monoclonal tumours and in cell lines with and without ERBB2 protein anti-ERBB2 antibody (trastuzumab/Herceptint)has overexpression.This ERBB2-specific gene expression shown benefits in metastatic patients (for a review, see signature (GES) contained 29 overexpressed genes Leyland-Jones, 2002). However, modifications of che- including the ERBB2 gene itself, five genes located in its motherapy and hormonal therapy strategies based on immediate vicinity on 17q12, non-17q genes such as ERBB2 status remain controversial. GATA4 and eight downregulated genes including oestro- Currently, ERBB2 status is primarily determined by gen receptor a (ER).Some correlations were validated at two different methods: fluorescence in situ hybridization the protein level using IHC on tissue microarrays.The (FISH), which reveals gene amplification, and immuno- GES was able to distinguish ERBB2-negative and histochemistry (IHC), which detects the overexpressed -positive cancer samples, as well as FISH-negative and protein (for recent reviews, see: Rampaul et al., 2002; FISH-positive ERBB2 2 þ IHC samples. Bilous et al., 2003). FISH is a good method for ERBB2 Oncogene (2004) 23, 2564–2575. doi:10.1038/sj.onc.1207361 testing but is technically more difficult to implement Published online 26 January 2004 than IHC, which is easier to perform but difficult to ONCOGENOMICS standardize (Pauletti et al., 2000). IHC is currently the Keywords: breast cancer; DNA microarrays; ERBB2/ only FDA-approved test for selection of patients for HER2; FISH; gene expression; expression profiles; treatment with trastuzumab, and American Society for tissue microarrays Clinical Oncology and National Comprehensive Cancer Network guidelines recommend the use of either FISH or the HercepTestt, a specific IHC test made by the Dako Corporation. This method semiquantitatively assesses positive staining on a scale ranging from 0 *Correspondence: D Birnbaum, De´ partement d’Oncologie Mole´ cu- (absence of ERBB2 protein overexpression) to 3 þ laire, Institut Paoli-Calmettes and UMR119 Inserm, IFR57, 217 Bd. (maximum of overexpression). For pathologists, inter- Leı¨ Roure, Marseille, France; E-mail: [email protected] 6These authors equally contributed. pretation of results is relatively straightforward in Received 20 August 2003; revised 29 October 2003; accepted 4 ERBB2-negative individuals (0–1 þ ) and in patients November 2003 who strongly overexpress the protein (3 þ ), but proble- ERBB2 GES in breast cancers F Bertucci et al 2565 matic for cases with intermediate level 2 þ ; for the latter mRNA expression profiles from 145 randomly chosen (10–15% of all breast cancers), the concordance with breast cancer samples (learning set), by comparing two FISH is, at best, 25%. Importantly, a proportion of 2 þ subgroups defined by their ERBB2 status as determined cases are bona fide ERBB2-overexpressing tumours to by standard IHC: samples scoring 0 and 1 þ (hereafter which Herceptin treatment should be applied. Thus, designated ERBB2À, 116 samples) were compared to accurate and standardized determination of ERBB2 samples scoring 3 þ (ERBB2 þ , 29 samples). At this status has not yet been achieved. The reliability of this stage, cases with equivocal 2 þ (n ¼ 10) or unavailable determination will greatly influence the selection of the (n ¼ 4) staining were excluded from analysis. To identify relevant cases and thus the efficacy of Herceptin. a molecular signature independent from the predefined Moreover, specific methods for patient selection for subgroups of tumours, we iteratively defined several ERBB2 antagonists may serve as a paradigm for guiding different subsets of samples and performed supervised clinical use of the new targeted approaches expected in analysis on each of these subsets independently. A total the near future. It is thus important to further document of 30 such iterations were carried out. The lists of genes the methods and parameters useful to assess ERBB2 identified as significant discriminators (these lists ranged status. from 80 to 274 clones) were then compared, revealing 37 Clinical outcome may vary between patients with the clones present in at least 25 lists: all these clones were same ERBB2 status and treatment, implying that other tag-resequenced to confirm their identity and defined factors may play a role in determining the level of our ERBB2-specific GES. sensitivity to trastuzumab. It may be necessary to Figure 1 shows the expression pattern of this associate other targeted therapies to anti-ERBB2 treat- signature in the 145 samples in a colour-coded matrix. ment and identification of complementary or secondary Samples are classified according to their correlation targets may prove useful. Although the common path- coefficient with the ERBB2 þ group. As shown, the ways such as the RAS/MAPK pathway and other resulting discrimination between ERBB2 þ and induced genes have been reported (Oh et al., 1999), ERBB2À samples was successful with only two mis- ERBB2-associated signalling cascades have yet to be classifications. The 37 clones corresponded to 36 unique elucidated. Thus, accurate measurement of ERBB2 sequences (two different clones represented ERBB2) status, as well as identification of associated molecular representing 29 characterized genes and seven other alterations are now intensively required. sequences or ESTs. In all, 29 were overexpressed and Emerging technologies may facilitate progress on eight were underexpressed in ERBB2 þ samples. Their both ERBB2 typing and target discovery. Among these, identity and chromosomal location are listed in Figure 1. DNA microarrays are currently prominent (for reviews, We next validated our GES in an independent set of see Bertucci et al., 2003a, b). This technology can be 54 breast cancer samples (validation set, 46 ERBB2À used to improve the prognostic classification of breast and eight ERBB2 þ ). As shown in Figure 2a, classifica- cancers (Golub et al., 1999; Bertucci et al., 2000; Perou tion of samples based on the GES successfully classified et al., 2000; Sorlie et al., 2001; Bertucci et al., 2002; Van them according to ERBB2 IHC status with only one de Vijver et al., 2002; Van’t Veer et al., 2002; Sorlie et al., ERBB2À sample misplaced in the ERBB2 þ group. 2003). Here, we have analysed 213 breast carcinomas and 16 breast cell lines using DNA microarrays B Validation of the ERBB2 GES from profiling of breast containing 9000 spotted PCR products from known tumour cell lines genes and expressed sequence tags (ESTs). Our aim was to identify differences in gene expression patterns We profiled a series of 16 breast cancer cell lines that between ERBB2-negative and ERBB2-positive breast included five cell lines (BT-474, HCC1569, MDA-MB- tumours. Using both supervised and unsupervised 453, SK-BR-3 and UACC-812) known to have ampli- analyses, we have identified a series of 37 discriminator fication and/or high mRNA expression of the ERBB2 genes/ESTs called ‘ERBB2 gene expression signature’, gene (Kauraniemi et al., 2001; Wilson et al., 2002). the expression of which permitted to distinguish ERBB2 GES successfully separated ERBB2 þ and ERBB2-negative and -positive samples, as well as ERBB2À cell lines (Figure 2b). FISH-negative and FISH-positive ERBB2 2 þ samples. Collectively, these analyses demonstrated that our Among the genes included in the signature were ERBB2 GES correctly classified breast tumours and cell potential additional targets, such as GATA4, which lines consistent with ERBB2 status evaluated with a was validated at the protein level using tissue microarray standard procedure (Herceptestt). (TMA) of breast cancers. Analysis of breast tumour samples using tissue microarrays Results Significant discriminator genes were further validated by immunohistochemical analysis of their corresponding Identification and validation of an ERBB2 gene B expression signature (GES) from tumour profiling proteins (Figure 3a). A total of 250 cases from TMA1 were available for the study of ERBB2, oestrogen Supervised analysis was utilized to identify a GES receptor a (ER, encoded by the ESR1 gene), GATA4 correlated with ERBB2 status. It was applied to the and Ki67. In ERBB2 GES, ERBB2, GATA4 and Ki67

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2566

Figure 1 Supervised classification of 145 breast tumours using ERBB2 GES. The classification of the learning sample set (145 cases) by supervised analysis on the basis of 37 clones defining our ERBB2 GES is shown. Bottom panel: Expression patterns of 37 cDNA clones in 145 samples. Each row represents a gene and each column represents a sample. Tumour samples are numbered from 1 to 145. Genes (right of panel) are referenced by their HUGO abbreviation as used in ‘Locus Link’ (http://www.ncbi.nlm.nih.gov/LocusLink/) and their chromosomal location (including which arm for chromosome 17). EST is used for clones without similarity with known gene or protein. Samples are ordered according to the correlation of their expression profile with the average profile of the ERBB2-positive group, and genes are ordered by their discriminating score. Each cell in the matrix represents the expression level of a transcript in a single sample relative to its median abundance across all samples and is depicted according to a colour scale shown at the bottom. Red and green indicate expression levels, respectively, above and below the median. The magnitude of deviation from the median is represented by the colour saturation. Grey indicates missing data. Top panel: The ERBB2 IHC status (HerceptTest) for each sample is shown: white square indicates sample scored 3 þ and black square indicates sample scored 0–1 þ

Figure 2 Validation of the ERBB2 GES. Our ERBB2 GES (37 genes/ESTs) was used for classifying independent series of breast cancer samples. (a) Same as for Figure 1, applied to the expression data of 54 additional breast cancers (validation set). Genes/ESTs located on 17q are marked with *(b) Same as for Figure 1, applied to the expression data of 16 breast cancer cell lines. Top panel: The ERBB2 status for each cell line is shown: white square indicates ampliﬁcation and/or high mRNA expression of the ERBB2 gene and black square indicates no ampliﬁcation and no overexpression

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2567 Table 1 Validation of GES using TMA1 for three non-17q-encoded proteins present in the signature ERBB2 (0–1+) ERBB2 (2–3+) P-value* n (%) n (%)

GATA4 Negative 169 (77%) 18 (47%) Positive 50 (23%) 20 (53%) o0.001

Ki67 o20 151 (72%) 21 (53%) X20 59 (28%) 17 (47%) o0.025

ER Negative 27 (13%) 18 (47%) Positive 179 (87%) 20 (53%) o0.0001

*Fisher exact test

A total of 68 (72%) of the 94 samples included in TMA2 were available for FISH analysis of ERBB2 locus. Examples of results are shown in Figure 3b. Of the 68 cases, 30 displayed ERBB2 ampliﬁcation, whereas 38 were not ampliﬁed.

Classification of breast tumours using ERBB2 GES Previous supervised analyses did not include the breast cancer samples with 2 þ (n ¼ 10) or unavailable (n ¼ 4) status for ERBB2 IHC. We reclassified them with all 145 samples previously analysed – which included the 68 cases with available FISH ERBB2 data – by using hierarchical clustering program based on ERBB2 GES. Results are displayed in Figure 4 that highlights clusters of correlated genes across clusters of correlated samples. The first large gene cluster contained 29 genes/ESTs, including ERBB2 (it was designated ‘ERBB2 cluster’). Figure 3 (a) Analysis of protein expression using immunohisto- The second gene cluster was globally negatively corre- chemistry on tissue microarray sections. On top, haematoxylin– lated with the previous one: it contained eight genes/ eosin staining (HES) of paraffin block section (25 Â 30 mm2) from ESTs, including ESR1 (it was designated ‘ER cluster’). TMA1 containing 552 tumours and control samples. Two examples Despite transcriptional heterogeneity between tu- of IHC staining for ERBB2, Ki67, ER and GATA4 are shown below. Case 1 shows a tumour sample with no detected ERBB2 mours for these genes, the combined expression patterns expression, no Ki67 expression, strong ER expression equal to defined at least three groups of tumours, designated A, Q ¼ 300 and no GATA4 expression. Case 2 shows a tumour sample B and C. Group A (73 cases, green branches of the with ERBB2 overexpression equal to 3 þ , a Ki67 expression equal dendrogram) displayed an overexpression of the ‘ER to P ¼ 20, no ER expression and a GATA4 overexpression equal to cluster’ and an underexpression of the ‘ERBB2 cluster’ Q ¼ 300. (b) Analysis of ERBB2 gene copy number in breast tumours using FISH on tissue microarray sections. On top, HES of overall compared to groups B and C. Conversely, the paraffin block section (25 Â 30 mm2) from TMA2 containing 94 ‘ERBB2 cluster’ and the ‘ER cluster’ were upregulated tumours. Below, two sections of invasive breast carcinomas are and downregulated in group C samples (36 cases, red shown, the first with ERBB2 amplification and the second with branches) overall, as compared to other groups. Finally, normal gene copy number. Red dots (arrows) represent ERBB2 copies and green dots represent centromere 17, on interphase group B (50 cases, black branches) displayed an chromosomes intermediate profile with heterogeneous expression of the ‘ERBB2 cluster’ and underexpression of the ‘ER cluster’. Correlations of tumour groups with ERBB2 status genes were overexpressed and ESR1 was underexpressed were analysed. As expected, group C strongly differed in ERBB2 þ samples. These correlations were con- from the other groups with respect to ERBB2 protein firmed at the protein level: overexpression of ERBB2 expression since 93% of all ERBB2 3 þ samples were protein was significantly associated with an upregulation located in this group. In group C, 77% of samples of GATA4 (Po0.001), Ki67 (Po0.025), and with scored 3 þ ,9%2þ and 14% 0–1 þ ; in contrast, in negativity of ER (Po0.0001) (Table 1). We found groups A and B, these rates were 0 and 5% (3 þ ), 3 and 40% of ERBB2 þ tumours in ER-negative tumours but 10% (2 þ ), and 97 and 85% (0–1 þ )(Po0.0001, w2 test, only 10% in ER-positive tumours. A vs B vs C groups), respectively. There was a strong

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2568

Figure 4 Unsupervised hierarchical classiﬁcation of 159 breast tumours using genes from the ERBB2 GES. (a) Hierarchical clustering of 159 breast tumours and 37 clones from the ERBB2 GES. Each row represents a clone and each column represents a sample (see Figure 1 for colour scheme). Dendrograms of samples (above data matrix) and genes (to the left of matrix) represent overall similarities in expression proﬁles. The orange vertical lines mark the subdivisions into three main tumour groups that are represented in the branches of dendrogram in green (A), black (B) and red (C), respectively. The dendrogram of genes is zoomed in (b). Between the dendrogram of samples and the data matrix are represented, according to a grey colour ladder relevant, histoclinical data for the tumours: ERBB2 IHC (HercepTest: 0–1 þ , white; 2 þ , light grey; 3 þ , black; unavailable; dark grey) and FISH (negative, white; positive, black; unavailable, dark grey) status, SBR grade (I, white; II, light grey; III, black; unavailable, dark grey), ER, PR and P53 IHC status (negative, white; positive, black, unavailable, dark grey), axillary lymph node invasion (negative, white; positive, black), pathological size of tumours (pT1, white; pT2, light grey; pT3, black). (b) Dendrogram of genes referenced by their HUGO abbreviation. Genes/ESTs located on 17q are marked with *. See the ‘ERBB2 cluster’ (red branches) and the ‘ER cluster’ (green branches)

correlation between tumour groups and FISH status Table 2 Data for assessment of sensitivity and specificity of ERBB2 with most of the FISH-positive cases clustered in group GES and IHC for prediction of ERBB2 FISH status 2 C(Po0.0001, w test, A vs B vs C groups). ERBB2 FISH FISH status Totala information and IHC status were both available in 64 cases out of 159. Interestingly, the three 2 þ tumours Negative Positive located in group C displayed ERBB2 amplification GES groups A+B 30 3b 33 (FISH positive), while the seven 2 þ tumours included C 42731 in groups A (two cases) and B (five cases) had no Totala 34 30 64 amplification (FISH negative). Our ERBB2 GES could c b separate FISH-positive and FISH-negative ERBB2 2 IHC status Negative 26 3 29 þ Positive 82735 tumours, providing more specific information with Totala 34 30 64 respect to FISH status as compared with ERBB2 IHC status (HercepTest). Indeed, the correlation between aConsidering 64 tumours with data available for IHC, FISH and GES- GES groups (C samples vs A þ B samples) and FISH based grouping; btwo samples are probably false-positive FISH results; result (negative vs positive) provided a sensitivity of cnegative: 0–1+ and positive, 2–3+ 90% and a specificity of 88% (concordance in 89% of cases). In comparison, the correlation between IHC- based grouping (0–1 þ vs 2-3 þ ) and FISH status absence of ERBB2 amplification (not shown). Taken showed an equal sensitivity of 90% but a weaker into account the two samples with false-positive FISH specificity of 76% (concordance in 82% of cases) results, the error rate was five out of 64 (with four false (Table 2). Sensitivity was even probably better for the positive and one false negative) for correlation between two comparisons; as shown in Figure 4, two samples our classification and FISH, whereas it was nine out of located in groups A and B and IHC negative for ERBB2 64 for correlation between standard IHC and FISH. were FISH positive; reviewing of the corresponding sections revealed in fact the presence of intraductal Correlation with histoclinical parameters carcinoma in one case and abundant necrosis in the other case, both of which might have lead to false We searched for correlations between tumour groups positive FISH results. Further verification using real- and relevant molecular and histoclinical parameters time quantitative PCR definitely demonstrated the of samples. As expected (Revillion et al., 1998), our

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2569 GES-based grouping correlated with SBR grade and There may be a whole group of loci whose transcription hormone receptor status. Group C did not contain is coordinately increased because the corresponding grade I samples; 44% of samples were grade II and 56% proteins belong to the same network. In total, four genes were grade III. In groups A þ B, 15% of samples were expressed in endothelial cells and platelets (encoding grade I, 48% were grade II and 37% were grade III three integrins ITGA2, ITGA2B, ITGB3, and an (P ¼ 0.02, w2 test). In group C, samples were likely to be adhesion molecule of the Ig family PECAM1) were ER negative (59%), compared with 27% in groups overexpressed in ERBB2 þ tumours (however, not all A þ B(P ¼ 0.001, w2 test). Similar, although not integrin genes from 17q present on the microarray were significant, correlation was found for PR status overexpressed since ITGA3 was not). Overexpression of (P ¼ 0.07, w2 test). No correlation was found with these genes may be associated with a specific (e.g. either pathological size of tumours, axillary lymph node status ER negative, basal-like or neoangiogenic) phenotype. and P53 IHC status. Other genes in the near vicinity of ERBB2 locus may be coamplified with ERBB2 gene, but may not be expressed due to the absence of an appropriate promoter or to Content of the ERBB2 GES repression. While amplification greatly influences gene In total, 29 genes/ESTs were significantly overexpressed expression (Pollack et al., 2002), only a proportion of in ERBB2 þ tumours. Their coexpression may indicate genes from a given amplicon are overexpressed (Platzer coamplification (same chromosomal location), regula- et al., 2002). tion by ERBB2, coregulation by common factors or Other overexpressed genes were not located on association with unknown phenotypic feature of disease. chromosome arm 17q. CDH15, also called M-Cadherin In addition to ERBB2 itself, there were six genes from or muscle cadherin, is expressed in myoepithelial cells. It region q12 of chromosome 17 in the signature (Figure 1); might suggest that ERBB2 þ tumours have a certain the six genes are all located within less than one degree of myoepithelial differentiation; alternatively, megabase on either side of ERBB2, defining a small they may be characterized by a high degree of ‘core’ region of coexpressed – probably coamplified – dedifferentiation with appearance of new markers. An genes (Figure 5). Genes from 17q12 that were absent interesting finding was GATA4, whose coexpression from the microarray (e.g. MLN64/STARD3) are not with ERBB2 was validated at the protein level. This shown in Figure 5. Although overexpression of 17q gene codes for a transcription factor of the GATA genes with ERBB2 is likely to be due simply to DNA family (Patient and McGhee, 2002). GATA4 is essential amplification, the functional affect of overabundant for cardiovascular development (Kuo et al., 1997), and transcripts of these genes may impact on the clinical may regulate genes critical for myocardial differentia- outcome in patients. Indeed, this may be the case for tion and function. Likewise, ERBB2 is essential for example for GRB7 or PPARBP. GRB7, a tyrosine heart development (Garratt et al., 2003, for review). It is kinase cytoplasmic adaptor substrate, has been impli- therefore possible that ERBB2 exerts some of its cated with different partners in integrin-mediated cell downstream effects through GATA4 or alternatively, migration (Shen et al., 2002). PPARBP has been shown that GATA4 stimulates ERBB2 gene transcription. to downregulate P53-dependent apoptosis (Frade et al., Incidentally, MAP2K6 is also strongly expressed in 2002). Other genes from the microarray and located on cardiac muscle (Han et al., 1996). The major adverse 17q but further apart from ERBB2 were not found in effect of Herceptin is cardiotoxicity. Investigation of the the signature except for ITGA2B/CD41, ITGB3/CD61, functional relationship between ERBB2, GATA4 and PECAM1/CD31 and MAP2K6. Overexpression of these MAP2K6 may enhance current understanding of genes may not be due to increased ERBB2 gene copy cardiotoxicities associated with ERBB2 antagonists, number per se, but may be triggered by intense ERBB2 and contribute to design ways to circumvent this side signalling; it might also be due to the presence of other effect. Activation of GATA4 is thought to occur telomeric, 17q-associated amplicons (Andersen et al., through RHO GTPases (Yanazume et al., 2002), which 2002; Hyman et al., 2002). ITGA2, whose gene is not are also central to the functions of integrins and on 17q, was also overexpressed in ERBB2 þ tumours. cadherins (for a review, see Arthur et al., 2002).

Figure 5 Localization of genes from chromosome 17q12–24 region represented on the DNA microarray. Genes identiﬁed in our ERBB2 GES and with expression upregulated in our ERBB2 þ breast cancer samples are indicated in bold. The other genes indicated were represented on the microarray but were not found in the ERBB2 signature. The list of genes is not thorough for genes located outside 17q12. From several studies, we can point to a ‘core’ of genes almost always cooverexpressed with ERBB2. @: gene cluster

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2570 The MKI67 gene encodes the proliferation marker when amplified. Willis et al. (2003) used an oligonucleo- Ki67/MIB1. It was upregulated in ERBB2 þ samples, tide chip to study mRNA from 12 breast tumours and suggesting that ERBB2 þ tumours are proliferative two cell lines also typed using comparative genomic tumours. Immunohistochemical results on B250 TMA1 hybridization. A total of 20 known genes showed tumours for ERBB2 and Ki67 staining confirmed such significant overexpression in tumours with gains of correlation at the protein level in agreement with recent region 17q12–23; these included ERBB2, GRB7, reports (Korsching et al., 2002; Callagy et al., 2003). The PPARBP, and also MLLT6, KRT10 and TUBG1, overexpression of the CSTA gene, which encodes which were not identified in our signature. In the study cystatin A, a cysteine protease inhibitor of the stefin of Wilson et al. (2002) using a microarray with B5.000 family that acts as endogenous inhibitor of cathepsins, cDNAs, only few genes from 17q were among those may be put in perspective with the finding of Oh et al. upregulated genes; these included RPL19 and LASP1. (1999) on the downregulation of cathepsin D in ERBB2- Dressman et al. (2003) studied 34 tumours and transfected MCF7 cells. Finally, the presence of genes established a gene expression signature specific of encoding two structurally related factors, lymphotoxin ERBB2 þ samples that contained several 17q genes A(LTA) and preB-cell colony-enhancing factor (PBEF), including GRB7, NR1D1, PSMB3 and RPL19. Sorlie and NFKBIE imply that specific immune and inflam- et al. (2003) have also defined ERBB2 þ signature with matory mechanisms may be associated with ERBB2 þ five genes from 17q12, including ERBB2, GRB7 and tumours. PPARBP. Taken together, these reports indicate that Only five genes with known function were down- genes located in the vicinity of ERBB2 are frequently regulated in ERBB2-positive tumours. Interestingly, one coupregulated following DNA amplification. Some of of these was ESR1, which encodes ER a. It is recognized the genes close to ERBB2 and present on our microarray that most ERBB2-amplified tumours are ER negative did not appear in our signature, whereas they were and resistant to hormone therapy (Keshgegian, 1995; found upregulated in other studies (i.e. LASP1, Konecny et al., 2003). Moreover, an interplay between MLLT6). This may be due to a different proportion ERBB2 and ER pathways has been demonstrated of tumours with variably sized amplicons in the (Pietras et al., 1995). There were also SCUBE2, a gene analysed panels. encoding a secreted protein with an EGF-like domain While amplification of region 17q12–21 affects (Yang et al., 2002), and CELSR2, which encodes a ERBB2 chromosomal neighbours, ERBB2 protein over- nonclassical cadherin. These two molecules might have expression will affect downstream targets and possibly antagonistic regulatory roles of ERBB2 activities at the also upstream regulators via positive feedback regula- cell membrane. SCUBE2 and NAT1 were associated to tory mechanisms. Balance in cadherins and integrins ESR1 in GES associated with ER positivity (Sorlie et al., and functional processes associated with cell–matrix 2003; personal data). adhesive systems seem particularly affected in ERBB2- positive tumours (Wilson et al., 2002; this study). This suggests that ERBB2 oncogenic activity may be Discussion associated with cell motility, as has been proposed previously (Tan et al., 1997; Spencer et al., 2000). A In this work, we have defined a set of genes that recent study, using DNA microarrays, containing B discriminate between ERBB2 þ and ERBB2À breast 6000 unique genes/ESTs, has described the transcrip- tumours. The identity of these genes may give insight in tional changes associated with a series of 61 genes the underlying biological mechanisms associated with following overexpression of a transfected ERBB2 gene ERBB2 status and with the aggressive phenotype of in a mammary cell line (Mackay et al., 2003). Previously, ERBB2 þ breast cancers. They may also provide new several studies had identified genes whose transcription diagnostic, prognostic and predictive factors, as well as is affected by ERBB2 overexpression or amplification new therapeutic targets. Some of the genes in the using differential screening (Tomasetto et al., 1995; signature reported in this study are known to be Oh et al., 1999). Some of these genes are located near associated with ERBB2 and cancer, while others the ERBB2 locus. Our GES shares no common gene constitute new associations. Two issues need to be with the list of Kumar-Sinha et al. (2003) established further discussed: the inclusion of these discriminator in comparing cell lines including ERBB2-transfected genes in similar studies and the interest of using GES for cell line; however, a gene related to fatty acid biology, the determination of ERBB2 status. FADS2, is part of our signature. Tiwari et al. (1991) reported a relationship between ERBB2, fatty acids 0 0 ERBB2 and microarrays and 2 ,5 oligoadenylate synthetases (OAS2), which is included in our ‘ERBB2 cluster’. Peroxisome prolifera- Several recent large-scale gene expression studies tor-activated receptors (PPARs) are known regulators have addressed the issue of ERBB2 status and function of lipid metabolism; their trans-activating capacity in breast cancer. A large-scale study of the ERBB2 depends on the recruitment of auxiliary proteins (for a amplicon was carried out on seven breast tumour cell review, see Gilde and van Bilsen, 2003); modifications lines by Kauraniemi et al. (2001) using a microarray of fatty acid metabolism in ERBB2 þ tumours may that included 217 cDNAs from 17q12 genes; ERBB2, thus be simply associated with overexpression of GRB7, PPP1R1B were consistently overexpressed PPARBP.

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2571 ERBB2 signature and assessment of ERBB2 status Materials and methods

The determination of ERBB2 status is important in Breast carcinoma samples breast cancer management, but is difficult in routine Using DNA microarrays, we studied 213 breast cancer samples clinics since both IHC and FISH have limitations obtained from women treated at the Institut Paoli-Calmettes and are influenced by many variables (for recent between 1988 and 2001. Inclusion criteria of samples were: (i) reviews, see Rampaul et al., 2002; Bilous et al., 2003). sporadic primary breast cancer treated with surgery, (ii) In routine practice, IHC, which more than FISH detects tumour material quickly dissected and frozen in liquid the actual target of Herceptint, is faster and more nitrogen and stored at À1601C. Exclusion criteria included economic but highly dependent on fixative conditions, locally advanced or inflammatory or metastatic forms. The main characteristics of patients and tumours are listed in staining procedures, scoring system, quality controls Table 3. Immunohistochemical parameters collected included and interlaboratory standardization. In addition, results ER, progesterone receptor (PR) and P53 status (positivity are often difficult to interpret since a number of cases cutoff values of 1%), and ERBB2 status (0–3 þ score as show only moderate overexpression of the protein illustrated by the HercepTest kit scoring guidelines). All and discrepancies in the results are subject to inter- observer variability. In contrast, FISH method is reliable, quantitative and sensitive (Press et al., 2002), Table 3 Characteristics of patients and tumours (n ¼ 213) but is also expensive, time consuming and requires Characteristic No (%)a specialized expertise and equipment. Finally, PCR- based methods are not reliable enough (Ginestier et al., Age (years) in press). As a consequence, there is still no consensus on Median (range) 53 (29, 83) the best method for assessing ERBB2 status. In this Histological type study, we report the potential of gene expression Ductal 165 (77) profiling to establish ERBB2 status, and to identify Lobular 24 (11) among ERBB2 2 þ cases, those with gene amplification Mixed 12 (6) and those without. These observations need confir- Tubular 4 (2) Medullary 3 (2) mation on larger series of tumours, notably for ERBB2 Other 5 (2) 2 þ samples. Nevertheless, the results in favour of gene expression profiling are encouraging and will Axillary lymph node status require comparative studies to other methods (van de Negative 56 (26) Vijver, 2002). Positive 157 (74) Currently however, DNA microarray testing is not Pathological tumour size designed to be used in a routine setting. It requires pT1 58 (27) sophisticated technology and is expensive. These pT2 115 (54) aspects, as well as standardization, will also have to be pT3 40 (19) taken into account during the necessary validation of a SBR grade gene expression-based approach prior to acceptance for I 32 (15) routine clinical application. However, in the near future, II 97 (46) biochip specifically engineered to suit the needs and III 83 (39) requirements of a hospital-based platform may be an Peritumoral vascular invasion interesting alternative method for establishing simulta- Absent 113 (53) neously in one test not only ERBB2 status but a series of Present 99 (47) parameters for breast cancer management. Another important next step will be to identify a GES able to ER status (IHC) Negative 68 (32) predict response to trastuzumab, and to compare it with Positive 142 (68) the one described here. In conclusion, we have identified a GES that can PR status (IHC) accurately identify ERBB2 alteration in breast tumours, Negative 78 (38) as well as provide interesting clues to enhance current Positive 130 (62) understanding of the role of ERBB2 in mammary ERBB2 status (IHC) oncogenesis. Its validity is reinforced by the iterative 0–1+ 162 (78) procedure of identification that we have used, its 2+ 10 (4) successful validation on two independent series of breast 3+ 37 (18) cancer samples and the identity of certain genes that P53 status (IHC) were further validated at the protein level. This Negative 143 (69) signature contains genes that are neighbours of ERBB2 Positive 64 (31) on 17q12, and includes potential regulators and/or ERBB2 status (FISH) downstream effectors of ERBB2 (GATA4) and eventual Negative 38 (56) targets (cadherin, integrins). It may be used both for Positive 30 (44) breast cancer management in clinical settings and as a research tool in academic laboratories. a% of evaluated cases

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2572 tumour sections were reviewed de novo by two pathologists et al., 2002) and the variability of experimental conditions (ECJ, JJ) prior to analysis and all samples contained more using nonlinear rank-based methods (Sabatti et al., 2002), then than 50% tumour cells. The series of 213 samples was divided log-transformed. We first applied supervised analysis to in two sets: a first set of 159 samples, from which was derived identify the optimal set of genes that best discriminated before supervised analysis a ‘learning’ set of 145 samples, and a between ERBB2-negative and -positive breast cancer samples. second set of 54 samples (‘validation’ set). The positivity cutoff of ERBB2 status was defined by protein A consecutive series of 552 women with localized invasive expression using IHC (HercepTest kit, Dako Corporation, breast carcinomas treated at the Institut Paoli-Calmettes Copenhagen, Denmark): positive status was defined as 3 þ between 1981 and 1999 was studied using a first TMA and negative status as 0 or 1 þ . Analysis was carried out in (TMA1). Of the 552 cases studied, 257 were available for two steps: the molecular signature was first derived through ERBB2, GATA4, ER and Ki67 staining. According to the training on the learning set of 145 samples including 116 WHO classification, there were 194 ductal, 26 lobular, 10 ERBB2À and 29 ERBB2 þ cases; samples with ERBB2 status tubular, three medullary carcinomas and 24 other histological 2 þ (n ¼ 10) or unavailable (n ¼ 4) were not included in the types. The median age at diagnosis was 60 years, median age supervised analysis. The signature was then validated on the 60 years (range 25–91). A total of 135 tumours were associated validation set of 54 samples including 46 ERBB2À and eight with lymph node invasion, and 199 were positive for ER. A set ERBB2 þ cases. of 94 tumours (chosen within those analysed by DNA ProfileSoftwaret Corporate (Ipsogen, Marseille) was microarrays and included in the learning set) was included in utilized for all analyses. This program uses a discriminating a second TMA (TMA2). score (DS) (Golub et al., 1999) combined with iterative random permutation tests. The DS was calculated for each Breast tumour cell lines gene as DS ¼ (M1–M2)/(S1 þ S2), where M1 and S1, respectively, represent mean and s.d. of expression levels of the Except for SUM-52, SUM-102 and SUM-149 (a gift of SP gene in subgroup 1 (ERBB2 positive), and M2 and S2 in Ethier, Ann Arbor, MI, USA; http://www.cancer.med.umi- subgroup 2 (ERBB2 negative). Statistical confidence levels ch.edu/breast_cell/clines/clines.html), the breast cancer cell were estimated by bootstrap resampling as previously de- lines (BT-474, HCC38, HCC1395, HCC1569, HCC1937, scribed (Magrangeas et al., 2003) with a false positive rate of 2/ MDA-MB-157, MDA-MB-231, MDA-MB-453, SK-BR-3, 10 000. The program allows the identification of a molecular SK-BR-7, T-47D, UACC-812, and ZR-75-1) were obtained signature as independent of the sample set used as possible. from ATCC (Rockville, MD, USA; http://www.atcc.org/). All Briefly, approximately two-thirds (n ¼ 106) of the samples cell lines were grown according to the recommendations of the from the learning set (n ¼ 145) were randomly selected to supplier. include at least 20 ERBB2 þ cases. They were then submitted to supervised analysis described above. The process was repeated 30 times (30 randomly defined subgroups of 106 RNA extraction samples), thus generating 30 lists of genes. These lists were Total RNA was extracted from frozen tumour samples and then compared and a gene was considered as discriminator if cell lines by standard methods using guanidinium isothiocya- present in at least 25 gene lists out of 30, allowing the nate solution and centrifugation on caesium chloride cushion identification of the most relevant genes, independent of the as previously described (Theillet et al., 1993). RNA integrity sample set used. was controlled by electrophoresis on agarose gels and by Unsupervised hierarchical clustering was applied to investi- Agilent analysis (Bioanalyzer, Palo Alto, CA, USA) before gate relationships between samples and relationships between labelling. genes identified by supervised analysis. The hierarchical clustering was applied to data log-transformed and median- centred on genes using the ProfileSoftwaret Corporate Construction of DNA microarrays program (Ipsogen, Marseille) (average linkage clustering using DiscoveryChip DNA microarrays were prepared in our core uncentred Pearson correlation as similarity metric) and results facility (Ipsogen, Marseille, France; http://www.ipsogen. fr/). were displayed with the same program. PCR products from a total of 9038 Image clones and hundreds 2 of control clones were spotted on 12 Â 8.5 cm nylon filters Construction of tissue microarrays with a Microgrid II robot (Biorobotics Apogent Discoveries, http://www.apogentdiscoveries.com/BioRobotics.asp). Image Two TMA, TMA1 (552 samples) and TMA2 (94 samples), clones represented 3910 ESTsEST and 5125 known genes, were prepared as described (Richter et al., 2000) with slight B3000 of which were related with oncogenesis. Control clones modifications (Ginestier et al., 2002). For each tumour, a included poly(A) þ stretches, plant cDNAs and PCR controls. representative tumour area was carefully selected by histo- Microarray spotting was performed as previously described pathological analysis of a haematoxylin–eosin-stained section (Bertucci et al., 2000). of a donor block. Core cylinders (one for each tumour for TMA2 and three for each tumour for TMA1), with a diameter of 0.6 mm for TMA1 and 2 mm for TMA2, were punched from DNA microarray hybridisations, data analysis and statistical this area and deposited into a recipient paraffin block using a methods specific arraying device (Beecher Instruments, Silver Spring, Hybridisations of microarrays membranes were carried out MD, USA). In addition to tumour tissues, the recipient block with radioactive [a-33P]dCTP-labelled probes made from 5 mg also included normal breast and established breast tumour cell of total RNA from each sample according to described lines to serve as internal controls: BT-474 known to have four- protocols (http:/tagc.univ-mrs. fr/pub/Cancer/). Membranes to eight-fold amplification of the ERBB2 gene, and MCF7, were then washed, exposed to phosphor-imaging plates and whose chromosomes 17 have each one copy of the ERBB2 scanned with a FUJI BAS 5000 machine. Signal intensities gene. Sections (5 mm) of the resulting array block were were quantified with ArrayGauge software (Fuji, Du¨ sseldorf, mounted onto glass slides and used for IHC (TMA1) and Germany), normalized for amount of spotted DNA (Bertucci FISH (TMA2) analyses.

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2573 Antibodies and washed in Dako wash buffer. Slides were pretreated by immersion in Dako pretreatment solution at 971C for 10 min The following antibodies were used for IHC: polyclonal and cooled to room temperature. Slides were then washed in antibody anti-ERBB2 (Dako-HercepTestt), used strictly Dako wash buffer and immersed in Dako pepsin at room following the guidelines described by the manufacturer; goat temperature for 10 min. Pepsin was removed with two changes polyclonal antibody anti-GATA4 (sc-1237, 1 : 50 dilution; of wash buffer. Slides were dehydrated in graded alcohol. Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA), HER2/CEN17 (centromere 17) (10 ml) Probe Mix (Dako) was anti-MIB1/Ki67 (1 : 100 dilution, Dako), anti-ER (clone 6F11, added to the sample area of each section. Sections were 1 : 60 dilution, Novocastra Laboratories). coverslipped and the edges were sealed with rubber cement. Slides were placed on a flat metal surface and heated at 821C Immunohistochemistry for 5 min to codenature the probe and target DNA, and IHC was done on 5-mm sections of TMA1. Briefly, tissues were transferred to a preheated humidified hybridization chamber deparaffinized in Histolemon (Carlo Erba Reagenti, Rodano, to hybridize the probe and DNA for 18 h at 451C. After Italy) and rehydrated in graded alcohol. Antigen retrieval was hybridization, the rubber cement and the coverslips were carried out by incubation at 981C in citrate buffer. Slides were removed from the slides. Sections were washed in wash buffer transferred to a Dako autostainer, except for Dako-Hercep- at 651C then at room temperature. Slides were dehydrated in Testt where guidelines are imposed by the manufacturer. graded alcohol and air-dried in the dark. Nuclei were Staining was carried out at room temperature as follows: after counterstained with 15 ml of DAPI/antifade and coverslipped. washes in phosphate buffer, endogenous peroxidase activity Slides were stored at À41C in the dark for up to 7 days prior to was quenched by treatment with 0.1% H2O2, slides were analysis. preincubated with blocking serum (Dako Corporation) for 10 min, then incubated with the affinity-purified antibody for FISH scoring 1 h. After washes, slides were incubated with biotinylated antibody against rabbit IgG for 20 min followed by strepta- Sections were examined with a fluorescent microscope (Zeiss- divin-conjugated peroxidase (Dako LSABR2 kit). Immunor- Axiophot) using the filter recommended by Dako. The invasive eactive complexes were visualized with the peroxidase lesion selected for the TMA2 was easily localized under the substrate, diaminobenzidine, counterstained with haematox- microscope. Approximately 40 malignant, nonoverlapping cell ylin, and coverslipped using Aquatex (Merck, Darmstadt, nuclei were scored for each case, and included and scored only Germany) mounting solution. Slides were evaluated under a if HER2 and CEN17 signals were clearly detected. A ratio of light microscope by three pathologists (ECJ, JH, JJ). HER2/CEN17 was calculated for each specimen that met this Immunoreactivities for GATA4 and ER were classified by inclusion criteria. ERBB2 was considered as amplified when estimating the percentage (P) of tumour cells showing the FISH ratio HER2/CEN17 was X2.0. Each assay was read characteristic staining (from undetectable level or 0%, to twice by two observers. Specimens were considered negative homogenous staining or 100%) and by estimating the intensity when less than 10% of tumour cells showed amplification of (I) of staining (weak staining or 1, moderate staining or 2, ERBB2. strong staining or 3). Results were scored by multiplying the percentage of positive cells by the intensity, that is by the so- Statistical analysis called quick score (Q)(Q ¼ P Â I; maximum ¼ 300). For Ki67, only the percentage (P) of tumour cells was estimated, since Correlations between hierarchical clustering-based tumour intensity does not vary and for ERBB2, the status was defined groups and molecular and histoclinical parameters were using the Dako scale. Expression levels allowed to group investigated by using the w2 test. Distributions of molecular tumours in two categories: no expression (Q ¼ 0 for GATA4 markers analysed by TMA1 were compared using Fisher exact and ER, Po20 for Ki67 and 0/ þ for ERBB2), and expression test. All P-values were two-sided at the 5% level of (Q40 for GATA4 and ER, PX20 for Ki67 and 2 þ /3 þ for significance. ERBB2). The average of the score of a minimum of two core biopsies was calculated for each case of TMA1. The reliability of the method was assessed by comparison with conventional Acknowledgements sections for the usual prognostic parameters (including ER and This article is dedicated to the memory of P Basset. We thank ERBB2); the value of the kappa test was 0.95 (Ginestier et al., F Birg, G Houvenaegel, D Maraninchi and C Mawas for 2002). encouragements, and M Chaffanet, S Deraco, R Houlgatte, C Nguyen and R Tagett for help and discussions. This work was supported by Ipsogen, Institut Paoli-Calmettes, Inserm, ERBB2 gene amplification detected by FISH Universite´ de la Me´ diterrane´ e, and grants from the ‘Associa- FISH for ERBB2 gene amplification was performed on TMA2 tion pour la Recherche sur le Cancer’, the ‘Ligue Nationale using the Dako ERBB2 FISH PharmDXt Kit, according to contre le Cancer’, the ‘Ministe` re de la Sante´ ’ and the ‘Ministe` re the manufacturer’s instructions. In brief, TMA2 sections were de la Recherche’ (PHRC 2001 No 24-01 and 2002 No 24-04). C baked overnight at 551C, deparaffinized in Histolemon (Carlo G is the recipient of a fellowship from the ‘Ministe` re de la Erba Reagenti, Rodano, Italy), rehydrated in graded alcohol Recherche’.

References

Andersen CL, Monni O, Wagner U, Kononen J, Arthur WT, Noren NK and Burridge K. (2002). Biol. Res., 35, Barlund M, Bucher C, Haas P, Nocito A, Bissig H, 239–246. Sauter G and Kallioniemi A. (2002). Am. J. Pathol., 161, Bertucci F, Houlgatte R, Benziane A, Granjeaud S, Adelaide 73–79. J, Tagett R, Loriod B, Jacquemier J, Viens P, Jordan B,

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2574 Birnbaum D and Nguyen C. (2000). Hum. Mol. Genet., 9, Magrangeas F, Nasser V, Avet-Loiseau H, Loriod B, Decaux 2981–2991. O, Granjeaud S, Bertucci F, Birnbaum D, Nguyen C, Bertucci F, Nasser V, Granjeaud S, Eisinger F, Adelaide J, Harousseau JL, Bataille R, Houlgatte R and Minvielle S. Tagett R, Loriod B, Giaconia A, Benziane A, Devilard E, (2003). Blood, 101, 4998–5006. Jacquemier J, Viens P, Nguyen C, Birnbaum D and Oh JJ, Grosshans DR, Wong SG and Slamon DJ. (1999). Houlgatte R. (2002). Hum. Mol. Genet., 11, 863–872. Nucleic Acids Res., 27, 4008–4017. Bertucci F, Viens P, Hingamp P, Nasser V, Houlgatte R and Patient RK and McGhee JD. (2002). Curr. Opin. Genet. Dev., Birnbaum D. (2003a). Int. J. Cancer, 103, 565–571. 12, 416–422. Bertucci F, Viens P, Tagett R, Nguyen C, Houlgatte R and Pauletti G, Dandekar S, Rong H, Ramos L, Peng H, Seshadri Birnbaum D. (2003b). Lab. Invest., 83, 305–316. R and Slamon DJ. (2000). J. Clin. Oncol., 18, 3651–3664. Bilous M, Dowsett M, Hanna W, Isola J, Lebeau A, Moreno Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, A, Penault-Llorca F, Ruschoff J, Tomasic G and Van De Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, Vijver M. (2003). Mod. Pathol., 16, 173–182. Fluge O, Pergamenschikov A, Williams C, Zhu SX, Lonning Callagy G, Cattaneo E, Daigo Y, Happerﬁeld L, Bobrow LG, PE, Borresen-Dale AL, Brown PO and Botstein D. (2000). Pharoah PD and Caldas C. (2003). Diagn. Mol. Pathol., 12, Nature, 406, 747–752. 27–34. Pietras RJ, Arboleda J, Reese DM, Wongvipat N, Pegram Dressman MA, Baras A, Malinowski R, Alvis LB, Kwon I, MD, Ramos L, Gorman CM, Parker MG, Sliwkowski MX Walz TM and Polymeropoulos MH. (2003). Cancer Res., 63, and Slamon DJ. (1995). Oncogene, 10, 2435–2446. 2194–2199. Platzer P, Upender MB, Wislon K, Willis J, Lutterbaugh J, Frade R, Balbo M and Barel M. (2002). Oncogene, 21, Nosrati A, Willson JKV, mack D, Ried T and Markowitz S. 861–866. (2002). Cancer Res., 62, 1134–1138. Garratt AN, Ozcelik C and Birchmeier C. (2003). Trends Pollack JR, Sorlie T, Perou CM, Rees CA, Jeffrey SS, Lonning Cardiovasc. Med., 13, 80–86. PE, Tibshirani R, Botstein D, Borresen-Dale AL and Gilde AJand Van Bilsen M. (2003). Acta Physiol. Scand., 178, Brown PO. (2002). Proc. Natl. Acad. Sci. USA, 99, 425–434. 12963–12968. Ginestier C, Charafe-Jauffret E, Bertucci F, Eisinger F, Geneix Press MF, Slamon DJ, Flom KJ, Park J, Zhou JY and J, Bechlian D, Conte N, Ade´ laide J, Toiron Y, Nguyen C, Bernstein L. (2002). J. Clin. Oncol., 20, 3095–3105. Viens P, Mozziconacci MJ, Houlgatte R, Birnbaum D and Rampaul RS, Pinder SE, Gullick WJ, Robertson JF and Ellis Jacquemier J. (2002). Am. J. Pathol., 161, 1223–1233. IO. (2002). Crit. Rev. Oncol. Hematol., 43, 231–244. Ginestier C, Charafe-Jauffret E, Penault-Llorca F, Ade´ laı¨ de J, Revillion F, Bonneterre Jand Peyrat JP.(1998). Eur. J. Geneix J, Chaffanet M, Mozziconacci MJ, Hassoun J, Cancer, 34, 791–808. Birnbaum D and Jacquemier J: J. Pathol., in press. Richter J, Wagner U, Kononen J, Fijan A, Bruderer J, Schmid Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, U, Ackerman D, Maurer R, Alund G, Kno¨ nagel H, Rist M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, Wilber K, Anabitarte M, Hering F, Hardmeier T, Scho¨ nen- Bloomﬁeld CD and Lander ES. (1999). Science, 286, berger A, Flury R, Ja¨ ger P, Fehr JL, Schrami P, Moch H, 531–537. Mihatsch MJ, Gasser T, Kallioniemi OP and Sauter G. Han J, Lee JD, Jiang Y, Li Z, Feng L and Ulevitch RJ. (1996). (2000). Am. J. Pathol., 157, 787–794. J. Biol. Chem., 271, 2886–2891. Sabatti C, Karsten SL and Geschwind DH. (2002). Math. Hayes DF and Thor AD. (2002). Semin. Oncol., 29, 231–245. Biosci., 176, 17–34. Holbro T, Civenni G and Hynes NE. (2003). Exp. Cell Res., Shen TL, Han DC and Guan JL. (2002). J. Biol. Chem., 277, 284, 99–110. 29069–29077. Hyman E, Kauraniemi P, Hautaniemi S, Wolf M, Mousses S, Slamon DJ, Clark GM, Wong SG, Levin WJ, Ullrich A and Rozenblum E, Ringner M, Sauter G, Monni O, Elkahloun McGuire WL. (1987). Science, 235, 177–182. A, Kallioniemi OP and Kallioniemi A. (2002). Cancer Res., Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen 62, 6240–6245. H, Hastie T, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen Kauraniemi P, Ba¨ rlund M, Monni O and Kallioniemi A. T, Quist H, Matese JC, Brown PO, Botstein D, Eystein (2001). Cancer Res., 61, 8235–8240. Lonning P and Borresen-Dale AL. (2001). Proc. Natl. Acad. Keshgegian AA. (1995). Breast Cancer Res. Treat., 35, Sci. USA, 98, 10869–10874. 201–210. Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel Konecny G, Pauletti G, Pegram M, Untch M, Dandekar S, A, Deng S, Johnsen H, Pesich R, Geisler S, Demeter J, Aguilar Z, Wilson C, Rong HM, Bauerfeind I, Felber M, Perou CM, Lonning PE, Brown PO, Borresen-Dale AL Wang HJ, Beryt M, Seshadri R, Hepp H and Slamon DJ. and Botstein D. (2003). Proc. Natl. Acad. Sci. USA, 100, (2003). J. Natl. Cancer Inst., 95, 142–153. 8418–8423. Korsching E, Packeisen J, Agelopoulos K, Eisenacher M, Voss Spencer KS, Graus-Porta D, Leng J, Hynes NE and Klemke R, Isola J, van Diest PJ, Brandt B, Boecker W and Buerger RL. (2000). J. Cell. Biol., 148, 385–397. H. (2002). Lab. Invest., 82, 1525–1533. Tan M, Yao Jand Yu D. (1997). Cancer Res., 57, 1199–1205. Kumar-Sinha C, Woods Ignatoski K, Lippman ME, Ethier SP Theillet C, Ade´ laı¨ de J, Louason G, Bonnet-Dorion F, and Chinnaiyan AM. (2003). Cancer Res., 63, 132–139. Jacquemier J, Adnane J, Longy M, Katsaros D, Sismondi Kuo CT, Morrisey EE, Anandappa R, Sigrist K, Lu MM, P, Gaudray P and Birnbaum D. (1993). Genes Chromosomes Parmacek MS, Soudais C and Leiden JM. (1997). Genes Cancer, 7, 219–226. Dev., 11, 1048–1060. Tiwari RK, Mukhopadhyay B, Telang NT and Osborne MP. Leyland-Jones B. (2002). Lancet Oncol., 3, 137–144. (1991). Anticancer Res., 11, 1383–1388. Mackay A, Jones C, Dexter T, la Silva R, Bulmer K, Jones A, Tomasetto C, Regnier C, Moog-Lutz C, Mattei MG, Chenard Simpson P, Harris RA, Jat PS, Neville AM, Reis LFL, MP, Lidereau R, Basset P and Rio MC. (1995). Genomics, Lakhani SR and O’Hare MJ. (2003). Oncogene, 22, 28, 367–376. 2680–2688. van de Vijver M. (2002). Oncology, 63 (Suppl 1), 33–38.

Oncogene ERBB2 GES in breast cancers F Bertucci et al 2575 van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Willis S, Hutchins AM, Hammet F, Ciciulla J, Soo WK, White Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton D, van der Spek P, Henderson MA, Gish K, Venter DJand MJ, Parrish M, Atsma D, Witteveen A, Glas A, Delahaye L, Armes JE. (2003). Genes Chromosomes Cancer, 36, 382–392. van der Velde T, Bartelink H, Rodenhuis S, Rutgers ET, Wilson KS, Roberts H, Leek R, Harris AL and Geradts J. Friend SH and Bernards R. (2002). N. Engl. J. Med., 347, (2002). Am. J. Pathol., 161, 1171–1185. 1999–2009. Yanazume T, Hasegawa K, Wada H, Morimoto T, Abe M, van’t Veer LJ, Dai H, van de Vijver M, He YD, Hart AA, Kawamura T and Sasayama S. (2002). J. Biol. Chem., 277, Mao M, Peterse HL, van der Kooy K, Marton MJ, 8618–8625. Witteveen AT, Schreiber GJ, Kerkhoven RM, Roberts C, Yang RB, Ng CK, Wasserman SM, Colman SD, Shenoy S, Linsley PS, Bernards R and Friend SH. (2002). Nature, 415, Mehraban F, Komuves LG, Tomlinson JE and Topper JN. 530–535. (2002). J. Biol. Chem., 277, 46364–46373.

Oncogene