ONCOGENOMICS Mammary Luminal Epithelial Cells

Home , KIFC3, Membrane glycoproteins

cDNA microarray analysis of genes associated with ERBB2 (HER2/neu) overexpression in human mammary luminal epithelial cells

Alan Mackay*,1,6, Chris Jones2,6, Tim Dexter2, Ricardo LA Silva3, Karen Bulmer2, Allison Jones2, Peter Simpson2, Robert A Harris1, Parmjit S Jat4, A Munro Neville1, Luiz FL Reis3, Sunil R Lakhani2,5 and Michael J O’Hare1

1LICR/UCL Breast Cancer Laboratory, University College London, London, UK; 2The Breakthrough Toby Robins Breast Cancer Research Centre, Institute of Cancer Research, London, UK; 3Ludwig Institute for Cancer Research, Sao Paolo, Brazil; 4Ludwig Institute for Cancer Research, University College London, London, UK; 5Royal Marsden Hospital, London, UK

To investigate changes in gene expression associated with the disease within their lifetime. Overexpression of the ERBB2, expression profiling of immortalized human proto-oncogene ERBB2 (HER2/neu) is observed in 25– mammary luminal epithelial cells and variants expressing 30% of all such cancers, and is an established adverse a moderate and high level of ERBB2 has been carried out prognostic factor (Slamon et al., 1987, 1989; Ross and using cDNA microarrays corresponding to approximately Fletcher, 1999; Menard et al., 2001) yielding a median 6000 unique genes/ESTs. A total of 61 significantly up- or survival of 3years, compared with 6–7 years when downregulated (>2.0-fold) genes were identified and unassociated with ERBB2. Overexpression also corre- further validated by RT–PCR analysis as well as lates with tumor size, lymph node metastases, high microarray comparisons with a spontaneously ERBB2- nuclear grade, high percentage of S-phase cells, aneu- overexpressing breast cancer cell line and ERBB2-positive ploidy and estrogen receptor (ER) and progesterone primary breast tumors. The expression and clinical receptor (PR) negativity (Ross and Fletcher, 1998). relevance of proteins predicted to be associated with ERBB2 overexpression may also predict resistance to ERBB2 overexpression in breast cancers were analysed both chemotherapy and endocrine therapy, although together with their clinical relevance by antibody screen- this remains controversial (Houston et al., 1999; Miles ing using a tissue array. Differentially regulated genes et al., 1999; Yu and Hung, 2000). The ERBB2 receptor include those involved in cell–matrix interactions including has become the first oncogene product to be targeted for proline 4-hydroxylase (P4HA2), galectin 1 (LGALS1) breast cancer therapy, by the humanized anti-ERBB2 and galectin 3 (LGALS3), fibronectin 1 (FN1) and p- monoclonal antibody trastuzumab (Herceptin) (Baselga cadherin (CDH3), and cell proliferation (CRIP1, et al., 1996; Cobleigh et al., 1999; Slamon et al., 2001). IGFBP3) and transformation (S100P, S100A4). A There is much evidence to suggest that ERBB2 is not number of genes associated with MYC signalling were the only gene activated by amplification at 17q12–q21 in also differentially expressed, including NDRG1, USF2 breast cancer (Tomasetto et al., 1995; Bieche et al., 1996; and the epithelial membrane proteins 1 and 3 (EMP1, Kauraniemi et al., 2001). Owing to the potential EMP3). These data represent profiles of the transcrip- contribution of these other genes within the ERBB2 tional changes associated with ERBB2-related pathways amplicon (17q12–q21) to breast tumorigenesis in vivo,it in the breast, and identify novel and potentially useful has been difficult to determine the precise role played by targets for prognosis and therapy. ERBB2-mediated signalling, and to establish the me- Oncogene (2003) 22, 2680–2688. doi:10.1038/sj.onc.1206349 chanisms by which anti-ERBB2 therapeutic agents exert their antiproliferative effects. Keywords: ERBB2 (HER2/neu); breast cancer; cDNA We have previously established a model of ERBB2 microarray; expression profiling; tissue array overexpression in conditionally immortalized human ONCOGENOMICS mammary luminal epithelial cells. Transfection of the HB4a human mammary epithelial cell line with ERBB2 cDNA resulted in two cell lines, C3.6 and C5.2, which Introduction showed ‘moderate’ and ‘high’ overexpression of ERBB2, respectively (Harris et al., 1999). These cells do not Breast cancer is one of the most common malignancies exhibit amplification at 17q12–q21, and thereby provide in the Western world, with one in 12 women developing an opportunity to examine ERBB2-related signalling independent of the effects of other related coamplified *Correspondence: A Mackay, LICR/UCL Breast Cancer Laboratory, genes. Charles Bell House, Riding House Street, London W1W 7EJ, UK; Accordingly, we have investigated the transcriptional E-mail: [email protected] changes associated with ERBB2 overexpression in 6These authors contributed equally to this work Received 16 October 2002; revised 23December 2002; accepted 3 breast epithelial cells using cDNA microarrays employ- January 2003 ing 9930 cDNA clones representing approximately 6000 ERBB2-related gene expression using microarrays A Mackay et al 2681 unique genes/ESTs. A total of 61 statistically significant expressed genes, 132 (46.7%) were found to be differentially expressed genes (greater than 2.0-fold) significant in both C5.2 and C3.6, including LGALS1, associated with ERBB2 overexpression were identified. S100P, NDRG1, CDH3, EMP1, FN1, IGFBP3 and Their clinical relevance was demonstrated at the protein LGALS3. level by carrying out immunohistochemistry on a tissue We also examined a breast cancer cell line, BT474, array comprising 48 ERBB2-positive and 47 ERBB2- using our cDNA microarrays. BT474 contains the negative breast carcinomas. 17q12–q21 amplicon, and allows for comparison with our ERBB2-transfected cell lines which do not. BT474 mRNA was reverse transcribed and hybridized with HB4a for direct comparison with the C3.6 and C5.2 Results data, in two replicated reverse-labelled experiments. In cDNA microarray analysis total, 274 statistically significant (SAM, 1% false discovery), differentially expressed (greater than 1.7- Using the normalized data from our 9930 clone cDNA fold) genes are published in supplementary table S3. microarrays, gene lists were generated by SAM (version Primary breast tumors 731 and 732 were also analysed 1.12) for differentially expressed genes in the compar- by cDNA microarrays, hybridized against the normal isons between C3.6 (moderately overexpressing ERBB2) luminal epithelial HB4a cells. This analysis generated and HB4a, C5.2 (highly overexpressing ERBB2) and lists of 187 and 91 differentially expressed genes, HB4a, as well as between C3.6 and C5.2. Significance respectively, corresponding to a total of 152 unique was assigned using a False Discovery Rate threshold of genes (supplementary Table S3). 1% in conjunction with a ratio threshold of 1.7. A set of The Venn diagram in Figure 2 shows a summary of 143significantly upregulated and 140 significantly overlaps of differentially expressed genes between the downregulated genes from the three comparisons was ERBB2-transfected breast luminal epithelial cell line generated. In individual experiments, 61 genes were C5.2, the ERBB2-amplicon-containing breast cancer cell found to be differentially expressed in C3.6 compared to line BT474, and ERBB2-overexpressing invasive ductal HB4a, 180 between C5.2 and HB4a and 91 between breast carcinomas examined by cDNA microarrays. The C5.2 and C3.6 (mean array ratios, SAM scores and s.d.’s only gene to segregate with ERBB2 (differential in all for the list of 283are provided in supplementary Table S2). analyses) was IGFBP3. Four genes, including FN1, By performing the three-way comparisons, the C5.2 PTRF and TIP-1, overlapped between C3.6/C5.2 and versus C3.6 ratios can be predicted from the individual BT474. None of the differentially expressed genes in the variant versus parental experiments. The actual : pre- C5.2 and C3.6 experiments are found at the 17q12–q21 dicted plot is shown in Figure 1, and gives a correlation amplicon. Eight genes that were found to be differential coefficient of 0.968, allowing for an array-based control through analysis of C3.6/C5.2 were also up- or down- of the reproducibility of the data. For journal format regulated in the tumor samples, but were not discovered brevity, a ratio threshold of 2.0 in the C5.2 versus HB4a in the BT474 breast cancer cell line. These include the comparison was used to produce a list of 36 upregulated upregulated genes LGALS1, CRIP1, VIM and PRDX2, and 25 downregulated unique genes, ranked according and the downregulated genes RPL17, MYL9, ADORA3 to ratio value (Table 1). Of the 283differentially and GTF3A.

Figure 1 Predicted versus actual array ratio plot for C5.2 versus C3.6 comparison, calculated from individual comparisons against parental cell line HB4a

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2682 Table 1 List of differentially expressed genes by cDNA microarray analysis C5.2 versus C3.6 versus C5.2 versus Symbol Ensembl Name HB4a HB4a C3.6 ERBB2 ENSG00000141736 v-erb-b2 erythroblastic leukemia viral oncogene homolog 2 15.72 1.13 6.47 LGALS1 ENSG00000100097 Lectin, galactoside-binding, soluble, 1 (galectin1) 12.39 1.41 6.01 PRO2605 ENSG00000130600 Hypothetical protein PRO2605 10.87 0.97 8.18 CRIP1 ENSG00000133512 Cysteine-rich protein 1 (intestinal) 9.30 1.53 5.07 S100P ENSG00000163993 S100 calcium binding protein P 5.48 6.16 0.70 CPS1 ENSG00000021826 Carbamoyl-phosphate synthetase 1, mitochondrial 4.33 3.08 1.14 SPAG1 ENSG00000104450 Sperm-associated antigen 1 3.97 4.88 0.67 VIM ENSG00000026025 Vimentin 3.59 1.03 2.86 NDRG1 ENSG00000104419 N-myc downstream regulated gene 1 3.55 1.32 2.62 CDH3 ENSG00000062038 Cadherin 3, type1, P-cadherin (placental) 3.30 1.85 1.25 S100A4 ENSG00000160673 S100 calcium binding protein A4 3.21 1.27 1.82 SLPI ENSG00000124107 Secretory leukocyte protease inhibitor (antileukoproteinase) 2.87 2.70 0.76 PRDX2 ENSG00000167815 Peroxiredoxin 2 2.76 1.79 1.43 PFKFB3 ENSG00000170525 6-Phosphofructo-2-kinase/fructose-2,6-biphosphatase 3 2.55 1.36 1.37 MGC11242 ENSG00000167183 Hypothetical protein MGC11242 2.49 2.59 0.84 SLC1A 6 ENSG00000105143 Solute carrier family 1 member 6 2.41 1.60 1.44 EDN2 ENSG00000127129 Endothelin 2 2.37 1.53 1.24 SIAT4C ENSG00000110080 Sialyltransferase 4C 2.33 1.71 1.06 SQLE ENSG00000104549 Squalene epoxidase 2.31 1.52 1.46 NNMT ENSG00000166741 Nicotinamide N-methyltransferase 2.30 2.69 0.80 HSPB1 ENSG00000106211 Heat-shock 27 kDa protein1 2.27 1.61 1.28 AGR2 ENSG00000106541 Anterior gradient 2 homolog (Xenopus laevis) 2.25 2.22 0.87 FLJ12443 ENSG00000153395 Hypothetical protein FLJ12443 2.25 1.82 1.09 BNIP3 ENSG00000104765 BCL2/adenovirus E1B 19 kDa interacting protein 3 2.23 1.31 1.89 P4HA2 ENSG00000072682 Proline 4-hydroxylase, alpha polypeptideII 2.21 1.27 1.46 STMN3 ENSG00000171780 Stathmin-like 3 2.17 1.55 0.99 RPL17 ENSG00000174228 Ribosomal protein L17 2.14 1.24 1.47 KIFC3 ENSG00000140859 Kinesin family member C3 2.08 1.02 1.52 DNAJB6 ENSG00000105993 DnaJ (Hsp40) homolog, subfamily B, member 6 2.07 1.72 1.15 PFN2 ENSG00000070087 Profilin 2 2.05 2.10 0.87 EMP1 ENSG00000134531 Epithelial membrane protein 1 2.05 2.13 0.94 TGM1 ENSG00000092295 Transglutaminase 1 2.04 2.07 0.86 ALDOC ENSG00000109107 Aldolase C, fructose-bisphosphate 2.02 0.78 1.59 NPTX2 ENSG00000106236 Neuronal pentraxin II 2.01 0.98 2.27 EMP3 ENSG00000142227 Epithelial membrane protein 3 2.00 1.37 1.18 ZFP36L2 ENSG00000152518 Zinc-finger protein 36, C3H type-like 2 2.00 1.32 1.19

MYL9 ENSG00000101335 Myosin, light polypeptide 9, regulatory 0.18 0.49 0.32 RRAD ENSG00000166592 Ras-related associated with diabetes 0.22 0.40 0.88 SPARC ENSG00000113140 Secreted protein, acidic, cysteine-rich (osteonectin) 0.27 0.53 0.74 FN1 ENSG00000115414 Fibronectin 1 0.27 0.29 0.94 IGFBP3 ENSG00000146674 Insulin-like growth factor binding protein 3 0.29 0.33 0.92 TAGLN ENSG00000149591 Transgelin 0.29 0.76 0.31 PTRF Polymerase I and transcript release factor 0.40 0.53 0.77 SPINT2 ENSG00000167642 Serine protease inhibitor, Kunitz type, 2 0.42 1.01 0.43 ADORA3 ENSG00000121933 Adenosine A3 receptor 0.44 0.47 0.65 SEC61A1 ENSG00000058262 Protein transport protein SEC61 alpha subunit isoform 1 0.45 0.65 1.00 p100 ENSG00000106334 EBNA-2 coactivator (100kDa) 0.46 0.55 0.93 MME ENSG00000114802 Membrane metallo-endopeptidase (CALLA, CD10) 0.47 0.53 1.07 GTF3A ENSG00000122034 General transcription factor IIIA 0.47 0.61 0.76 PCK2 ENSG00000100889 Phosphoenolpyruvate carboxykinase 2 (mitochondrial) 0.47 0.67 0.82 GAS6 Growth arrest-specific 6 0.48 0.53 0.91 PSMD8 ENSG00000099341 Proteasome (prosome, macropain) 26S subunit 0.48 0.96 0.52 PEAS ENSG00000124702 Testis intracellular mediator protein 0.48 0.93 0.60 stSG89556 ENSG00000169261 40S ribosomal protein 0.49 0.89 0.60 RPS10 ENSG00000123445 Ribosomal protein S10 0.49 0.88 0.59 CHST2 ENSG00000175040 Carbohydrate sulfotransferase 2 0.50 0.58 1.09 WFS1 ENSG00000109501 Wolfram syndrome 1 (wolframin) 0.50 0.72 0.74 NR1H2 ENSG00000131408 Nuclear receptor subfamily 1, group H, member 2 0.50 0.62 0.99 C4BPB ENSG00000123843 Complement component 4 binding protein, beta 0.50 0.64 0.90 TIP-1 ENSG00000108392 Tax interaction protein 1 0.50 0.79 0.57 USF2 ENSG00000105698 Upstream transcription factor 2, c-fos interacting 0.50 1.12 0.49

Upregulated genes are shown in red, downregulated in green. Array ratio figures in italics reach formal statistical significance (SAM, 1% false discovery rate)

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2683

Figure 2 Venn diagram showing overlap of genes differentially expressed in C5.2/C3.6, BT474 breast cancer cell line and ERBB2- overexpressing tumors 731/732 against HB4a cells. Genes labelled green were upregulated, those labelled red were downregulated with respect to HB4a

a a a a 4 .6 .2 4 .6 .2 4 .6 .2 4 .6 .2 B 3 5 B 3 5 B 3 5 B 3 5 H C C H C C H C C H C C ERBB2 ERBB2 ERBB2 ERBB2

LGALS1 PRDX2 MYL9 PSMD8

WFS1 CRIP1 PFKFB3 SPARC

C4BPB S100P MGC11242 FN1

USF2 CPS1 SQLE IGFBP3

CCND3 VIM NNMT TAGLN

LGALS3 NDRG1 BNIP3 ADORA3

S100A4 EMP1 p100 MMP7

SLPI MYC MME MT1L

GAPDH GAPDH GAPDH GAPDH

Figure 3 RT–PCR analysis of differentially expressed genes from cDNA microarray data

Conﬁrmation of differential expression the relative levels of expression in the cell lines. In all, 51 out of 56 genes (91%) analysed showed a pattern of For a subset of differentially expressed genes generated transcript abundance consistent with the microarray from the cDNA microarray experiments, RT–PCR data in the cell lines C3.6 and C5.2. Figure 3 shows analysis of cell line RNA was carried out to conﬁrm representative gels for a subset of 16 upregulated genes

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2684 and 16 downregulated genes of genes analysed by RT– significance (Po0.05) by Fischer’s exact test. FN1, PCR. showing predominantly stromal cell localization, ex- hibited a statistically significant inverse correlation with our microarray data. One ERBB2-positive core (0049) Breast tumor tissue array showed a pattern of immunohistochemical staining entirely consistent with our ERBB2-postive model Where there were appropriate commercial antibodies system for all antibodies tested. A full list of all tumors available, relative protein levels of the differentially on the array and the immunohistochemical results are expressed mRNA levels associated with our expression provided in supplementary Table S4. profiling data were probed by immunohistochemistry on a breast tumor tissue array. As representative examples, antibodies to HSPB1 (HSP27), LGALS3, FN1, ERBB2, MYC and IGFBP3, were obtained and were analysed by Discussion immunohistochemistry on a tissue array containing 48 ERBB2-positive and 47 grade-matched ERBB2-negative The ERBB2 protein is overexpressed owing to gene invasive ductal breast carcinomas (Figure 4). Statistical amplification in 25–30% of invasive ductal breast analysis was carried out examining associations between carcinomas, and correlates with a poor clinical prog- antibody staining in the ERBB2-positive and -negative nosis. However, comparatively little is known mechan- tumor groups (Table 2). Staining with all antibodies istically about the exact role of ERBB2 in breast except FN1 showed a pattern of over- or underexpres- tumorigenesis, other than that directly attributable to sion in the ERBB2-positive/negative tumor groups its role as a member of the ERBB receptor tyrosine consistent with cDNA microarray data from our cell kinase family. Investigations into ERBB2 action provide lines, although only LGALS3achieved formal statistical a paradigm for how molecular analyses can lead to the development of a novel clinical therapy (trastuzumab, Herceptin) targeted against a single gene product. However, although a humanized antibody against the receptor (Herceptin) showed some clinical benefit to around 25% of patients with ERBB2-positive metastatic breast cancer (Pegram et al., 1998), most were not helped. Since little is known about the more global effects of ERBB2 overexpression in breast cancer and how these may impact on attempts to antagonize ERBB2 activity, we have undertaken expression profiling of ERBB2 overexpressing cell lines derived from a ERBB2 conditionally immortalized human mammary luminal epithelial cells, as a longitudinal model of ERBB2 transformation. This analysis revealed a large number of genes consistently up- or downregulated coordinately with ERBB2 overexpression. Some of these are known to be associated with transformation and proliferation in breast and other tissues, while others represent novel markers of ERBB2 function. Examples of known coordinately upregulated genes include members of the S100 calcium-binding protein family (S100P, S100A4), which have been previously HSP27 LGALS3 implicated in breast tumorigenesis (Guerreiro Da Silva et al., 2000; Mazzucchelli, 2002) and specifically related Figure 4 Tissue array containing 48 ERBB2-positive and 47 ERBB2-negative primary breast tumors. Immunohistochemical to ERBB2 status (Perou et al., 2000; Rudland et al., staining for ERBB2, HSP27 and LGALS3(all same tumor core, 2000; Sørlie et al., 2001). These were upregulated in our HB049) is show model system, and therefore provide a validation of the

Table 2 Tissue array data. Percentage of ERBB2-positive and -negative cases showing immunoreactivity with selected antibodies chosen from differentially expressed genes from cDNA microarray analysis ERBB2 positive (48) ERBB2 negative (47) Signiﬁcance (Fischer’s exact)

ERBB2 48/48 (100%) 0/47 (0%) Po0.001 MYC 37/40 (92.5%) 30/36 (83.3%) P=0.190 IGFBP3 36/37 (97.3%) 36/36 (100%) P=0.507 LGALS325/43(58.1%) 32/41(78.0%) P=0.042* FN1 11/36 (30.6%) 4/37 (10.8%) P=0.035* HSP27 18/42 (42.9%) 9/38 (23.7%) P=0.057

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2685 system and method of analysis. The most significant An intriguing aspect of our data involves MYC,a upregulated gene (after ERBB2 itself), not previously transcription factor whose amplification and overex- associated with ERBB2 status, was LGALS1 (galectin 1, pression correlates to metastasis and poor prognosis in GBP). The galectins are a family of beta-galactoside- breast cancer patients (Scorilas et al., 1999). MYC was binding proteins implicated in modulating cell–cell and also found to be differentially expressed in our model cell–matrix interactions (Hughes, 2001). LGALS1 has system coordinately with ERBB2 expression. A recent been reported to act as an autocrine negative growth report has shown that coamplification of MYC and factor that regulates cell proliferation (Goldstone and ERBB2 is associated with a significant reduction in Lavin, 1991; Baldini et al., 1993) and to bind oncogenic patient survival (Cuny et al., 2000). As in addition to H-Ras to mediate Ras membrane anchorage and cell MYC itself, a number of other genes known to interact transformation (Paz et al., 2001). Interestingly, LGALS3 with MYC were also differentially expressed in our (galectin 3) was found to be significantly downregulated model system, for example, USF2, NMYC and the in our model, agreeing with published data (Andre et al., NMYC downstream regulated gene NDRG1 (Cap43), 1999) indicating a correlation of increased LGALS1 and further implicating myc family member signalling in reduced LGALS3 in breast carcinomas with metastatic ERBB2-associated breast tumorigenesis. Other MYC- propensity. LGALS3 is thought to regulate many associated genes, epithelial membrane proteins 1 biological processes including cell adhesion, migration, (EMP1, TMP, CL20) and 3( EMP3, YMP) were also cell growth, tumor progression, metastasis and apopto- significantly upregulated in our model system. EMP1 sis (Moon et al., 2001). The mechanisms by which the and EMP3 are members of the PMP22/EMP/MP20 galectins exert these effects remain largely unknown family of membrane glycoproteins, involved in squa- (Perillo et al., 1998), and to our knowledge, this is the first mous cell differentiation (Chen et al., 1997; Ben-Porath report associating galectin up- or downregulation with et al., 1998, 1999; Liehr et al., 1999). Their precise ERBB2 expression in breast carcinomas and thus possibly functions in the breast are unknown but associations contributing significantly to its impact on prognosis. with cell proliferation and tumorigenesis have been Other differentially expressed genes associated with proposed, and differential expression of EMP1 has been cell–matrix interactions include matrix metalloprotei- reported in gastric cancer (Hippo et al., 2001) and in nases, P4HA2 (proline 4-hydroxylase), FN1 (fibronectin estrogen-resistant MCF7ADR breast cancer cells 1) and CDH3 (p-cadherin). Silencing of fibronectin (Schiemann et al., 1997). Given its trans-membrane expression has been described as one of the key localization within epithelial cells, EMP1 (and possibly mechanisms underlying metastatic behavior in adeno- EMP3) represents novel and potentially useful targets carcinomas (Werbajh et al., 1998), and has previously for prognosis and targeting for therapy. been reported to be crucial for ERBB2-induced invasive In summary, expression profiling of our model cell potentiation (Ignatoski et al., 2000). P-cadherin, a lines has revealed 61 genes that are significantly up- or calcium-dependent cell–cell adhesion glycoprotein, was downregulated by greater than 2.0-fold in response to also upregulated in erbB2-overexpressing cell lines and ERBB2 overexpression in vitro, independent of other has been previously shown to correlate with poor genes on the 17q12–q21 amplicon. We have attempted, prognosis in breast cancers (Peralta Soler et al., 1999; by means of RT–PCR screening of breast tumor Gamallo et al., 2001). One of the most significant samples and immunohistochemistry on breast tumor overexpressed genes in our study was hypothetical tissue arrays, to demonstrate a potential clinical role for protein PRO2605, which shows strong sequence simi- the up- or downregulation of some of these genes and larity to the laminin receptor protein 1. High expression their protein products. Further validation will involve of laminin has also been previously identified as a both retrospective studies of tumor samples, and marker of poor prognosis in breast cancer (De Iorio targeted inhibition in this in vitro model. It is hoped, et al., 2001). therefore, that some of the genes identified in this study One of the most significantly downregulated genes will provide new targets for the development of new associated with ERBB2 overexpression was insulin-like treatment regimens to enhance the existing methods of growth factor binding protein 3( IGFBP3). IGFBP3 is a abrogating ERBB2 activity in overexpressing breast major determinant of circulating levels of IGFs, which cancers and thereby improving their efficacy and scope. are potent mitogenic agents in breast cancer cells (Stoll, 1997). Breast tumors are reported to show decreased levels of IGFBP3, and this correlates inversely with ER Materials and methods status (Rocha et al., 1996). IGFBP3 has been purported Tissue culture and tumor preparations to be a proapoptotic protein, regulated by blocking IGF–1R regulation of HIF1 activation of p53(Shen and The cell lines C5.2 and C3.6 were established by transfecting Glazer, 1998). Loss of IGFBP3, and subsequent the normal mammary luminal epithelial cell line HB4a with increased IGF–1R signalling, has been reported to play ERBB2 cDNA derived from the BT474 breast cancer cell line (Harris et al., 1999). HB4a, C3.6 and C5.2 were cultured in a role in resistance to trastuzumab (Herceptin) in vitro RPMI-1640 with 10% (v/v) FCS, 5 mg/ml hydrocortisone, (Lu et al., 2001), suggesting that strategies targeting insulin, 2 mm glutamine, 100 IU/ml penicillin and 100 mg/ml these pathways may prevent or delay development of streptomycin. Cell lines MCF-7, BT474 and SKBR3were trastuzumab resistance in ERBB2-overexpressing breast grown in Dulbecco’s modified Eagle’s medium supplemented tumors. with 10% FCS 100 IU/ml penicillin and 100 mg/ml streptomy-

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2686 cin. All cells were cultured in an atmosphere of 5% (v/v) and plotted on an MA plot (Yang et al., 2002), shown in carbon dioxide in a humidified incubator at 36.51C. Total supplementary Figure S1. The interquartile range (IQR) of the RNAs were extracted from cultured cells by the guanadinium - average intensity (A) values for each hybridization was used as phenyl chloroform extraction method (Chomczynski and a quality measure. Lower IQR values correlate with higher Sacchi, 1987). average background, a higher proportion of flagged spots and Breast tumors were collected directly after surgical removal, imply a lower resolution of the expression measurements. The flash frozen in liquid nitrogen prior to storage at À801C. To six hybridizations for each cell line comparison, which had the extract RNA, tumors were pulverized under liquid nitrogen highest IQR values and which preserved balanced dye labeling, and homogenized in a high-speed homogeniser (Ultra-Turrax) were selected for further analysis. More than half of the in guanadinium isothiocyanate prior to phenol chloroform hybridizations used had IQR values greater than 1.0 and none extraction as above (Chomczynski and Sacchi, 1987). Tumors selected were below 0.5. As some of the better hybridizations 731 and 732 were obtained from Hospital do Cancer AC showed obvious dye bias, the log expression ratios were Camargo, Sao Paulo, Brazil, and RNA was prepared for normalized using Lowess local regression as described by Yang cDNA microarray analysis. et al. (2002), using the implementation in the S-plus statistical analysis package (Insightful.com). The resulting sets of log ratios (i.e. six replicates per comparison) showed similar cDNA microarray hybridizations median absolute deviation (MAD) and were not normalized The cDNA microarrays used in this study were constructed at further. All flagged spots had values recorded as missing. For the Sanger Centre as part of the LICR/CRUK Microarray each cell line comparison, differentially expressed genes were Consortium, containing 9930 sequence-validated cDNA clones identified by application of the statistical analysis of micro- representing approximately 6000 unique sequences. Informa- array (SAM, version 1.12) Excel add-in, one class analysis. A tion regarding clone set and array preparation can be obtained false discovery rate threshold of 1% in conjunction with a ratio from http://www.sanger.ac.uk/Projects/Microarrays/. All an- threshold of 1.7 was used to define a set of significantly up- and notation in published gene lists is derived from an Ensembl downregulated genes from the three comparisons. The BT474 gene build based on the June 2002 NCBI 30 Golden Path tumor-derived line and the tumor samples 731 and 732 each assembly. had four replicates (two fluor flip replicates). Normalized log In each of the five experiments, RNA preparations from the ratios were derived by Lowess local regression (as above), but C3.6, C5.2 and parental HB4a cell lines were each labelled because of some variation in the MAD of the log ratios separately with both Cy3and Cy5 dyes and hybridized to between the replicates, these were rescaled by use of the robust microarrays in all combinations of cell line and dye (i.e. six estimate of the scale factor used by Yang et al. (15) for print tip hybridizations per experiment, 30 in total). RNA from the normalization. Differentially expressed genes were identified BT474 breast cancer cell line and two grade III invasive ductal by application of SAM as above. breast carcinomas were hybridized in replicated reverse- labelled experiments against RNA from HB4a cells as a RT–PCR reference. Total RNA (25 mg) was used to prepare direct Cy3- and Total RNA (10 mg) was reverse transcribed from an oligo-dT Cy5-labelled first-strand cDNA probes using a single-base primer under conventional conditions (Superscript II, Invitro- anchored oligo dT17 primer (Sigma-Genosys) and Superscript gen). The resulting reaction was diluted 10-fold in water and II reverse transcriptase (Invitrogen). The probes were purified 2 ml was used as a template for PCR amplification. PCR was using AutoSeq G-50 columns (Amersham), and Cy3- and Cy5- performed under standard conditions in 50 ml; 10 mm Tris pH labelled probes were coprecipitated with 8 mg human Cot 1 8.3, 50 mm KCl, 1.5 mm MgCl2, 200 mm dNTPs, 0.2 mm each DNA (Invitrogen) and 8 mg poly dA (Sigma). The pellets were primer (sense and antisense) and 1 U AmpliTaq polymerase resuspended in hybridization buffer (4 Â SSC, 5 Â Denhardt’s (Applied Biosystems) with PCR cycles of 941C for 1 min, then solution, 50 mm Tris-HCl, pH 7.6, 0.1% sarkosyl, 40% 25–40 cycles of 941C for 30 s, 55–601C for 1 min and 721C for formamide) and split across both arrays for hybridization at 1–2 min with a final cycle of 721C for 5 min. Products were 471C overnight. The slides were washed in 2 Â SSC (once), resolved by standard agarose gel electrophoresis and visualized 0.1 Â SSC, 0.1% SDS (four changes) and 0.1 Â SSC (once), by ethidium bromide staining under UV light. Primer and dried in a centrifuge before scanning. Fluorescent images sequences are found in supplementary Table S1. of hybridized microarrays were captured using either the GenePix 4000 (Axon) dual color confocal laser scanner and Tissue array construction and immunohistochemistry GenePix software or a GSI Lumonics 4000 scanner and ScanArray software; data were recorded as paired 16-bit TIFF Breast tumors were selected from the archives of the Royal images. Images were quantitated and background subtracted Marsden Hospital, London, UK with appropriate local using GSI Lumonics Quantarray 3.0 software. All raw Ethical Committee approval. A total of 48 ERBB2-positive fluorescence intensity data and microarray image files are and 47 grade-matched ERBB2-negative (as determined by deposited within the public repository for microarray-based routine immunohistochemistry) invasive ductal carcinomas gene expression data ArrayExpress ( www.ebi.ac.uk/arrayex- were retrieved and reviewed by an experienced pathologist press), complying with minimun information about a micro- (SRL). The paraffin blocks were marked and punched with array experiment (MIAME) and microarray gene expression 0.6 mm2 tumor cores taken from the donor blocks for inclusion database (MGED) group standards (Brazma et al., 2001). in duplicate recipient tissue array blocks using a precision The ArrayExpress accession number for this experiment is tissue array instrument (Beecher Instruments) (Kononen et al., E-SNGR-8. 1998). Antibodies to differentially expressed genes were obtained commercially, where available, for tissue array profiling. All Image processing and data analysis tissue array sections were dewaxed in xylene overnight, taken The background-corrected fluorescence intensity measure- to ethanol (99.7–100% v/v) and blocked for endogenous ments from each experiment were log transformed (base 2) peroxidase in methanol for 10 min. For FN1 and LGALS3,

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2687 each section was subjected to a high-temperature unmasking of Cancer Research. We thank the staff of the Sanger Institute technique (2 min pressure cooking in citrate buffer, pH 6.0). Microarray Facility (http://www.sanger.ac.uk/Projects/Micro- Antigen unmasking was not necessary for HSP27. The sections arrays/) for the supply of arrays, lab protocols, and technical were blocked in normal horse serum (2.5%, Vector Labs) for advice (David Vetrie, Cordelia Langford, Adam Whittaker, 20 min, and primary antibodies applied for 30 min (MYC, Neil Sutton), Quantarray/GeneSpring dataﬁles and all data 1 : 150 (Vector Labs); IGFBP3, 1 : 20 (R&D Systems); FN1, analysis and databases relating to elements on the arrays (Kate 1 : 15 000 (Dako); LGALS3, 1 : 750 (Vector Labs); HSP27, Rice, Rob Andrews, Adam Butler, Harish Chudasama). The 1 : 200 (Vector Labs)). All antibodies were diluted in Tris- human I.M.A.G.E. cDNA clone collection was obtained from buffered saline (TBS). The primary antibodies were rinsed off the MRC HGMP Resource Centre (Hinxton, UK). All cDNA in 0.1% Tween 20 in TBS, developed using Vectastain clone resequencing was performed by Team 56 at the Sanger Universal ABC kit (Vector Labs), and visualized with Institute. We are indebted to Professor Ricardo Brentani at the diaminobenzidine (DAB, Dako). Hospital do Cancer A.C. Camargo, Sao Paulo, Brazil for the tumor RNA. We are grateful to Professors Munro Neville and Alan Ashworth for continued support and helpful discussions pertaining to this project. This work was supported by Acknowledgements Breakthrough Breast Cancer. The microarray consortium is funded by the Wellcome Trust, the Imperial Cancer Research Fund and the Ludwig Institutes

References

Andre S, Kojima S, Yamazaki N, Fink C, Kaltner H, Kayser Hippo Y, Yashiro M, Ishii M, Taniguchi H, Tsutsumi S, K and Gabius, HJ. (1999). J. Cancer Res. Clin. Oncol., 125, Hirakawa K, Kodama T and Aburatani H. (2001). Cancer 461–474. Res., 61, 889–895. Baldini A, Gress T, Patel K, Muresu R, Chiariotti L, Houston SJ, Plunkett TA, Barnes DM, Smith P, Rubens RD Williamson P, Boyd Y, Casciano I, Wells V, Bruni CB, and Miles DW. (1999). Br. J. Cancer, 79, 1220–1226. Mallucci L and Siniscalco M. (1993). Genomics, 15, 216–218. Hughes RC. (2001). Biochimie, 83, 667–676. Baselga J, Tripathy D, Mendelsohn J, Baughman S, Benz CC, Ignatoski KM, Maehama T, Markwart SM, Dixon JE, Dantis L, Sklarin NT, Seidman AD, Hudis CA, Moore J, Livant DL and Ethier SP. (2000). Br. J. Cancer, 82, Rosen PP, Twaddell T, Henderson IC and Norton L. (1996). 666–674. J. Clin. Oncol., 14, 737–744. Kauraniemi P, Ba¨ rlund M, Monni O, Kallioniemi A. (2001). Ben-Porath I, Kozak CA and Benvenisty N. (1998). Genomics, Cancer Res., 61, 8235–8240. 49, 443–447. Kononen J, Bubendorf L, Kallioniemi A, Barlund M, Schraml Ben-Porath I, Yanuka O and Benvenisty N. (1999). Mol. Cell. P, Leighton S, Torhorst J, Mihatsch MJ, Sauter G and Biol., 19, 3529–3539. Kallioniemi OP. (1998). Nat. Med., 4, 844–847. Bieche I, Tomasetto C, Regnier CH, Moog-Lutz C, Rio MC Liehr T, Kuhlenbaumer G, Wulf P, Taylor V, Suter U, Van and Lidereau R. (1996). Cancer Res., 56, 3886–3890. Broeckhoven C, Lupski JR, Claussen U and Rautenstrauss Brazma A, Hingamp P, Quackenbush J, Sherlock G, Spellman B. (1999). Genomics, 58, 106–108. P, Stoeckert C, Aach J, Ansorge W, Ball CA, Causton HC, Lu Y, Zi X, Zhao Y, Mascarenhas D and Pollak M. (2001). J. Gaasterland T, Glenisson P, Holstege FC, Kim IF, Natl. Cancer Inst., 93, 1852–1857. Markowitz V, Matese JC, Parkinson H, Robinson A, Mazzucchelli L. (2002). Am. J. Pathol., 160, 7–13. Sarkans U, Schulze-Kremer S, Stewart J, Taylor R, Vilo J Menard S, Fortis S, Castiglioni F, Agresti R and Balsari A. and Vingron M. (2001). Nat. Genet., 29, 365–371. (2001). Oncology, 61, 67–72. Chen Y, Medvedev A, Ruzanov P, Marvin KW and Jetten Miles DW, Harris WH, Gillett CE, Smith P and Barnes DM. AM. (1997). Genomics, 41, 40–48. (1999). Int. J. Cancer, 84, 354–359. Chomczynski P and Sacchi N. (1987). Anal. Biochem., 162, Moon BK, Lee YJ, Battle P, Jessup JM, Raz A and Kim HR. 156–159. (2001). Am. J. Pathol., 159, 1055–1060. Cobleigh MA, Vogel CL, Tripathy D, Robert NJ, Paz A, Haklai R, Elad-Sfadia G, Ballan E and Kloog Y. Scholl S, Fehrenbacher L, Wolter JM, Paton V, Shak S, (2001). Oncogene, 20, 7486–7493. Lieberman G and Slamon DJ. (1999). J. Clin. Oncol., 17, Pegram MD, Lipton A, Hayes DF, Weber BL, Baselga JM, 2639–2648. Tripathy D, Baly D, Baughman SA, Twaddell T, Glaspy JA Cuny M, Kramar A, Courjal F, Johannsdottir V, Iacopetta B, and Slamon DJ. (1998). J. Clin. Oncol., 16, 2659–2671. Fontaine H, Grenier J, Culine S and Theillet C. (2000). Peralta Soler A, Knudsen KA, Salazar H, Han AC and Cancer Res., 60, 1077–1083. Keshgegian AA. (1999). Cancer, 86, 1263–1272. De Iorio P, Midulla C, Pisani T, Valli C, Grillo L, Di Cocco B, Perillo NL, Marcus ME and Baum LG. (1998). J. Mol. Med., Nofroni I and Vecchione A. (2001). Anticancer Res., 21, 76, 402–412. 1395–1399. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Gamallo C, Moreno-Bueno G, Sarrio D, Calero F, Hardisson Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, D and Palacios J. (2001). Mod. Pathol., 14, 650–654. Fluge O, Pergamenschikov A, Williams C, Zhu SX, Lonning Goldstone SD and Lavin MF. (1991). Biochem. Biophys. Res. PE, Borresen-Dale AL, Brown PO and Botstein D. (2000). Commun., 178, 746–750. Nature, 406, 747–752. Guerreiro Da Silva ID, Hu YF, Russo IH, Ao X, Salicioni Rocha RL, Hilsenbeck SG, Jackson JG, Lee AV, Figueroa JA AM, Yang X and Russo J. (2000). Int. J. Oncol., 16, 231– and Yee D. (1996). J. Natl. Cancer Inst., 88, 601–606. 240. Ross JS and Fletcher JA. (1998). Oncologist, 3, 237–252. Harris RA, Eichholtz TJ, Hiles ID, Page MJ and O’Hare MJ. Ross JS and Fletcher JA. (1999). Semin. Cancer Biol., 9, 125– (1999). Int. J. Cancer, 80, 477–484. 138.

Oncogene ERBB2-related gene expression using microarrays A Mackay et al 2688 Rudland PS, Platt-Higgins A, Renshaw C, West CR, M, Baselga J and Norton L. (2001). N. Engl. J. Med., 344, Winstanley JH, Robertson L and Barraclough R. (2000). 783–792. Cancer Res., 60, 1595–1603. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen Schiemann S, Ruckels M, Engelholm LH, Schwirzke M, H, Hastie T, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen Brunner N and Weidle UH. (1997). Anticancer Res., 17, 13– T, Quist H, Matese JC, Brown PO, Botstein D, Eystein 20. Lonning P and Borresen-Dale AL. (2001). Proc. Natl. Acad. Scorilas A, Trangas T, Yotis J, Pateras C and Talieri M. Sci. USA, 98, 10869–10874. (1999). Br. J. Cancer, 81, 1385–1391. Stoll BA. (1997). Br. J. Cancer, 76, 1652–1654. Shen L and Glazer RI. (1998). Biochem. Pharmacol., 55, 1711– Tomasetto C, Regnier C, Moog-Lutz C, Mattei MG, Chenard 1719. MP, Lidereau R, Basset P and Rio MC. (1995). Genomics, Slamon DJ, Clark GM, Wong SG, Levin WJ, Ullrich A and 28, 367–376. McGuire WL. (1987). Science, 235, 177–182. Werbajh SE, Urtreger AJ, Puricelli LI, de Lustig ES, Bal de Slamon DJ, Godolphin W, Jones LA, Holt JA, Wong SG, Kier Joffe E and Kornblihtt AR. (1998). FEBS Lett., 440, Keith DE, Levin WJ, Stuart SG, Udove J, Ullrich A and 277–281. Press MF. (1989). Science, 244, 707–712. Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J and Slamon DJ, Leyland-Jones B, Shak S, Fuchs H, Paton V, Speed TP. (2002). Nucleic. Acids Res., 30, e15. Bajamonde A, Fleming T, Eiermann W, Wolter J, Pegram Yu D and Hung MC. (2000). BioEssays, 22, 673–680.

Oncogene