Gene profile and response to treatment

2 downloads 53 Views 134KB Size Report
therapy for a large fraction of breast cancer patients, including all patients with primary axillary ... size, histopathological grade and axillary lymph node status.
Annals of Oncology 16 (Supplement 2): ii195 – ii202, 2005 doi:10.1093/annonc/mdi737

Gene profile and response to treatment J. Smeds1, L. D. Miller2, J. Bjo¨hle1, P. Hall3, S. Klaar1, E. T. Liu2, Y. Pawitan3, A. Ploner3 & J. Bergh1 1

Department of Oncology and Pathology, Radiumhemmet, Karolinska Institute and University Hospital, Stockholm, Sweden; 2Genome Institute of Singapore, Singapore; 3Department of Medical Epidemiology and Biostatistics, Karolinska Institute, Stockholm, Sweden

Introduction Breast cancer is the most common malignancy in women, with more than 1 million new individuals diagnosed annually [1]. It is now firmly established that the use of adjuvant chemotherapy and/or endocrine therapies in breast cancer management will result in clinically meaningful reductions in breast cancer recurrence and improved overall survival [2, 3]. Current international guidelines recommend adjuvant systemic therapy for a large fraction of breast cancer patients, including all patients with primary axillary lymph node-positive disease and up to 85 –90% of patients with lymph node-negative disease [4– 6]. Although the effect of adjuvant therapies will be as much as a 50% relative reduction of disease recurrence, the other half of the patients would have been cured without the use of adjuvant therapies or would have relapsed despite their use. These shortcomings of over- and undertreatment are owing to limitations of the presently accepted prognostic and therapy-predictive factors, which lack sufficient sensitivity and specificity. Better prognostic and therapy-predicitive factors (besides oestrogen and progesterone receptors and HER2/neu) are urgently needed.

Presently used prognostic and therapypredictive factors The St Gallen guidelines recommend the use of age, tumour size, histopathological grade and axillary lymph node status for prognostication [5]. These factors are claimed to be true prognostic factors, although they have never been investigated in prospective and randomised studies and accordingly they do not fulfil the requirement for evidence level 1. The only accepted therapy-predictive factor in the St Gallen recommendations is receptor status [5]. At the most recent St Gallen consensus conference, vascular invasion and HER2/neu were accepted as additional risk factors for prognostication. Overexpression/amplification of the oncogene HER2/neu occurs in  20 –30% of primary breast cancers and these patients have a worse prognosis [7, 8]. Overexpression/amplification of HER2/neu has also been associated with a better response to anthracyclines compared with CMF-based therapy, as well as a better response to aromatase inhibitors compared with tamoxifen [9 –11]. Furthermore, overexpression of HER2/neu has been described to be associated with a worse response to tamoxifen [12–14], while other studies have failed q 2005 European Society for Medical Oncology

to demonstrate this relationship [15]. Although these (prognostic and predictive) factors perform quite well using groupbased statistical analyses, the outcome for the individual patient contains a rather high degree of uncertainty.

Limitations and problems for prognostication and therapy-predictive factors Ideally, a prognostic factor should be evaluated in untreated patients. Today, however, the majority of breast cancer patients will receive adjuvant therapy that will often influence outcome, and primary tumour specimens from untreated breast cancer patients (particularly from more recent time periods) are likely to be biased towards those with a relatively low risk for recurrence, or representative of a group with poor outcome owing to other serious medical conditions contra-indicating the use of adjuvant therapies. In addition, almost all breast cancer patients receive therapy for metastatic disease, which also influences survival [16–20]. Thus, the ideal setting for testing prognostic factors is virtually unattainable. A further concern is that many investigators would argue that prognostic and therapy-predictive factors need to be tested in randomised and prospective studies. A randomised study is of course highly relevant for testing a few pre-defined questions, but randomised studies will frequently comprise a highly selected cohort of patients not representative of the general breast cancer population. For example, the inclusion and exclusion criteria for randomised studies are often very strict. Accordingly, patients with co-morbidities or other malignancies who have an expected shorter survival will automatically be excluded. The randomised patient population will thus very likely be comprised of individuals with better prognosis and outcome than those that are truly unselected and representative of the population. We therefore suggest that a new prognostic factor or therapy-predictive factor should be tested on unselected and population-based materials, with a complete coverage over a predefined time period and geographical region.

Microarray-based profiling for biological understanding and prognosis Recently, gene expression profiling of primary breast cancers has been used to improve prognostication for primary breast

ii196 cancer patients, and has had particular success in reducing the requirement for adjuvant therapy [6, 21 –24]. Two Dutch studies investigated 295 and 286 primary breast cancers, respectively [6, 22]. The Amsterdam group identified a 70-gene set that separated both node-negative and nodepositive patients into good and poor prognosis groups better than the St Gallen and NIH consensus risk criteria [21, 22]. The Rotterdam group identified a 76-gene signature for untreated node-negative patients, which performed better in the multivariate analysis compared with classical breast cancer prognostic factors [6]. Sixteen genes were needed for the oestrogen receptor (ERa)-negative group, while 60 genes were needed for the ERa-positive group, further underlining the importance of ERa status in breast cancer biology [6]. Though only three genes overlap between the expression signatures uncovered by these two groups, both signatures contain genes belonging to same molecular pathways, such as proliferation and apoptosis, and are similarly effective in prognosis. Thus, they both have the potential to guide the reduced use of adjuvant chemotherapy for node-negative patients. In Europe, a prospective and randomised study called MINDACT has been planned that seeks to validate the Amsterdam gene signature [25], and the signature has been tested in a preliminary analysis of  300 node-negative frozen samples from six different institutions, which resulted in a somewhat lower prognostic performance [25]. We, also, have recently analysed untreated and treated primary breast cancers using the Affymetrix U133A/B microarray platform for more than 400 patients [26]. The prognostic gene signature that we derived was similar in gene number to the Dutch signatures, but only three genes overlapped with the Amsterdam signature and a single gene with the Rotterdam signature [6, 21]. That several similar studies consistently identify prognostic gene signatures of variable composition might be explained by a number of contributing factors, including: (i) different microarray platforms; (ii) cohort biases owing to dissimilar patient selection procedures and sample size; and (iii) different normalisation approaches and statistical methods for gene selection.

Microarray profiling reveals that breast cancer consists of many diseases Microarray profiling has also been used to discover distinct molecular subgroups of breast cancer that are not discernable by morphological classification [27–29]. For example, the clinical entity of breast cancer can be separated into five or six distinct subgroups with statistically significant prognostic separations both for relapse-free and overall survival [28, 30]. The presumed important breast cancer-related genes p53, HER2/neu and ERa were observed to present in high frequency in certain of these microarray-based subtypes of breast cancer [28]. Five of seven and nine of 11 of the tumours in the HER2/neu and basal cell-like subclasses, respectively, had p53 mutations [28]. In the large luminal subtype A class, only 13% of the tumours contained a p53 mutation (Figure 1),

consistent with this group’s significantly better outcome. The ERa gene on the other hand, was expressed in highest levels in the luminal subtype A. The same research consortium from Norway and USA expanded their initial study of 78 malignant breast tumours to a total of 115 breast cancers and subdivided them into the basal-, HER2/neu overexpressing-, luminal Aand B- and normal-like subtypes [30]. Excluding the normallike subtype, they confirmed previous observations and validated the prognostic implications for the other subtypes using the publicly available gene list from the Amsterdam group consisting of 78 breast cancers [21, 30]. In this study, the consortium also studied 18 and two tumours from patients with hereditary breast cancers with BRCA1 and BRCA2 mutations, respectively. The BRCA1-mutated tumours clustered to the basal-like type, consistent with the poor prognosis associated with BRCA1 mutants [30]. Together, these findings are intriguing in that they suggest that breast tumour subclasses, defined by distinct molecular configurations, may have prognostic primacy over conventional single-gene factors. Gene expression profiling has also been used to discriminate tumour classes associated with known biomarkers of high pathological and therapeutic relevance. In one study, the molecular profiles of spontaneous breast cancers were compared with those with BRCA1 and BRCA2 mutations [31]. Almost 200 genes were found to be significantly differently expressed between tumours with BRCA1 and BRCA2 mutations [31]. Interestingly, one patient with a sporadic breast cancer was demonstrated to have a gene expression profile that was highly similar to that of BRCA1 mutants, suggesting a possible defect elsewhere in the BRCA1 pathway. In a similar vein, we recently identified a gene expression signature capable of discriminating p53 mutant and wild-type tumours with reasonable accuracy. Closer observation revealed that the p53 wild-type tumours with expression signatures more related to that of the mutants possessed significantly lower levels of the p53 transcript, as well as lower transcript levels of a number of p53 direct target genes (compared with the other wildtype tumours), suggesting the possibility of p53 deficiency in these otherwise p53-sequencewild-type tumours. Moreover, Cox’s proportional hazards analysis confirmed a significantly poorer disease-specific survival in this subset of tumours compared with the other p53 wildtypes that was consistent with the survival curve seen in p53 mutants compared with all wild-types [32]. These findings highlight the intriguing view that mechanistically derived gene expression signatures may reflect the operational configuration of clinically relevant pathways in cancer, and thus may provide more robust prognostic information than the biomarkers from which they are derived. The fundamental importance of receptors for breast cancer biology was studied in 58 node-negative breast cancers, where the expression of ERa markedly discriminated tumours into an ERa-positive subgroup and a corresponding negative group with associated genes [33]. The separation of breast cancers into ERa-positive and -negative subgroups using microarray profiling has also been demonstrated by a number of other

ii197 78 carcinomas, 3 fibroadenomas and 4 normal breast samples

correlation coefficient

TP53 status: WT = GREEN TP53 mutation = RED not tested = BLACK

0.1 0.15 0.2 0.4 0.5 0.6

NEW YORK 3 STANFORD 48 NORWAY 21-BE NORWAY 37-AF NORWAY 41-BE STANFORD 14 STANFORD 46-LN STANFORD 23 NEW YORK 2 NORWAY 81-AF NORWAY H5 NORWAY 63-BE NORWAY 109-BE NORWAY 80-BE NORWAY 14-BE NORWAY 65-2ndT STANFORD 2 NORWAY 53-BE NORWAY 101-BE NORWAY 57-BE NORWAY 47-BE NORWAY H6 STANFORD 16 STANFORD 45 STANFORD 44 NORWAY 55-BE Benign STF 37 NORWAY 112-BE NORWAY 65-BE NORWAY 61-BE NorwNormBrst1 NormBrst3 NormBrst2 NormBrst1 Benign STF 20 Benign STF 11 NORWAY H2 NORWAY 104-BE NORWAY 48-BE NORWAY 102-BE NORWAY 26-BE NORWAY 11-BE NORWAY 19-BE NORWAY 15-BE NORWAY 95-BE NORWAY 22-BE NORWAY 7-BE NORWAY 51-BE NORWAY 10-BE NORWAY 39-BE NORWAY 17-BE NORWAY 100-BE NORWAY 12-BE NORWAY 83-BE NORWAY 96-AF NORWAY 90-BE NORWAY 85-BE NORWAY H4-T1 NORWAY 43-AF NORWAY 32-BE NORWAY 4-BE NORWAY 111-BE NORWAY 27-BE NORWAY 18-BE STANFORD 24 NORWAY 74-BE NORWAY 16-BE NORWAY 56-BE NORWAY 64-BE NORWAY 8-BE NORWAY 6-BE NORWAY 98-BE NORWAY 24-BE NORWAY H3 STANFORD A NEW YORK 1 STANFORD 6 STANFORD 40 STANFORD 18 STANFORD 4-LN STANFORD 35 STANFORD 38 STANFORD 31 STANFORD 17 NORWAY 29-BE

0.8

Basal-like

TP53 status: 9/11 (82%)

correlation coefficient

ERBB2+ 5/7 (71%)

Normal Breast-like 2/6 (33%)

Luminal Luminal Subtype C Subtype B 8/10 (80%) 2/5 (40%)

Luminal Subtype A 4/30 (13%)

51 Norway carcinomas, 3 fibroadenomas and 4 normal breast samples

0.1 0.15 0.2 0.4 0.5 0.6

NORWAY 55-BE NORWAY 112-BE NorwNormBrst1 NormBrst3 NormBrst2 NormBrst1 NORWAY 65-BE NORWAY 61-BE Benign STF 11 Benign STF 20 Benign STF 37 NORWAY 104-BE NORWAY 12-BE NORWAY 10-BE NORWAY 39-BE NORWAY 17-BE NORWAY 100-BE NORWAY 14-BE NORWAY 53-BE NORWAY 101-BE NORWAY 57-BE NORWAY 47-BE NORWAY 102-BE NORWAY 26-BE NORWAY 11-BE NORWAY 7-BE NORWAY 95-BE NORWAY 22-BE NORWAY 15-BE NORWAY 19-BE NORWAY 48-BE NORWAY 51-BE NORWAY 41-BE NORWAY 21-BE NORWAY 37-AF NORWAY 81-AF NORWAY 63-BE NORWAY 109-BE NORWAY 80-BE NORWAY 85-BE NORWAY 24-BE NORWAY 43-AF NORWAY 32-BE NORWAY 111-BE NORWAY 4-BE NORWAY 16-BE NORWAY 56-BE NORWAY 8-BE NORWAY 74-BE NORWAY 6-BE NORWAY 27-BE NORWAY 18-BE NORWAY 64-BE NORWAY 96-AF NORWAY 98-BE NORWAY 29-BE NORWAY 90-BE NORWAY 83-BE

0.8

Normal Breast-like

Luminal ERBB2+ Subtype B

Luminal Subtype C

Basal-like

Luminal Subtype A

Figure 1. The upper dendrogram shows status of TP53 in the five (six) subgroups identified in all 85 samples analysed by Sorlie et al. [28], indicated by the colour of the terminal dendrogram line. Red lines indicate tumours with mutated TP53 genes, green lines wild-type (WT) TP53 and black lines samples not tested. The lower dendrogram shows hierarchical clustering analysis using the 51 Norway carcinomas, three benign tumours and four normal breast tissues. It closely resembles the upper dendrogram. The subgroups are coloured accordingly and show that the group of tumours highlighted in orange changed position compared with the upper dendrogram. Furthermore, the basal-like tumours shown in red are inserted in between luminal subtypes A and C. To the left are shown the correlation coefficients for the dendrogram branches. Reproduced with permission.

groups [21, 27, 29, 34]. ERa gene expression has recently been analysed together with ERb, aiming at understanding their functional interactions in human breast cancer [35]. It was revealed that ERb had an inhibitory effect on tumour cell proliferation and that the gene expression signature linked to low ERb expression was associated with better disease outcome.

Microarray profiling for therapy prediction Gene expression profiling has also been used to identify gene patterns that potentially identify genes related to response to

certain cytotoxic agents like docetaxel [36]. This discriminatory array profile was based on a 92-gene set. In this limited series, one patient had progressive disease. The authors defined a positive tumour response as 25% or less residual disease, and resistant disease was defined as 26% or more residual disease. Conventional UICC and RECIST criteria define a 50% and 30% reduction, respectively, for a response [37]. If the UICC criteria had been applied, some of the tumours in the gene expression-predicted resistant group would have belonged to the UICC-responding group, despite the fact that they had a completely different expression profile for the selected gene signature.

ii198 This study utilised core biopsy samples taken before the initiation of the neo-adjuvant docetaxel therapy. Although the authors could extract only 3–6 mg total RNA from each biopsy, they acquired enough material for microarray analysis on the Affymetrix HgU95-Av2 gene chip. The important wider implication is that core biopsies, which represent a non-invasive tumour sampling method, give enough high-quality RNA for reliable expression profiling. Others have also reported the usefulness of core biopsies in microarray analysis [38, 39]. In the ongoing and randomised European Organisation for Research and Treatment of Cancer p53 study, we can confirm that for most patients this core biopsy strategy using 14 gauge/18 gauge needles will result in sufficient RNA quantities [40]. Fine needle aspirates, as used at the Karolinska Institute, Stockholm, Sweden, can also be used to obtain sufficient RNA quantities (2 mg/sample). The manual skills to obtain sufficient tumour material seem to be even more important with the fine needle biopsy technique, which very likely gives a purer tumour cell collection compared with the core biopsy technique. This will very likely have implications on the gene signatures, while the cytological material may contain a higher proportion of tumour cells versus stroma cells relative to material obtained with the core biopsy technique. The broader view of the Chang et al. study [36] is the potential of optimised and better tailored drug selections based on a gene signature. Repeated biopsies may be used for early therapy evaluations in studies focused on genes regulating proliferation and apoptosis. The combined use of gene expression signatures for prognosis, therapy selection and continuous follow-up in vivo may help to shift our present broad adjuvant strategies to better tailored and optimised neo-adjuvant therapy concepts. Regarding gene expression patterns related to endocrine treatment of breast cancer patients, a recent study showed that the expression of two genes, HOXB13 and IL17BR, extracted from a tamoxifen outcome predictive expression signature, could predict clinical outcome in a group of tamoxifen-treated breast cancer patients with better performance than existing biomarkers and with the same accuracy as the signature [41]. Such an expression ratio of only two genes would simplify application to clinical use relative to expression signatures based on large numbers of genes.

empirically determined expression signatures may simply recapitulate the prognostic aspects of certain conventional factors, and that gene selection methods are needed to derive signatures with strong independent prognostic capabilities.

Critical review of gene expression profiling

RNA from breast cancer samples have been studied on different microarray platforms, utilising either long cDNA clones as probes [28], Affymetrix 25mer oligo probes [6] or 60mer oligo probes [21]. The diversity of microarray platforms utilised for gene expression analysis and the variability in microarray data emphasise the need for quality assurance. High quality of the RNA samples is essential for microarray analysis, and each RNA preparation must be rigorously assessed for quality. Careful measures need to be undertaken during all steps of the RNA extraction method to prevent the RNA from degrading and to maximise the yield. DNase treatment should be

A Swedish group analysed clinical, histopathological and cell biological parameters in the previously discussed Amsterdam cohort (i.e. 97 patients) in order to determine whether these conventional variables were able, or not, to provide the same level of prognostic information as the microarray expression signature described by the authors [21, 42]. The Swedish group demonstrated a very similar outcome for both the good prognosis group and poor prognosis group using artificial neural networks, as well as the Nottingham prognostic index, compared with gene expression profiling. This suggests that

How do microarrays function? The Affymetrix and Agilent systems are among the dominating microarray platforms on the commercial market (Affymetrix Inc., Santa Clara, CA, USA; Rosetta Inpharmatics, licensed to Agilent Technologies, Palo Alto, CA, USA). Affymetrix chips are synthesised by photolithography and solidphase chemistry. Each perfect-match oligonucleotide is coupled with a mismatched oligonucleotide to measure nonspecific hybridisation. Several (11–20) pairs of 25mer oligoprobes are used for each gene or transcript. These bind to different regions of the cRNA, and improve signal-to-noise ratio and the dynamic range of detection. The multiple oligonucleotides also reduce cross-hybridisation effects. Biotinlabelled antisense cRNA amplified from sample mRNA is hybridised to the chip. For the Agilent technology, antisense cRNA or cDNA from sample and reference are labelled with the fluorescent dyes Cy3 and Cy5 and hybridised simultaneously to the array (Figure 2). Disadvantages with the labelling method may be the two-colour dye bias and the sensitivity of Cy5 to decay with exposure to atmospheric ozone, although Agilent recently modified the wash protocol to reduce the problem with ambient ozone levels. The Agilent arrays are synthesised by inkjet technology. The array contains between 11 000 and 44 000 spots with millions of copies of a 60mer oligonucleotide or cDNA. Each gene or transcript is represented by only one 60mer. Agilent allows for flexible design, with possibility to choose between customer-designed probes between 25 –60 bases long and 60mer probes designed by Agilent. The 60mer oligo format is advantageous relative to short 25mer length oligos in terms of sensitivity; however, shorter oligos are considered to be more specific. Agilent arrays are more tolerant of mismatches, which results in simplified analysis of polymorphic regions, but on the other hand decreases specificity.

Technical aspects of gene expression profiling: what can go wrong

ii199 RNA from Sample A

RNA from Sample B mRNA

AAAA 3·





MMLV-RT Oligo dT-Promoter Primer

MMLV-RT Oligo dT-Promoter Primer 1 st strand 5· cDNA

Promoter



TTTT



AAAA Anti-sense Promoter

3· 2 nd strand cDNA MMLV-RT

TTTT

3· 5·

Promoter

1 st strand 5· cDNA

3· 2nd strand cDNA MMLV-RT

AAAA Anti-sense Promoter

T7 RNA polymerase Cy5-CTP, NTPs

T7 RNA Polymerase Cy3-CTP, NTPs T7 RNA polymerase

T7 RNA polymerase Cy5

Cy3

Cy5

C

C

5· cRNA

UUUU



AAAA

Anti-sense Promoter

3· 2nd strand cRNA

Cy3

C

C



Cy5

C

C

Anti-sense Promoter

3· 2nd strand cRNA

Purify cRNA Cy3

Cy5 UUUU

5· cRNA

UUUU AAAA

Purify cRNA



mRNA

AAAA 3·

5· cRNA (antisense)



Cy3

C

C

UUUU

5· cRNA (antisense)

B

A

A+B

Oligo microarrays

Figure 2. The principle of the Agilent microarray analysis. Reference and tumour RNAs are labelled with Cy3 and Cy5 fluorescent dyes, respectively, then pooled together and hybridised to the array. The relative abundance of mRNA from tumour versus reference is obtained from the Cy5/Cy3 intensity ratios. MMLV-RT, Moloney murine leukemia virus reverse transcriptase.

included in the protocol to ensure the absence of contaminating DNA. The addition of proteinase K during RNA extraction markedly increases the yield [43]. Usually, RNA integrity is controlled for by measurement of the 28S/18S ribosomal RNA ratio on a capillary electrophoresis system such as the Agilent 2100 Bioanalyzer (Figure 3). A number of factors influence the correlation between different microarray platforms and complicate cross-platform comparisons. Different factors may affect the variability in microarray gene expression results and analyses: technical factors such as microarray manufacturing, cDNA or oligo probe type, selection of genes represented on the arrays, sample collection, RNA extraction, synthesis of cDNA or cRNA, labelling with fluorescent dye, and hybridisation; instrumental parameters such as image acquisition and quantitation; computational factors including preprocessing, normalisation and analysis of data; and biological factors [44–46].

cDNA clones and oligonucleotides differ in their hybridisation properties. GC content, probe sequence length, degree of cross-hybridisation and average signal intensity influence the correlation between microarray platforms. The length of the cDNA clones makes them more prone to hybridise to homologous sequences on other genes, while short oligonucleotides are more specific. Alternatively spliced transcripts that differ in their hybridisation ability to short oligo- or long cDNA-based arrays may contribute to the differences between the platforms [46]. The Affymetrix platform is perhaps the most standardised and evaluated microarray platform, with high reproducibility; however, some studies have reported a lower reproducibility (using the same RNA preparations) [47–51]. Yauk et al. [47] reported good reproducibility for Agilent cDNA and 60mer oligonucleotide arrays. Poor correlation between Affymetrix chips and customised cDNA microarrays have been reported in several studies, while use of only sequence-verified and -matched genes for comparison of Affymetrix and Agilent

ii200 (a)

45 40

Fluorescence

35 30 25 20 15

18S

5 0

(b)

19

24

29

34

39

19

24

29

34

39

28S

10

44 49 Time (seconds)

54

59

64

69

54

59

64

69

10 9

Fluorescence

8 7 6 5 4 3

18S

1

28S

2 0

44

49

Time (seconds) Figure 3. Assessment of the integrity of RNAs isolated from breast tumours on an Agilent 2100 Bioanalyzer system. The ratio of the 28S/18S ribosomal RNA peaks is measured in an electropherogram. The upper graph illustrates the 18S and 28S rRNA peaks of intact RNA and the lower graph shows RNA with marked degradation.

cDNA microarrays resulted in a better correlation [52–56]. Jarvinen et al. [46] observed a better correlation between Affymetrix and Agilent cDNA than with custom-made cDNA arrays. Better concordance has also been reported for Affymetrix and Agilent oligo or customised oligo microarrays [57, 58].

the King Gustav V Jubilee Fund, Stockholm, Singapore Agency for Science and Technology and Research (A* STAR) and Bristol-Myers Squibb.

References Conclusions RNA expression profiling of human breast cancer can separate patients into subgroups with good prognosis signatures versus poor prognosis signatures. Different prognostic gene expression signatures with very little overlap have been reported in breast cancer, underlining the importance of improved analytical procedures and the use of large numbers of patients for validation. Gene signatures will likely be used in the future for optimised and tailored drug selections.

Acknowledgements Our array studies have been supported by grants from the Swedish Cancer Society, the Cancer Society in Stockholm and

1. Schwartsmann G, Ratain MJ, Cragg GM et al. Anticancer drug discovery and development throughout the world. J Clin Oncol 2002; 20: 47S–59S. 2. Early Breast Cancer Trialists’ Collaborative Group. Polychemotherapy for early breast cancer: an overview of the randomised trials. Lancet 1998; 352: 930–942. 3. Early Breast Cancer Trialists’ Collaborative Group. Tamoxifen for early breast cancer: an overview of the randomised trials. Lancet 1998; 351: 1451–1467. 4. Eifel P, Axelson JA, Costa J et al. National Institutes of Health Consensus Development Conference Statement: adjuvant therapy for breast cancerNovember 1–3, 2000. J Natl Cancer Inst 2001; 93: 979–989. 5. Goldhirsch A, Wood WC, Gelber RD et al. Meeting highlights: updated international expert consensus on the primary therapy of early breast cancer. J Clin Oncol 2003; 21: 3357–3365.

ii201 6. Wang Y, Klijn JG, Zhang Y et al. Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet 2005; 365: 671 –679. 7. Slamon DJ, Clark GM, Wong SG et al. Human breast cancer: correlation of relapse and survival with amplification of the HER-2/neu oncogene. Science 1987; 235: 177 –182. 8. Sjogren S, Inganas M, Lindgren A et al. Prognostic and predictive value of c-erbB-2 overexpression in primary breast cancer, alone and in combination with other prognostic markers. J Clin Oncol 1998; 16: 462 –469. 9. Muss HB, Thor AD, Berry DA et al. c-erbB-2 expression and response to adjuvant therapy in women with node-positive early breast cancer. N Engl J Med 1994; 330: 1260–1266. 10. Paik S, Bryant J, Park C et al. erbB-2 and response to doxorubicin in patients with axillary lymph node-positive, hormone receptor-negative breast cancer. J Natl Cancer Inst 1998; 90: 1361–1370. 11. Smith I, Dowsett M, on behalf of the IMPACT, Royal Marsden Hospital, United Kingdom. In Comparison of anastrozole vs tamoxifen alone and in combination as neoadjuvant treatment of estrogen receptor-positive (ER+) operable breast cancer in postmenopausal women: the IMPACT trial. In Lippman ME (ed.). 82. 26th Annual San Antonio Breast Cancer Symposium 2003; S6, (Abstr 1). 12. Borg A, Baldetorp B, Ferno M et al. ERBB2 amplification is associated with tamoxifen resistance in steroid-receptor positive breast cancer. Cancer Lett 1994; 81: 137 –144. 13. Carlomagno C, Perrone F, Gallo C et al. c-erbB2 overexpression decreases the benefit of adjuvant tamoxifen in early-stage breast cancer without axillary lymph node metastases. J Clin Oncol 1996; 14: 2702–2708. 14. Houston SJ, Plunkett TA, Barnes DM et al. Overexpression of c-erbB2 is an independent marker of resistance to endocrine therapy in advanced breast cancer. Br J Cancer 1999; 79: 1220– 1226. 15. Elledge RM, Green S, Ciocca D et al. HER-2 expression and response to tamoxifen in estrogen receptor-positive breast cancer: a Southwest Oncology Group Study. Clin Cancer Res 1998; 4: 7–12. 16. Fossati R, Confalonieri C, Torri V et al. Cytotoxic and hormonal treatment for metastatic breast cancer: a systematic review of published randomized trials involving 31,510 women. J Clin Oncol 1998; 16: 3439–3460. 17. A’Hern RP, Ebbs SR, Baum MB. Does chemotherapy improve survival in advanced breast cancer? A statistical overview. Br J Cancer 1988; 57: 615– 618. 18. Nabholtz JM, Senn HJ, Bezwoda WR et al. Prospective randomized trial of docetaxel versus mitomycin plus vinblastine in patients with metastatic breast cancer progressing despite previous anthracyclinecontaining chemotherapy. 304 Study Group. J Clin Oncol 1999; 17: 1413–1424. 19. Slamon DJ, Leyland-Jones B, Shak S et al. Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. N Engl J Med 2001; 344: 783 –792. 20. O’Shaughnessy J, Miles D, Vukelja S et al. Superior survival with capecitabine plus docetaxel combination therapy in anthracyclinepretreated patients with advanced breast cancer: phase III trial results. J Clin Oncol 2002; 20: 2812–2823. 21. van’t Veer LJ, Dai H, van de Vijver MJ et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002; 415: 530–536. 22. van de Vijver MJ, He YD, van’t Veer LJ et al. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 2002; 347: 1999–2009. 23. Ahr A, Karn T, Solbach C et al. Identification of high risk breastcancer patients by gene expression profiling. Lancet 2002; 359: 131 –132.

24. Huang E, Cheng SH, Dressman H et al. Gene expression predictors of breast cancer outcomes. Lancet 2003; 361: 1590–1596. 25. Piccart MJ, Loi S, van’t Veer L, et al. Multi-center external validation study of the Amsterdam 70-gene prognostic signature in node negative untreated breast cancer: are the results still outperforming the clinical-pathological criteria? Breast Cancer Res Treat 2004; 88 (Suppl 1): (Abstr 38). 26. Pawitan Y, Bjohle J, Amler L et al. Gene expression profiling spares early breast cancer patients from adjuvant therapy–derived and validated in two population based cohorts. Manuscript 2005. 27. Perou CM, Sorlie T, Eisen MB et al. Molecular portraits of human breast tumours. Nature 2000; 406: 747–752. 28. Sorlie T, Perou CM, Tibshirani R et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA 2001; 98: 10869– 10874. 29. Sotiriou C, Neo SY, McShane LM et al. Breast cancer classification and prognosis based on gene expression profiles from a populationbased study. Proc Natl Acad Sci USA 2003; 100: 10393–10398. 30. Sorlie T, Tibshirani R, Parker J et al. Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci USA 2003; 100: 8418–8423. 31. Hedenfalk I, Duggan D, Chen Y et al. Gene-expression profiles in hereditary breast cancer. N Engl J Med 2001; 344: 539 –548. 32. Miller LD, Smeds J, George J et al. An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects and patient survival. 2005; Manuscript. 33. Gruvberger S, Ringner M, Chen Y et al. Estrogen receptor status in breast cancer is associated with remarkably distinct gene expression patterns. Cancer Res 2001; 61: 5979–5984. 34. West M, Blanchette C, Dressman H et al. Predicting the clinical status of human breast cancer by using gene expression profiles. Proc Natl Acad Sci USA 2001; 98: 11462–11467. 35. Lin C-Y, Strom A, Kietz S et al. Microarray analysis reveals inhibitory mechanisms of estrogen receptor beta on tumor cell proliferation and identifies markers predictive of clinical outcome in primary breast cancer. 2005; Manuscript. 36. Chang JC, Wooten EC, Tsimelzon A et al. Gene expression profiling for the prediction of therapeutic response to docetaxel in patients with breast cancer. Lancet 2003; 362: 362– 369. 37. Therasse P, Arbuck SG, Eisenhauer EA et al. New guidelines to evaluate the response to treatment in solid tumors. European Organization for Research and Treatment of Cancer, National Cancer Institute of the United States, National Cancer Institute of Canada. J Natl Cancer Inst 2000; 92: 205–216. 38. Ellis M, Davis N, Coop A et al. Development and validation of a method for using breast core needle biopsies for gene expression microarray analyses. Clin Cancer Res 2002; 8: 1155–1166. 39. Sotiriou C, Khanna C, Jazaeri AA et al. Core biopsies can be used to distinguish differences in expression profiling by cDNA microarrays. J Mol Diagn 2002; 4: 30–36. 40. Farmer P, Iggo R, Becette V et al High quality gene expression microarray data from a multicentre prospective trial: results of the first microarray analysis in the EORTC 10994/BIG 00-01 study. 4th European Breast Cancer Conference. Eur J Cancer 2 2004; S99 (Abstr 155). 41. Ma XJ, Wang Z, Ryan PD et al. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell 2004; 5: 607–616. 42. Eden P, Ritz C, Rose C et al. ‘Good old’ clinical markers have similar power in breast cancer prognosis as microarray gene expression profilers. Eur J Cancer 2004; 40: 1837– 1841.

ii202 43. Egyhazi S, Bjohle J, Skoog L et al. Proteinase K added to the extraction procedure markedly increases RNA yield from primary breast tumors for use in microarray studies. Clin Chem 2004; 50: 975 –976. 44. Shi L, Tong W, Goodsaid F et al. QA/QC: challenges and pitfalls facing the microarray community and regulatory agencies. Expert Rev Mol Diagn 2004; 4: 761– 777. 45. Ein-Dor L, Kela I, Getz G et al. Outcome signature genes in breast cancer: is there a unique set? Bioinformatics 2005; 21: 171 –178. 46. Jarvinen AK, Hautaniemi S, Edgren H et al. Are data from different gene expression microarray platforms comparable? Genomics 2004; 83: 1164–1168. 47. Yauk CL, Berndt ML, Williams A, Douglas GR. Comprehensive comparison of six microarray technologies. Nucleic Acids Res 2004; 32: e124. 48. Tan PK, Downey TJ, Spitznagel EL Jr et al. Evaluation of gene expression measurements from commercial microarray platforms. Nucleic Acids Res 2003; 31: 5676– 5684. 49. Piper MD, Daran-Lapujade P, Bro C et al. Reproducibility of oligonucleotide microarray transcriptome analyses. An interlaboratory comparison using chemostat cultures of Saccharomyces cerevisiae. J Biol Chem 2002; 277: 37001– 37008. 50. Unger MA, Rishi M, Clemmer VB et al. Characterization of adjacent breast tumors using oligonucleotide microarrays. Breast Cancer Res 2001; 3: 336–341.

51. Bakay M, Chen YW, Borup R et al. Sources of variability and effect of experimental approach on expression profiling data interpretation. BMC Bioinformatics 2002; 3: 4. 52. Kothapalli R, Yoder SJ, Mane S, Loughran TP Jr. Microarray results: how accurate are they? BMC Bioinformatics 2002; 3: 22. 53. Li J, Pankratz M, Johnson JA. Differential gene expression patterns revealed by oligonucleotide versus long cDNA arrays. Toxicol Sci 2002; 69: 383–390. 54. Kuo WP, Jenssen TK, Butte AJ et al. Analysis of matched mRNA measurements from two different microarray technologies. Bioinformatics 2002; 18: 405–412. 55. Lee JK, Bussey KJ, Gwadry FG et al. Comparing cDNA and oligonucleotide array data: concordance of gene expression across platforms for the NCI-60 cancer cells. Genome Biol 2003; 4: R82. 56. Mecham BH, Klus GT, Strovel J et al. Sequence-matched probes produce increased cross-platform consistency and more reproducible biological results in microarray-based gene expression measurements. Nucleic Acids Res 2004; 32: e74. 57. Barczak A, Rodriguez MW, Hanspers K et al. Spotted long oligonucleotide arrays for human gene expression analysis. Genome Res 2003; 13: 1775–1785. 58. Chiorino G, Acquadro F, Mello Grand M et al. Interpretation of expression-profiling results obtained from different platforms and tissue sources: examples using prostate cancer data. Eur J Cancer 2004; 40: 2592–2603.