We studied a consecutive series of 832 tumours with early (stage I, II or III) breast cancer treated in our institution between October 1987 and December 2001 and with sufficient cancer tissue available for inclusion in TMA. The stage of disease was defined according to the tumour node metastasis (TNM) classification. Tumours were all invasive adenocarcinomas. The patients were treated according to guidelines used in our institution: all had primary surgery that included complete resection of the tumour (modified radical mastectomy in 28% of cases, lumpectomy in 72%) and axillary lymph node dissection; 96% were treated with breast-conservative surgery received adjuvant local-regional radiotherapy; 51.3% were given adjuvant chemotherapy (anthracyclin-based regimen in most cases); 56.5% received adjuvant hormone treatment (tamoxifen in most cases) and 54.9% of these received adjuvant chemotherapy. After completion of treatment, the patients were evaluated at least twice a year for the first five years and at least annually thereafter. The median follow-up was 86 months after diagnosis; 162 patients experienced metastatic relapse as a first event (local recurrence was not taken into account as first event). The five-year metastasis-free survival (MFS) rate was 83.2% (95% confidence interval (CI) = 80.4 to 85.8). The experimental part of this study concerning paraffin-embedded samples was completed before informed consent was necessary but was approved and executed in compliance with our institutional review board. Each sample was assigned an anonymous unique identification that was linked to an anonymous clinical board approved data base containing follow-up information. The study was performed with the intent of benefiting treatment planning in future patients.

Tissues were collected from 143 patients with invasive adenocarcinoma who underwent initial surgery at the Institut Paoli-Calmettes (Marseilles, France). Each patient gave written informed consent. Samples were macro-dissected and frozen in liquid nitrogen within 30 minutes of removal.

Nucleic acids were extracted from frozen samples by using guanidium isothiocyanate and cesium chloride gradient, as previously described 16. RNA integrity was controlled on the Agilent Bioanalyzer (Agilent Technologies, Massy, France).

Gene expression was analysed in 143 breast cancer samples and four normal breast samples with Affymetrix U133 Plus 2.0 human oligonucleotide microarrays (Affymetrix Santa Clara, CA, USA). Preparation of c-RNA, hybridisations, washes and detection were performed as recommended by the supplier. For each sample, synthesis of the first-strand c-DNA was done from 3 μg total RNA by T7-oligo(dT) priming, followed by second-strand cDNA synthesis. After purification, in vitro transcription associated with amplification generated cRNA-containing biotinylated pseudouridine. Biotinylated cRNA was purified, quantified and chemically fragmented (95°C for 35 minutes), then hybridised to microarrays in 200 μL hybridisation buffer at 45°C for 16 hours. Automated washes and staining with streptavidin-phycoerythrin were performed as recommended. Double signal amplification was achieved by biotinylated antistreptavidin antibody with goat-IgG blocking antibody. Scanning was performed with Affymetrix GeneArray scanner and quantification with Affymetrix GCOS software.

Affymetrix data were analysed by the Robust Multichip Average method in R using Bioconductor and associated packages 17. The Robust Multichip Average performed background adjustment, quantile normalisation and summarisation of 11 oligonucleotides per gene. Before analysis, a filtering process removed the genes with low and poorly measured expression, as defined by an expression value inferior to 100 units in all breast cancer tissue and normal tissue samples, from the dataset. All data was then log₂-transformed for display and analysis.

Basal and luminal breast cancers were distinguished by the differential expression of clusters of genes. Sub-classification of the luminal cases was done as previously described 5. Kinase gene expression identified two subgroups of luminal A breast cancers, that is luminal Aa and Ab.

TMAs were prepared as previously described 18 from formalin-fixed and paraffin-embedded tissue. For each tumour, three representative areas were selected from a H&E-safran-stained section of a donor block. Core cylinders with a diameter of 0.6 mm each were punched from each of these areas and deposited into three separate recipient paraffin blocks using a specific arraying device (Alphelys, Plaisir, France).

Immunohistochemistry of 5 μmm TMA sections was performed as previously described using Dako LSAB^R2 Kit in the autoimmunostainer (Dako Autostainer, Glostrup, Denmark). Sections were deparaffinised in Histolemon (Carlo Erba Reagenti, Rodano, Italy) and rehydrated in graded ethanol solutions. Results were evaluated under a light microscope by two pathologists (EC-J, JJ) and scored by the quick score (QS) 19. The QS was used to combine the impact of the percentage and the intensity of the immunostaining. QS multiplies the percentage by the intensity and represents a range of 0 to 300. For each antibody, a sample was considered as positive when the QS was strictly superior to 0. However, the Ki67 status was expressed in terms of percentage of positive cells, with a threshold of 20% of positive cells. The ERBB2 status was evaluated with the Dako scale (HercepTest kit scoring guidelines, DakoCytomation, Copenhagen, Denmark). The level of 3+ was considered as positive and all 2+ cases were evaluated by chromogen in situ hybridisation (only the case with a ratio higher than 2.2 were considered as positive).

For each tumour, the mean of the score of a minimum of two core biopsies was calculated. The list of antibodies used is given in Table 1.

Survival rates were estimated by using the Kaplan-Meier method 20. The endpoint was the MFS, which was defined as the time from the date of breast cancer diagnosis until the date of the first distant relapse. Patients without relapse were censored at the time of last follow-up. Survival analysis was computed with a stratification on treatment by chemotherapy. Relative risks of metastasis according to the baseline factors were estimated by using the Cox proportional-hazards regression models 21 in univariate and multivariate analyses. In univariate analysis, differences in MFS were analysed by the Log-Rank test. Factors with a P value less than 0.15 in univariate analysis were included in the multivariate analysis, with a backward selection of variable procedure to minimise the Akaike information criterion 22. Results are presented as mean (95% CI). Statistical analyses were performed with the R.2.7.1. Statistical language 23.

A total of 135 of the 143 cases showed good-quality RNA and profiles. The correlation between expression of each parameter between microarrays and QS was excellent and highly significant (Table 2). The lowest level of the Rho coefficient was observed for Ki67, P53 and ERBB2.

The frequency of the different subtypes was: 25% basal, 12% ERBB2, 46% luminal A (68.1%% Aa and 31.8%% Ab), 2% luminal B and 9% normal-like. Molecular subtype and ER status were strongly correlated. Only 8% of basal breast cancers were ER-positive for 23.5% of ERBB2, 95.5% of luminal A, 100% of luminal B and 53.8% of normal-like breast cancers (Table 3).

We studied the impact of 16 histoclinical and immunohistochemical factors on disease-free survival. Hormonal therapy, size of the lesion, histoprognostic grade, vascular peritumoural invasion (VPI), ER, BCL2, GATA3, Ki67 and P53 had significant impact (Table 4). Only age, CCND1, PR, FOXA1 did not have any significant value in MFS.

For ERBB2 there was a significant difference in terms of disease-free survival at 60 months with 83.7% for the negative cases and 69.1% for the 3+ cases and amplified 2+ (P = 0.017). However, when the analysis was stratified on the presence or not of chemotherapy, no significant difference between the two groups was noted.

Eight factors were retained by the multivariate analysis including histopronostic grade, axillary lymph node invasion, VPI, size, then Ki67, P53, BCL2 and hormonal therapy (Table 5).

In the restricted ER-positive population studied by gene expression profiling we observed that 67 of 81 (86.4%) were luminal cases (Table 6). There was no difference in ER-positivity level (with a cut off QS of 120) between luminal Aa (18 of 43 above 120, 41.8%) and luminal Ab (8 of 21 above 120, 38%) cases. However, a significant difference was observed for proliferation: luminal Ab showed a higher grade (P = 4.710; Table 6) and a higher Ki67 index (P = 0.02) than luminal Aa. The three luminal B were ER-positive (the percentage of ER-positive cells was below 5% for two cases) but grade 3. Two were positive for P53.

One-third of the luminal A were Ab (31.8%) and two-thirds (68.1%) were luminal Aa cases. The four ERBB2-subtype cases were ER-positive but the level of ER expression was lower than the median value of QS 120; all the cases were grade 3 and PR-negative. The seven normal-like cases were grade 1 (n = 5) or 2 (n = 2) and four showed a low level of ER protein expression.

We then restricted the study to the ER-positive cases treated by hormonal therapy (n = 384). Subtype status was available for only a small series of these cases (n = 43). MFS was different between luminal Aa and Ab cases (P = 0.042; Figure 1). Of the 14 factors studied in univariate analysis (Table 7) only 6 showed a different distribution: size, grade, VPI, lymph node invasion, GATA3 and Ki67. Oestrogen-related proteins such as FOXA1 and AR had no significant impact whatever their quantitative value. The ER and PR level of expression had no significant MFS value in univariate analysis. The multivariate analysis in terms of MFS retained four factors: VPI, Ki67, P53 and GATA3 (Table 8).

Grossly, our 135 subtyped cases showed a similar distribution of subtypes as found in previous studies 1224. Our series contained 25% of basal cases, which is within a published range of 17 to 37%. The number of ERBB2 subtype (12%) was slightly higher than in most series. In contrast, the 45% frequency of luminal A was high and the proportion of luminal B was low.

In a previous study 5 we focused on the kinome of luminal A breast cancers. The breast cancer kinome differs between basal and luminal A cases. Within luminal A cases, it allowed the identification of luminal Aa and Ab. Here, we have confirmed the difference in outcome between luminal Aa and Ab by using immunohistochemistry on 43 luminal A cases treated by tamoxifen. The difference between luminal Aa and Ab was due to proliferative factors translated by a higher grade and Ki67 index in luminal Ab than in luminal Aa cases. The fact that a difference could be seen already with a small series suggests the importance of proliferation to distinguish outcome in ER-positive cases whatever their percentage of ER-positive cells.

Four factors, VPI, GATA3, P53 and Ki67, were retained by the multivariate analysis.

Two parameters were added in the 9th St Gallen meeting compared with the 8th edition: ERBB2 status and VPI. The volume of data published in the past few years provides compelling evidence for the importance of VPI 25 but the specific impact on luminal cases had never been described. A meta-analysis of microarray data revealed the importance of GATA3 26. Its expression in 10-year follow-up 27 demonstrates that its protective effect is more pronounced in patients who received tamoxifen. We showed the prognostic impact of P53 in two previous studies of luminal cases 45. Ki67 higher than 20% is one of the parameters able to distinguish luminal A from luminal B 28 but its specific prognostic impact in luminal cases had not been described.

An important question is whether the combination of VPI, GATA3, P53 and Ki67 predicts pure prognosis or responsiveness to endocrine therapy or both. Few studies using profiling of ER-positive breast cancers treated by tamoxifen have established a signature able to predict the prognosis. The oncotype DX RS 29 is a commercially available assay (Genomic Health, Redwood City, CA) that predicts recurrence in ER-positive cases. It is a PCR-based assay on paraffin-embedded tissue using 16 cancer-related genes and 5 controls. It was validated on 668 node-negative cases in the National Surgical Adjuvant Breast and Bowel Project (NSABP) trial receiving tamoxifen only. The histopronostic grade and recurrence score (RS) were significant. This RS was subsequently validated for chemotherapy and tamoxifen in 645 patients from the NSABP-14 30. Only four of the 16 genes are common with the factors we tested here (ER, PR, KI67 and BCL2). A more recent series of 255 ER-positive cases established a signature validated on an independent set of 362 cases coming from different institutions and treated by tamoxifen alone 31. A total of 181 genes belonging to 13 clusters strongly prognostic (HR = 3.26, P = 0.0002). These 13 cluster genes were the most important factor in multivariate analysis.

Immunohistochemistry has been involved in the search for a multiparametric score in ER-positive cases on a series of 257 ER-positive cases treated by tamoxifen; a multimarker model was established from nine markers and five of them were retained in a mathematic model: ER, PR, P53, ERBB2 and MYC. This model was more prognostic than the Nottingham prognostic index 4.

A previous study has looked at oestrogen-regulated genes in the MCF7 breast cancer cell line treated by 17β–oestradiol 32. These genes were then used to develop an outcome predictor on a training set of 65 luminal breast cancers and then validated on three independent published data sets. Interestingly, two groups of low risk (expressing XBP1, FOXA1 and PR) and high risk (expressing MYBL2 and CCNB2) were distinguished.

The study of a series of 140 cases used 23 antibodies and identified a prognostic score for ER-positive breast cancer without any notion of hormonal therapy 33. Five factors were retained by Cox analysis (P53, NDRG1, CEACAM5, SLC7A5 and HTF9c) but regression tree analysis retained six factors (P53, PR, Ki67, NAT1, SLC7A5 and HTF9c). The best HR was obtain by the Cox model (HR = 2.21, P = 0.0008).

P53 and Ki67 are the two factors common with our series. This again underlines the impact of proliferation in luminal cases. However, our analysis, with four factors, could be an easier manner to study ER-positive cases.

The fact that in our series the patients all received adjuvant tamoxifen stratified on chemotherapy suggests also that these factors could be more than prognostic in cases receiving hormonal therapy.