Materials and methods

Patients and samples
After an initial diagnostic biopsy, including characterization for p53 gene alterations, 92 women who were diagnosed with locally advanced breast carcinoma and who underwent primary chemotherapy were included in this study. The median clinical follow up was 78months (range 10-120months). Three of these patients had bilateral lesions, and in these three cases both lesions were examined. No family history for breast cancer was recorded in the 92 women. The patients received chemotherapy treatment (four or six courses, each lasting 21days) with a regimen containing cyclophosphamide, doxorubicine and 5-fluorouracil. The criteria for inclusion were as follows: inflammatory carcinomas, positive nodes and/or large (T3, T4) tumours. Clinical response to primary chemotherapy was categorized according to World Health Organization criteria and was considered as objective response (complete or partial response) or no response (stabilization or progression). In all cases neither radiotherapy nor hormone therapy were applied before chemotherapy.
Tumours were characterized before treatment by the clinical tumour size (categorized as T2, T3 or T4); the clinical nodal involvement (categorized as N0, N1 or N2); the histologic grade of Scarff, Bloom and Richardson (categorized as SBR1, SBR2 or SBR3); the hormone receptors (categorized as HR- or HR+, and considered as positive for oestrogen and/or progesterone receptors); the pathological tumour size (categorized as pT0, pT1, pT2 or pT3); and the number of involved axillary nodes (categorized as pN-, pN1+ and pN3+ for none, one or two, and three or more involved nodes, respectively).
Peripheral blood lymphocytes were obtained from each patient. Tumour samples were frozen in liquid nitrogen and stored at -80°C until analysis. Genomic DNA was extracted using proteinase, followed by phenol extraction and ethanol precipitation according to standard procedures [24].

Polymerase chain reaction method
The GSTM1-null genotype was determined by coamplifi-cation with the interferon-β gene, which served as an internal control. Primers for amplification of the GSTM1 gene corresponding to exon 4, intron 5 and exon 5 were 5' -ctgccctacttgattgatggg-3' and 5' -ctggattgtagcagatcatgc-3' (amplified product size, 271 base pairs). Primers for amplification of a part of the interferon-β gene, producing a constant 170-base pair band in all samples, were 5' -ggcacaacaggtagtaggcg-3' and 5' -gccacaggagcttctgacac-3' . Because the primers for the GSTM1 locus anneal to sites inside the coding region of the gene, the presence of the gene was determined by the presence of the band, whereas the null-genotype was determined by the lack of the band, using agarose gel electrophoresis (2%).
PCR was performed using 250ng template DNA in 10mmol/l Tris-HCl (pH8.4), 50mmol/l potassium chloride, 1.5mmol/l magnesium chloride (Bioprobe Systems, Illkirch, France), 0.2mmol/l concentrations of each deoxynucleotide triphosphate (dNTP), 500nmol/l concentrations of each primer and 2.5units of Taq DNA polymerase (Bioprobe Systems). The reaction (total volume 50 μ l) was amplified on a Omnigene thermal cycler (Hybaid Ltd, Ashford, Kent, UK). After an initial denaturation at 94°C for 3 min the reaction proceeded for 25 cycles of 50s at 94°C, 50s at 55°C and 50s at 72°C, concluded by a final extension step of 10min at 72°C. To test for contamination, negative controls (tubes containing the PCR mixture without the DNA template) were included in every run.

Reverse transcription polymerase chain reaction
Total RNA was obtained by the acid guanidine thiocyanate-phenol-chloroform extraction method [25]. Of the total RNA 1 μ g was dissolved in 20 μ l reverse transcriptase buffer (Gibco/BRL, Cergy Pontoise, France) containing 200 μ mol/l dNTP, 500 ng random hexamer and 200U avian myeloblastosis virus (AMV) reverse transcriptase (Superscript; Gibco/BRL). The reaction was incubated at 42°C for 50min and heated to 70°C for 15min. 
The PCR reaction was carried out in a 50 μ l volume of PCR buffer (Promega, Madison, Wisconsin, USA) containing 200 μ mol/l dNTP, 2.5 μ l complementary DNA template, 2.5U Taq DNA polymerase, and 500nmol/l of both GSTM1 and β2- microglobulin primers. The reaction was initiated by a heat step at 95°C for 2min, and carried out for 25 cycles of denaturation at 94°C for 50s, annealing at 55°C for 50s and extension at 72°C for 50s. Blanks for each reaction were included with all samples. Of PCR product 10 μ l were analysed on a 2% agarose gel with ethidium bromide staining. Band intensities were determined with a gel doc 1000 UV system (Biorad, Ivry-Sur-Seine, France), and the ratio of GSTM1 to β2- microglobulin was calculated.

Determination of p53 mutations
The determination of p53 mutations was performed as previously described [23]. Briefly, breast tumour samples were characterized before and after treatment for p53 gene mutations by PCR single-strand confirmation polymorphism and/or direct sequencing of exons 5-9 of the p53 gene.

Hormonal receptor assay
The oestrogen and progesterone receptor levels were determined in cytosolic tumours using enzyme immunoassay methods (Abbott Laboratories, Rungis, France). The cutoff level used for oestrogen and progesterone was 20fmol/mg cytosolic proteins.

Statistical analysis
The Pearson χ2 test was used as a homogeneity test for proportion. A stepwise logistic regression model with the logoistic regression (LR)-BMDP program (University of California Press, Berkeley, California, USA) [26] was used to assess the contribution of each independent factor to the GSTM1-null genotype and the probability of response to primary chemotherapy. The log-rank test using the Kaplan-Meier method was used to study the relationship between each factor and the probability of disease-free survival (median 50months) and overall survival (median 78months). A multivariate analysis using the Cox proportional hazards model was performed to assess the contribution of each independent factor to the probability of survival. For logistic regression or Cox regression models the enter and remove limits were 0.1 and 0.15, respectively. Overall significance of each factor (P value) was given by the likelihood ratio test. Results are expressed as odds ratios (OR) and corresponding 95% confidence interval (CI), with the significance of ORs being derived from the ratio of the coefficient divided by its standard error (Wald test). Statistical analysis for RNA GSTM1 expression was performed using the SAS/STAT program (SAS Institute Inc, Cary, North Carolina, USA) by using student t-test or analysis of variance, Pearson χ2 test and log-rank test. For all tests, optimality of the selected models was verified by all-possible-subsets analyses.