Supplementary Method

Supplementary method

Array data handling

After hybridisation and scanning the data were analysed using Bioconductor ( packages within the open source R statistical environment ( The quality control metrics recommended by Affymetrix, as well as boxplots and intensity histograms were used for quality assessment. After background correction by robust multi-array analysis (RMA), a filter using the standard deviation of gene expression values was applied to select the top most variable 10 000 genes and a double cut-off of false discovery rate (FDR) < 0.05 and a fold change ≥ 2 was applied. Differential expression analysis was examined using Limma {Wettenhall, 2004 #1676}; expression data for wild type T47D and ZR75-1 cells were accessed from GEO (GSE3156; {Bild, 2006 #1675}). More detailed analysis and validation of the array study will be published elsewhere.

Real-time PCR

Total RNA was extracted using RNAeasy Minikit (Qiagen) and treated with RNAse-free DNAse (Qiagen). RNA (1µg) was reverse transcribed with random hexamers and multiscript reverse transcriptase (Applied Biosystems) on the ABI Prism 7500, and expression normalised to GAPDH mRNA. Probe and primer Taqman sequences for MAGEA2 (assay ID Hs00606323_m1) were used.

Immunofluorescence

Cells were washed twice with PBS and fixed for 10 min in 4% paraformaldehyde in PBS. After washing with PBS, cells were permeabilised with 0.1% Triton-X-100 (Sigma-Aldrich) in PBS for 3 min. Samples were blocked for 45 min with 3% BSA (Sigma-Aldrich) in PBS. Primary and secondary antibodies were diluted in 3% BSA in PBS and incubated for 1 hr. Samples were washed three times with 0.1% Tween-20 in PBS after incubation with primary and secondary (FITC and Alexa-labeled) antibodies. The fixed samples were mounted in DAPI-containing mounting agent (ProLong Gold, Invitrogen) and analysed by confocal microscopy using a Ziess Axiovert 200M equipped with a LSM510 META detector (Zeiss) and a Plan-Neofluar 63X/1.4 DIC objective.

Missing values imputation and Cox proportional hazards model

There were 22 and 36 missing values for the grade and node variables, respectively. The binary Node missing values were imputed by predicted probabilities of a fitted logistic regression with cohort and age (continuous) as covariates (see Schemper & Heinze 1997 and Schemper & Henderson 2000). For the grade variable, missing values were imputed by the predicted mean value (rounded to the nearest actual data value) from an ordinal logistic regression with cohort and age (continuous) as covariates.

Incorporating the imputed values, a stepwise Cox’s model with backward elimination steps was performed using the MageA (0 or 1), Grade (1, 2 or 3) and Node (0 or 1) variables. The Node variable was removed from the model at the first step (p-value= 0.9515). The estimated coefficient, the hazard ratio (HR) with 95% CI and the p-value from the Wald test for each variable in the final model are shown in Table 1. The p-value of the likelihood ratio test, which tests the global null hypothesis, is 0.0001.

Table 1: model information

VariablecoefficientHR (95% CI) p-value
MageA 0.518 2.027 (1.228, 3.345) 0.006
Grade 0.410 1.847 (1.195, 2.853) 0.001

In terms of discrimination between those having and not having a death of breast cancer, the c-index for the risk score was 0.67. The c-index is a generalization of the area under the ROC curve.

In order to investigate whether there was significant overfitting during the development of the final model, the selected variables MageA and grade were repeatedly fitted (B=1000) in a bootstrap sample and the performance of the model was evaluated on the original sample (see Harrell 2001). The c-index for the original, training and test data is found in Table 2. A corrected c-index of 0.64 means prediction is better than random.

Table 2: original, training, test and corrected c-index using bootstrap
Original Training Sample Test Sample Corrected

c-index 0.67 0.68 0.66 0.65

References

Harrell F (2001). Regression Modeling Strategies with Applications to Linear Models, Logistic Regression, and Survival Analysis. Springer-Verlag.

Schemper M and Heinze G (1997). Probability imputation revisited for prognostic factor studies. Statistics in Medicine, 16:73-80.

Schemper M and Henderson R (2000). Predictive accuracy and explained variation in Cox regression. Biometrics, 56:249-255.

FIGURE LEGENDS

Supplementary Figure S1: T47D derived MAGEA2-expressing clones also continue to proliferate in tamoxifen-containing media

(A) Western blot (20μg WCE/lane) comparing MAGEA2 expression level in T47D-derived vector alone (VA) control cells with two MAGEA2-expressing stable clones, C30 and C34. Blot was also probed for total p53, acetylated p53, p21cip and Hsc70 (loading control), as indicated. (B) T47D wild type (WT), VA and clone C30 and C34 cells were seeded (3.5x105cells/well in triplicate) onto 6-well plates and treatment with 10-7M tamoxifen began 24 hrs later. Cells were counted (Coulter cell counter) daily for eight days. Growth rates at later time points declined due to cell confluence. C30 growth at days 6, 7 and 8 compared to VA control using Student’s t test; ***p<0.001.

Supplementary Figure S2: MAGEA3-overexpressing cells also proliferate in tamoxifen media

A MAGEA3 cDNA clone was obtained from the IMAGE consortium (ID: 3345801) and the insert transferred to pcDNA3.1. Stable MAGEA3-expressing clones C3 and C4 were established in MCF-7 cells as for MAGEA2. Each clone and VA control cells were seeded (3x104cells/well) onto 12 well plates and treated with 10-7M tamoxifen. Triplicate wells were counted daily for eight days. Cell numbers were averaged for both clones and compared to VA using Student’s t test; ***p<0.001.

Supplementary Figure S3: MAGEA2-overexpressing cells cannot proliferate in media supplemented with Fulvestrant

MCF-7 VA and C24 cells were seeded (3x104 cells/well) onto 12 well plates and given media containing the indicated dose of Fulvestrant (F) 24 hours later. Triplicate samples were counted (Coulter counter) daily. No significant difference in cell proliferation was observed between MAGEA2-expressing and VA cells. Experiment performed in parallel to that shown in Fig. 2C.

Supplementary Figure S4: Reduced apoptosis in MAGEA2-expressing MCF-7 and T47D lines in tamoxifen media

(A) Full cell cycle profile after FACS Calibur analysis of Annexin V-Propidium Iodide labelled stable MAGEA2-expressing MCF-7 clones (C18 and C24) compared with WT and VA control cells after 144 hrs in media with and without 10-7M tamoxifen. Triplicate wells were seeded for each line and condition and data show one of three replicate experiments. Graph showing percentage of early apoptotic cells shown in Figure 2D. (B) Experiment in (A) repeated using T47D clones (C30 and C34) compared with their WT and VA control cells; the percentage of early apoptotic cells for each line and condition is displayed. **p<0.01, Students t-test comparing samples indicated.

Supplementary Figure S5: Exogenous MAGEA2 binds p53 and localizes to the nucleus in MAGEA2-expressing and TR cells

(A) Lysates from T47D VA control and C30 MAGEA2-expressing cells were immunoprecipitated (IP) for p53 (left panels) or MAGEA2 (middle) and then blotted (WB) for each protein as indicated, to show complex formation between exogenous MAGEA2 and endogenous p53. 5% of IP input protein was blotted to show levels of p53 are equivalent in VA and C30 cells (right panel). (B) Nuclear (N) and cytoplasmic (C) extracts from T47D WT and TR cells were probed for MAGEA2, Acp53 and total p53 as indicated. The integrity of each fraction was checked using antibodies against compartment-specific proteins laminin A/C and actin. (C) Confocal imaging of

immunofluorescent staining of MCF-7 C24 cells grown +/- tamoxifen (TAM) for 48 hrs revealed that MageA2 became localized to the nucleus after tamoxifen treatment. Scale bar 20 µm

Supplementary Figure S6: MAGEA2 interacts with estrogen receptor in T47DTR cells

WCE lysates from T47D WT and TR cells were immunoprecipitated (IP) for ERα or

MAGEA2 as indicated. IPs were the blotted (WB) for the same proteins as indicated (left panel). 5% of the input proteins used for IP were also blotted for MAGEA2 and ERα plus Hsc70 as a loading control (right panel). The results confirmed that endogenous MAGEA2 protein also interacts with estrogen receptor in tamoxifen resistant T47D cells.

1