Previously untreated operable primary breast cancer in 127 Caucasian women (age range, 32 to 89 years; median age, 60 years) with sufficient tumour tissue surplus for diagnostic requirements and complete clinical and pathological data was analysed. Tumour tissues were macrodissected by a specialist breast pathologist and snap-frozen in liquid nitrogen prior to storage at -80°C. Samples were examined following Local Research Ethics Committee approval under delegated authority from the Tayside Tissue Bank. The Tayside Tissue Bank has received ethical approval for its activities (REC Reference 07/S1402/90).

Approximately 10 mg of tumour tissue (>40% of tumour cells) was homogenised in 750 μL of QIAzol lysis reagent (Qiagen Ltd, Crawley, West Sussex, UK), and total RNA was extracted (Qiagen). RNA quality was assessed using the BioAnalyzer 2100™ (Agilent Technologies, Palo Alto, CA, USA) prior to reverse transcription-polymerase chain reaction (RT-PCR) analysis, and all samples with a 28S:18S ratio <1.2 were discarded. RT was performed with 0.5 μg of total RNA using the Cloned AMV Reverse Transcription Kit (Invitrogen, Paisley, UK), and cDNA quality was confirmed by PCR amplification of actin. Samples for which actin could not be amplified after 30 cycles of PCR were discarded. p53 isoform cDNA was amplified by two consecutive PCR assays (nested PCR) of 30 cycles each, and the PCR primers used were specific for each of the p53 isoforms analysed. The different primers used and their corresponding sequences are indicated in Table S1 in Additional file 1. To determine p53γ mutation status, the entire open reading frame of the isoform was sequenced using the Sanger method (BigDye Terminator, ABI 3730 Genetic Analyser; (Applied Biosystems, Warrington, UK) with the primers JWF (5'-AGCCAAGTCTGTGACTTGCA) and MP9ER (5'-TCTCCCAGGACAGCACAAACACG).

The p53 mutation status was determined using 100 ng of genomic DNA extracted from homogenised frozen tissues as described previously using the AmpliChip p53 Test (Roche Diagnostics, Pleasanton, CA, USA) 21 .

Immunohistochemical staining was carried out on 4-μm sections of formalin-fixed, paraffin-embedded tumours with the mouse monoclonal anti-oestrogen receptor α (ER) antibody 6F11 (Novocastra Laboratories Ltd, Newcastle upon Tyne, UK), progesterone receptor (PR) antibody clone 16 (Novocastra Laboratories Ltd) and mouse monoclonal anti-HER2 antibody CB11 (Novocastra Laboratories Ltd). Additional analyses were performed according to histological tumour grade (graded by a specialist consultant breast pathologist); pathological tumour size (pT1, tumours <2 cm; pT2 and pT3 cancers, tumours ≥2 cm) 22 . ER status (ER negative 0 to 3 versus ER positive 4 to 18) was determined using the quickscore method 23 . Briefly, immunoreactivity scored semiquantitatively for both the intensity and the proportion of cells staining. Intensity was given scores from 0 to 3 (no staining = 0, light staining = 1, moderate staining = 2 and strong staining = 3) and proportion was given scores from 1 to 6 (0% to 4% = 1, 5% to 20% = 2, 21% to 40% = 3, 41% to 60% = 4, 61% to 80% = 5 and 81% to 100% = 6). The two scores were then multiplied to obtain the final result of 0 to 18. Human epidermal growth factor receptor 2 (HER2) scoring was performed as previously described 24 .

The primary outcomes in this study were breast cancer-specific overall survival (abbreviated to overall survival) and breast cancer-specific disease-free survival (abbreviated to disease-free survival or cancer recurrence throughout the text), and accordingly, non-breast cancer deaths were censored at the time of death (that is, at the time of death, the women were considered to have survived breast cancer but died as a result of other causes). Statistical analysis was performed using Minitab version 15.1.0.0 statistical software (Minitab Inc., PA 16801-3008, USA) for χ², two-sided Fisher's exact test and Kaplan-Meier analyses. These univariate analyses test for associations between variables in a pairwise manner (for example, A versus B), but they do so without adjusting for influences exerted by other associated variables (for example, both A and B may be associated with confounding variables C, D and E, casting doubt on the validity of the relationship between A and B).

To clarify the univariate analyses and adjust for possible confounding variables, the selected variables were interrogated using the multivariate methods of binary logistic regression (BLR) with associated odds ratios (OR) and the Cox proportional hazards regression model (CR) with associated risk ratios (RR), both utilising the backwards stepwise elimination method. (For more detailed methods, read the "method" section in Additional file 2.)

In the tables of results for these multivariate analyses, the β value is a regression coefficient that indicates the strength of association between the predictor and response variables, where a large β indicates a strong association. A positive β indicates a positive association between the predictor and response variables, whilst a negative β indicates a negative association.

The OR is used to assess the risk of a particular outcome if a certain factor (or exposure) is present, indicating how much more likely it is that someone who is exposed to the factor under study will develop the outcome as compared to someone who is not exposed. If the odds are greater than 1, then the event is more likely to happen than not, whilst if the odds are less than 1, then the event is less likely. One 'reads' the risk ratios in precisely the same way.

The results of the univariate and multivariate analyses were consistent, and for clarity and brevity only the results of BLR, CR and Kaplan-Meier analyses are presented. Throughout the analyses the null hypothesis was rejected at an α level of 10% (P < 0.10), and observations considered to be marginal (that is, worthy of further analysis) for an α level between 5% and 10% (0.05 ≤ P ≤ 0.10) and significant at 5% (P < 0.05). The P value represents the probability of error that is involved in accepting our observed result as valid. For example, P = 0.05 indicates that there is a 5% probability that the relation between the variables found in the sample occurred by chance.

Cancers from 127 women (median age, 60.0 years; age range, 32.1 to 89.1 years) were examined. The majority of cancers were ductal carcinomas at 84% (107 of 127). Of these cases, 77% (98 of 127) were ER-positive, 62% (79 of 127) were PR-positive, 14% (17 of 119) were HER2-positive and 22% (28 of 127) had a tumour containing mutant p53. Approximately 50% (63 of 127) of the patients had axillary lymph node metastasis; tumours were grade 1 (16 cancers), grade 2 (48 cancers) or grade 3 (61 cancers), respectively. This patient population was therefore representative of symptomatic primary breast cancers in a Western country.

Expression of p53β and p53γ was successfully analysed in the 127 primary breast cancers by using RT-PCR (Figure 1). On testing in triplicate, breast cancers consistently demonstrated p53β expression (36%; 46 of 127) and p53γ expression (37%; 47 of 127). Only 19% (24 of 127) of tumours expressed both p53β and p53γ.

Univariate statistical analysis determined that both p53β and p53γ were associated with clinical markers (data not shown). To clarify these associations and adjust for possible confounding variables, BLR analyses were performed to examine the associations of the various clinical markers with p53β and p53γ (Tables S2 and S3 in Additional file 1). p53β isoform expression was independently associated with p53γ expression (P = 0.008, BLR) (Table 1), and p53γ expression was independently associated with p53 mutation (P = 0.002, BLR) (Table 1). However, after adjusting for other associated clinical markers, p53β expression was not associated with p53 mutation (P = 0.970, BLR) (Table 1). For tumours bearing the p53 mutation, p53γ cDNA was directly sequenced using Sanger methods and was found to contain the same p53 mutation identified by the AmpliChip p53 Test, indicating that p53γ was expressed by tumour tissue from the same allele as the p53 mutation and not by stromal tissues. Most p53 mutations were hemizygous missense mutations affecting the DNA-binding domain of p53. Since the p53 gene was mutated at different codons in our cohort of breast cancer, there were not enough cases with the same p53 mutation for the statistical analysis.

p53β expression was independently associated with ER status (P = 0.033, BLR) (Table S4 in Additional file 1), but not with PR status. p53γ expression was not associated with either ER or PR status. p53β and p53γ isoform expression was not associated with tumour type, menopausal status, age of cancer onset or HER2 status (data not shown).

As expected, p53 mutation was independently associated with cancer recurrence and death (P = 0.013 and P = 0.017, respectively; BLR) (Table 1). However, neither p53β nor p53γ isoform expression was associated with cancer recurrence (P = 0.198 and P = 0.636, respectively) (Table 1) or death (P = 0.082 and P = 0.783, respectively) (Table 1).

To determine whether the associations between p53 mutation and the various clinical markers were different in p53β- or p53γ-positive tumours, data were stratified by p53β and p53γ expression status. BLR analyses were performed to examine the associations between markers in the p53β- and p53γ-positive and p53β- and p53γ-negative cohorts (Table 2, and Tables S2 and S3 in Additional file 1).

Regarding p53β, p53 mutation status was marginally associated with cancer recurrence and death in the p53β-negative cohort (P = 0.059 and P = 0.072, respectively; BLR) (Table S2 in Additional file 1), while in the p53β-positive cohort p53 mutation status was not associated with death but was associated with cancer recurrence (P = 0.018, BLR; Table S2 in Additional file 1).

Regarding p53γ, p53 mutation status was independently associated with cancer recurrence and death in the p53γ-negative cohort (P = 0.001 and P = 0.002, respectively; BLR) (Table 2, and Table S3 in Additional file 1). Interestingly, p53 mutation status was not associated with cancer recurrence or death in the p53γ-positive cohort (P = 0.579 and P = 0.282, respectively; BLR) (Table 2, and Table S3 in Additional file 1), despite the greater proportion of grade 3 cancers with p53 mutations (61.5%; 16 of 26) in the p53γ-positive cohort compared with the p53γ-negative cohort (25.7%; 9 of 35). These data suggest that p53γ expression delineates two subpopulations of mutant p53 breast cancer patients with markedly different outcomes.

To investigate the association between p53β or p53γ isoform expression and the clinical markers in relation to survival and cancer recurrence, we performed CR analyses that included p53β, p53γ, p53 mutation status and clinical markers (Table 3). These analyses demonstrated the expected associations between prognosis (death and cancer recurrence) and ER, PR, tumour grade, p53 mutation, HER2 or lymph node status (Table 3, rows a-f), but did not show any independent association for p53β or p53γ isoforms (Table 3, rows g and h).

Further, to determine the degree of interdependence between the variables p53β, p53γ and p53 mutation with respect to survival and cancer recurrence, we aggregated these variables into combined variables and reran the CR analyses. We thus formed a binary variable (p53m&p53β) that was positive when p53β was expressed and the p53 gene was mutated, but negative otherwise. Similarly, we formed a binary variable (p53m&p53γ) that was positive when p53γ was expressed and the p53 gene was mutated, but negative otherwise. The results of using such combined variables (Table 3, row i and row j) allowed us to determine that p53 mutation is no longer associated with death or cancer recurrence when p53β or p53γ is expressed (Table 3, compare row c with row i or row j). This effect was independent of ER status and therefore independent of endocrine therapy (in this study, all ER-positive patients were treated with tamoxifen 20 mg for 5 years as standard adjuvant therapy). Moreover, there was also no significant difference in the ER status of patients bearing p53 mutations between the p53γ-positive and p53γ-negative cohorts (P = 0.254, Fisher's exact test).

Furthermore, given the association between p53β and p53γ expression (Table 1), we performed CR analysis to determine the combined effects of p53γ and p53 mutation in the absence of p53β and reciprocally the combined effects of p53β and p53 mutation in the absence of p53γ. We formed a binary variable (p53m&p53β⁺ p53γ^-) that was positive when p53β was expressed in the absence of p53γ expression and the p53 gene was mutated (Table 3, row k). We also formed a binary variable (p53m&p53γ⁺ p53β^-) that was positive when p53γ was expressed in the absence of p53β expression and the p53 gene was mutated (Table 3, row l). These analyses revealed that p53 mutation in patients expressing p53β but not p53γ retained the association with death and cancer recurrence (Table 3, compare row c, row i and row k), while p53 mutation in patients without p53β but with p53γ was not associated with death and cancer recurrence (Table 3, row c, row j and row l).

Therefore, the apparent abrogation of the association of p53 mutation with poor prognosis in the p53γ-positive population (but not in the p53β-positive population) indicates that only p53γ allows the identification of a subpopulation of breast cancer patients expressing mutant p53 with a better prognosis than expected.

Using Kaplan-Meier log-rank analyses, patients with mutant p53 breast cancer had significantly worse disease-free survival and overall survival than those with wild-type p53 breast cancer (Kaplan-Meier log-rank test, χ² = 10.51, 1 df, P = 0.001; and χ² = 6.55, 1 df, P = 0.010, respectively) (Figure S1 in Additional file 3), with a more than three times increased risk of recurrence and death (hazard ratio (HR) = 3.48 and HR = 3.16, respectively). Expression of p53β or p53γ was not associated with cancer recurrence (Kaplan-Meier log-rank test, χ² = 0.05, 1 df, P = 0.817; Figure S2 in Additional file 4, bottom; and χ² = 0.15, 1 df, P = 0.694; Figure S3 in Additional file 5, bottom, respectively) or with overall survival (Kaplan-Meier log-rank test, χ² = 0.37, 1 df, P = 0.544; Figure S2 in Additional file 4, top; and χ² = 0.31, 1 df, P = 0.575; Figure S3 in Additional file 5, top, respectively).

Patients bearing mutant p53 tumours and expressing the p53γ isoform had disease-free survival and overall survival that were not different from patients bearing wild-type p53 tumours, with a low comparative risk of recurrence and a similar risk of death (HR = 1.72 and HR = 1.04, respectively) (Figure 2A: Kaplan-Meier log-rank test, χ² = 0.76, 1 df, P = 0.384; and Figure 2B: Kaplan-Meier log-rank test, χ² < 0.01, 1 df, P = 0.958, respectively). However, patients bearing mutant p53 tumours without p53γ isoform expression had a high risk of recurrence and subsequent high risk of death (HR = 7.21 and HR = 11.23, respectively) compared with patients bearing wild-type p53 tumours (Figure 2; Kaplan-Meier log-rank test: χ² = 18.33, 1 df, P < 0.001; and χ² = 20.70, 1 df, P < 0.001, respectively). Consistent with the CR analysis, the Kaplan-Meier log-rank analyses indicate that p53γ allows the identification of a subpopulation of breast cancer patients expressing mutant p53 with a better prognosis than expected.