MATERIALS AND METHODS

Patients and tumours

All samples analysed were biopsies taken from patients with stage II–IIIa breast cancer and at least four involved lymph nodes (n = 66); the biopsies had been previously fixed in formalin and preserved in paraffin wax in the course of routine pathological assessment. All patients were entered into the Anglo Celtic I trial, and randomised in equal numbers to two “arms”. Arm 1 received induction treatment, consisting of four cycles of Doxorubicin, followed by conventional CMF chemotherapy. Arm 2 received the same induction treatment, followed by high dose cyclophosphamide and subsequent bone marrow transplantation. The particular patient subgroup of those entered into the Anglo Celtic I trial comprised those who were managed at the Northern Centre for Cancer Treatment, Newcastle General Hospital, Newcastle, UK.

For each patient, data were available concerning patient survival (including times to relapse and/or death). Also available were data describing tumour grade, tumour size, the proportion of involved nodes, and the presence of vascular invasion for each patient.

Assessment of antigen expression by immunohistochemistry

Immunohistochemistry was carried out using an indirect streptavidin–avidin biotin complex method. Briefly, sections were dewaxed and hydrated before blocking of endogenous peroxide using methanol/hydrogen peroxide. Antigen retrieval was carried out at high temperature and pressure using citrate buffer (200mM citric acid, 500mM NaOH, pH 6.0). Sections were then washed in 1× Tris buffered saline (TBS; 140mM NaCl, 50mM Tris/HCl, pH 7.6) and blocked using 20% normal rabbit serum (NRS) (in 1× TBS) for 10 minutes, before incubation with the primary antibody solution for one hour (table 1 lists the primary antibodies used and their dilution factors). After washing in TBS, sections were incubated with biotinylated rabbit antimouse secondary antibody (Dako, Ely, Cambridgeshire, UK) at a 1/500 dilution (in 20% NRS) for 30 minutes. After further washing, streptavidin–biotin (Dako) was added at a 1/100 dilution (in 20% NRS), and the sections incubated for 30 minutes. The final reaction was visualised using 3,3′-diaminobenzidine (Sigma, Poole, Dorset, UK) in a TBS/hydrogen peroxide solution for three minutes. Finally, sections were counterstained using Harris’s haematoxylin before being blued, and then dehydrated and mounted in DPX.

T1GFR status was assessed according to membrane and cytoplasmic expression, with the exception of c-erbB-4, which was assessed according to cytoplasmic and nuclear expression (essentially, no membrane staining was seen). The ER and progesterone receptor (PGR) were assessed according to nuclear expression. Evaluation of immunohistochemical staining was carried out by two histopathologists. Each estimated the proportion of tumour cells staining, and the average intensity of staining for the whole slide. Staining intensity was assigned a value of either 0 (negative), 1 (weak positive), 2 (intermediate positive), or 3 (strong positive). An average of the results for the two histopathologists was then taken. Cases in which the scores were discrepant by more than 10% (positive cells) or ⩾ 1 (intensity) were discussed using a conference microscope, and a value agreed. For each patient, the data were then combined to yield a final value reflecting both the proportion of positive cells and the intensity of staining. This was carried out according to the method of Detre,²⁰ to yield a value between 0 (completely negative) and 18 (very strong positive). The data were classed as continuous for the purpose of statistical analysis.

Statistical analysis

To assess the relations between variables, two statistical tests were applied. Inter-relations between the molecular antigenic variables (T1GFRs, ER, and PGR (continuous data)) were assessed using Spearman’s test for rank correlations (bivariate). Further analysis was carried out after categorisation of continuous variables into two groups (high and low). The cut off value for the T1GFRs was the median value, and for ER and PGR, the cut off for positivity was a Detre value of greater than 2.²⁰ Subsequent comparison with tumour grade, size, nodal involvement, and vascular invasion (categorical data) was then carried out using Fisher’s exact test. With the data in this categorical form, survival analysis was also carried out with respect to each variable. Kaplan Meier step graphs were constructed and survival analysis was carried out using the logrank test. All analysis was carried out using SPSS statistical software.

RESULTS

General statistics

Of the 66 patients examined, five patients had grade 1 tumours, 15 grade 2, and 46 patients grade 3. All patients had involved nodes, ranging from 18% to 100% of nodes examined, with a mean of 69.6%. Tumour diameter ranged from 0.9 to 8 cm, with a mean of 3.24 cm. Of the 66 patients examined, 39 patients had vascular involvement, whereas 23 did not. Regarding menopausal status, 43 patients were premenopausal, eight patients were perimenopausal, and 15 were postmenopausal. Thirty eight patients were assessed as ER positive (Detre score, > 2) and 38 were assessed as PGR positive (Detre score, > 2).

Expression of type 1 growth factor receptors

Immunostaining for EGFR, c-erbB-2, and c-erbB-3 was both membranous and cytoplasmic, whereas labelling for c-erbB4 was cytoplasmic and sometimes also nuclear, as described below. With regard to the expression of individual T1GFRs, the assessment of EGFR membrane staining showed 44 patients with some degree of staining (Detre score, > 0), and the median Detre score was 2. Fifty eight samples showed some cytoplasmic staining (median value, 5). With respect to c-erbB-2 staining, 64 patients showed some membrane staining (median value, 6) and 65 showed some cytoplasmic staining (median value, 6). Analysis of c-erbB-3 expression revealed that 48 (median value, 2) and 60 (median value, 5) patients had some degree of membrane and cytoplasmic staining, respectively. Regarding c-erbB-4 cytoplasmic expression, 61 cases showed cytoplasmic labelling (median value 6; fig 1), whereas 56 showed a degree of nuclear staining (median value, 2).

For c-erbB-2, the criteria for overexpression when assessed by immunohistochemistry is generally classed as being intermediate to strong membrane staining in > 10% of tumour cells (Dako Herceptest™). We did not use this cut off value in our statistical analysis, but purely as a comparative statistic, we found that 34 of the 66 patients overexpressed c-erbB-2 when assessed in this way. This is consistent with other studies, given that we have a high proportion of grade 3 tumours.¹⁶

Associations between variables

With regard to associations between the molecular antigenic variables assessed using Spearman’s test, positive associations were found between c-erbB-3 and c-erbB-4 cytoplasmic expression (p = 0.003). As expected, ER and PGR expression showed a significant correlation (p < 0.001). Negative associations were found between EGFR membrane expression and both ER (p = 0.001) and PGR (p = 0.023) expression.

With regard to additional assessment carried out after categorising the above variables into high and low groups (about the median) and comparison of the molecular variables with grade, tumour size, nodal involvement, and vascular invasion, a positive association was seen between EGFR membrane staining and tumour grade (p = 0.025). No other significant associations were found.

Survival analysis

Survival analysis was carried out with respect to each variable. We found that high tumour grade and size were associated with poor survival, although significance was not quite reached. ER and PGR expression were both associated with improved survival, with ER reaching significance (p < 0.001; fig 2D). Of the T1GFRs, both EGFR and c-erbB-2 cytoplasmic expression were associated with poor survival, but the results did not reach significance (p = 0.167 and p = 0.171, respectively; fig 2A,B). c-erbB-3 expression showed no association with survival. However, upon analysis of c-erbB-4, patients with high cytoplasmic expression were found to fare significantly worse than those with low expression (p = 0.002; fig 2C).

DISCUSSION

In our study, despite the small number of patients studied, we have observed several established trends. First, the negative association seen between ER and EGFR expression (p = 0.001) is now well documented.¹⁷ Second, the survival trend observed with respect to ER expression (p < 0.001) is also well established.¹⁸

Regarding the T1GFR family, as stated previously, c-erbB-2 expression has been well characterised, and expression is known to be associated with a poor prognosis in breast cancer. Our results are consistent with this, although the result was not significant (p = 0.171). Concerning EGFR, our results showed a non-significant advantage for patients expressing low amounts of this receptor (p = 0.167). This result is in agreement with many previous studies.^2,9,10

Very little information exists with respect to the prognostic implications of c-erbB-3 and c-erbB-4 expression. A large study of c-erbB-3 expression and its association with clinical outcome was performed by Travis et al.¹¹ This study of 501 patients with breast cancer showed that 15% and 35% of patients with primary operable and advanced disease, respectively, had increased expression relative to normal tissue. However, no association with survival was found. Other studies by Lemoine and colleagues¹² and Quinn and colleagues¹³ showed no significant relation between c-erbB-3 expression and survival, and our results are also consistent with these findings.

We found that patients expressing high amounts of cytoplasmic c-erbB-4 fared significantly worse that those expressing low amounts (p = 0.002). To our knowledge, this association has not been seen before. In a previous clinical study performed by Kew and colleagues¹⁴ no association between c-erbB-4 expression and survival was demonstrated in a study of 127 cases of breast cancer. The scoring system used was different from that used in our study because the proportion of positive cells was not taken into account, but rather a system based on average intensity was used. In addition, a heterogenous patient group was studied (for example, node positive and node negative patients, with treatment differences), whereas in our study we have investigated node positive patients with one or two treatment strategies entered into a clinical trial. Another clinical study by Srinivasan et al found no association between nuclear c-erbB-4 expression and patient survival.¹⁵ Knowlden et al found that c-erbB-4 protein expression was associated with a prognostically favourable ER positive phenotype,¹⁸ and using the reverse transcription polymerase chain reaction for c-erbB4 mRNA, Suo et al also found an association with positive ER status.¹⁷ Pawlowski and colleagues¹⁹ also found an association between c-erbB-4 mRNA values and ER, in addition to a significant association with better prognosis; these observations may again relate to the patients studied and the methods of assessment.

“Our results are largely consistent with previous studies, but our observation of an association between c-erbB-4 and poor prognosis is new”

The observed nuclear expression of c-erbB-4 was unexpected on the basis of current models of how growth factor receptors function, but has been noted previously by other authors.¹⁴ Experiments are currently under way using methods with higher resolution than immunocytochemical staining to determine whether this represents genuine intranuclear staining or perinuclear localisation.

We also found several additional associations between variables. First, we found a positive association between c-erbB-3 and c-erbB-4 cytoplasmic expression (p = 0.003). This may reflect a preferred heterodimerisation partnership in some cancers, although the cell membrane is the actual site for dimerisation. In addition, we found that EGFR expression was positively associated with tumour grade (p = 0.025). This is consistent with the fact that high EGFR expression and high grade are both associated with poor prognosis.

We acknowledge that our study has certain drawbacks, the most obvious of which is the small number of patients examined, although the series is relatively homogeneous. Despite this, however, we have demonstrated several established trends, which help to strengthen our observations. Our results are largely consistent with previous studies, but our observation of an association between c-erbB-4 and poor prognosis is new. The strength of the association suggests that the trend may well be genuine. If so, c-erbB-4 may be useful as a prognostic indicator in breast cancer, identifying those patients in whom alternative or more aggressive treatment may be more effective, at least in the node positive group. It may also represent a new target for anticancer treatment in a similar manner to c-erbB-2 targeting by Herceptin®.⁵ There is evidence to suggest that c-erbB-2 may predict response to anthracycline treatment,²¹ and one study in non-small cell lung cancer showed that c-erbB-4 may predict the response to gemcitabine-cisplatin²²; the role of c-erbB-4 as a predictive factor in breast cancer needs to be explored. Further investigations involving larger numbers of patients, including node negative patients, are clearly warranted. In these studies, it may be of value to investigate the significance not only of the expression of growth factors individually, but also to examine the profile of expression—for example, coexpression of c-erbB-2 with one or more heterodimerisation partners.