Introduction
A variety of genes involved in breast cancer biology have been studied and proposed as prognostic or predictive biomarkers, but only a few of them, such as hormone receptors and ERBB2, are used today to classify breast cancer patients and to make treatment decisions in the clinical routine [
1,
2]. The introduction of microarray analysis recently lead to a better characterization of breast cancer on a molecular level, underlining its biological heterogeneity and revealing that breast tumors can be grouped into different subtypes with distinct gene expression profiles and prognosis [
3]. Some of these subtypes confirmed the relevance of established differences between phenotypes such as the estrogen receptor (ER) and ERBB2 status, but also identified novel breast cancer subtypes or prognostic signatures of potential clinical value [
3‐
7]. Although little overlap was observed between these gene signatures at the level of individual genes, recent data indicate that the underlying biological processes and pathways might be common [
8‐
10].
In terms of tumor biology, proliferation has been recognized as a distinct hallmark of cancer and as an important determinant of cancer outcome [
11‐
13]. Increased tumor cell proliferation is accompanied by cell matrix remodeling and neo-angiogenesis, which together form the basis for an aggressive tumor phenotype [
14,
15]. This observation was further underlined by recent reports showing that several genes involved in gene signatures discriminating clinically relevant breast cancer subtypes were related to proliferation [
3,
4,
9,
16,
17].
In the context of breast cancer molecular screening, we recently investigated by quantitative RT-PCR the expression of 60 tumor-related genes in various subsets of breast cancers from the Stiftung Tumorbank Basel (STB) [
18,
19]. This gene set also comprised several genes involved in proliferation such as thymidilate synthase (TYMS), thymidine kinase 1 (TK1), topoisomerase 2-alpha (TOP2A), survivin (BIRC5) and the transcription factor E2F1. Since these genes strongly correlated to one another and since the assessment of a single gene able to accurately predict breast cancer patients' outcome would represent major advantages for standard clinical use, we focused our efforts on the evaluation of E2F1 transcript levels as surrogate marker for proliferation. This transcription factor is well known for being involved in the cyclin/cyclin-dependent kinase/retinoblastoma pathway and for controlling the expression of more than 1,000 genes involved in cell proliferation, differentiation and apoptosis [
20‐
23]. In a set of 317 primary breast cancers patients with known clinical outcome (STB data set), we evaluated E2F1 mRNA expression levels with respect to other proliferation markers, ER and ERBB2 status and clinical outcome. All results obtained in our collective were subsequently validated in The Netherlands Cancer Institute (NKI) microarray data set comprising 295 breast cancer patients. Moreover, the prognostic value of E2F1 was compared with the 70-gene prognostic signature, and with other gene expression-based predictors such as the intrinsic subtypes, the wound response signature and the recurrence score available as reported by Fan and colleagues using the same NKI data set [
8].
Discussion
In the present study we demonstrated that the assessment of E2F1 mRNA as a surrogate proliferation marker is a strong determinant of breast cancer outcome, particularly suitable for identifying patients at very low risk of metastasis, comparable with gene expression-based signatures such as the 70-gene signature. The prognostic component of the ER and ERBB2 status as well as different gene signatures were found to be strongly related to tumor proliferation. In fact, a large subset of patients classified with very favorable outcome shared a common molecular tumor phenotype characterized by ER-positive and/or ERBB2-negative status and low proliferation (low levels of E2F1 as well as BIRC5,TYMS,TOP2A and TK1). Moreover, the results obtained in our data set analyzed by quantitative RT-PCR were successfully validated in an independent breast cancer data set using microarray technology.
Sotiriou and colleagues developed a gene expression grade index able to reclassify breast cancer patients with tumor histological grade 2 into groups with high risk of recurrence versus low risk [
9]. The gene expression grade index was developed on the basis of the analysis of five breast cancer microarray data sets including more than 600 tumors, from which the authors extracted a list of 242 genes associated with tumor grade and predicting patient outcome. Most of these genes were related to proliferation and cell survival, such as E2F1 and MKI67, BIRC5, TOP2A and STK6, all being highly correlated and providing similar prognostic information. In our study, we demonstrated that the detection of a single gene is sufficient to select tumors at low proliferation. A single gene assessment requires high RNA quality from fresh (frozen) tissue, however, and might be insufficient in cases of more heterogeneous RNA quality (for example, RNA from paraffin-embedded tissues).
Breast cancer has been successfully classified using microarrays into clinically relevant subgroups based on variations in gene expression patterns. Sorlie and colleagues showed that ER-negative tumors grouped into basal-like and ERBB2 subtypes, both with poor prognosis [
3]. In contrast, ER-positive breast cancers could be classified into luminal A and luminal B subtypes with significantly distinct prognosis: luminal A tumors displayed favorable outcome, whereas survival of patients with luminal B tumors was poor and comparable with those of the ER-negative ERBB2 and basal subtypes [
3]. Our classification in the NKI data set revealed that 81% of the tumors expressing low E2F1 levels (below this study's cutoff point) corresponded with luminal A subtype as defined by Fan and colleagues [
8], and subsequently had similar prognostic value (Figure
3b).
Van de Vijver and colleagues used a 70-gene prognostic signature to discriminate patients with good prognosis and poor prognosis [
5], which according to our analysis strongly correlated with E2F1 expression levels. As shown in Figure
2, patients defined as of good prognosis by the 70-gene signature had tumors expressing low E2F1 levels and were mainly ER-positive. Despite all observed correlations, multivariate Cox analysis of the NKI data set showed that E2F1 levels and the 70-gene prognostic signature retained additive significance when both covariates were included (Table
5). This is probably due to the fact that both markers classified, in addition to the overlapping patients at very low risk, patients at similar but higher risk who would not have been selected by either classifier alone (Table
3). Furthermore, we found that almost all ERBB2-positive and ER-negative tumors expressed high levels of E2F1 and were classified as of poor prognosis according to the 70-gene signature – suggesting an explanation of why Espinosa and colleagues were unsuccessful in improving the accuracy of the 70-gene signature by incorporating additional genes such as ERBB2 [
26].
Fan and colleagues [
8] recently demonstrated that the different gene-expression-based predictors including the 70 gene-signature, the intrinsic subtypes, the wound signature and the recurrence score were highly concordant to evaluate breast cancer outcome. Our analysis revealed that low proliferation as quantified by low levels of E2F1 represented a common determinant of patients with good prognosis (Figures
2 and
3). It has to be noted that the prognostic value of E2F1 was independent of the nodal status. Indeed, 40% of the STB tumors and 50% of the NKI tumors with low E2F1 expression levels belonged to nodal-positive patients at very low risk of metastases, reconfirming the impact of proliferation recently reported in a study evaluating breast cancer patients with 10 and more positive lymph nodes [
27,
28].
The STB and NKI data sets differed in adjuvant treatment modalities; in general, patients of the STB collective were older and consequently received more hormone therapy but less chemotherapy as compared with patients of the NKI collective. In this context, it has to be emphasized that treatment regiments were chosen independent of the E2F1 status (Additional file
1) and that E2F1 levels retained predictive survival significance in patients with and without different adjuvant treatments (Additional file
2). Multivariate analyses, however, revealed different treatment impacts in the two data sets (Tables
4 and
5). In the STB collective, chemotherapy was particularly significant in univariate Cox analysis but was nonsignificant in multivariate Cox models, suggesting that information about the higher risk cases receiving chemotherapy is already included in the combination of the other covariates. Since E2F1 is co-expressed or regulates genes such as TYMS, TK1 and TOP2A, which were mechanistically linked with response to 5-fluorouracil and anthracycline-based therapy [
16,
29‐
32], however, our results with respect to specific chemotherapy response should be further investigated.
Acknowledgements
VV and PU contributed equally to the work. This work was supported by the Stiftung Tumor Bank Basel, the Swiss National Foundation (Grant 3100-059819.99/1) and the NCCR Molecular Oncology of the Swiss National Science Foundation. OncoScore AG paid the costs for quantitative RT-PCR analysis. The authors are very grateful to Sabine Ehret for technical support and data management. They thank pathologists for providing tumor tissues, in particular Prof. H. Moch, Prof. M. Mihatsch and Prof. W. Wegmann. The authors are indebted to all clinicians for their collaboration in collecting clinical data, especially to Prof. D. Fink, PD Dr E. Wight and Dr K. Lüscher.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
VV, ML and SE-C designed the study. VV and ML contributed to the selection of the genes, selected primers, and supervised the RNA extraction and quantitative RT-PCR. VV and PU performed statistical analysis under the supervision of MD and PW. VV, PU, MD, PW and SE-C contributed to data interpretation. RF performed the pathological analysis of several samples and asserved the surgical samples for molecular analysis. CCB, HD, RF, FS and UE participated in designing the study and writing the manuscript. VV, PU and SE drafted the manuscript. All authors read and approved the final manuscript.