Two new studies suggest that tests based on Cyclin E or microarray analysis have the potential to outperform conventional criteria predicting the outcome of breast cancer.
Breast cancer continues to be a major cause of premature death in women of Western societies, despite progress in early detection and treatment and advances in our understanding of cancer's molecular basis. Besides the obvious need to develop even more efficient and cancer-cell−specific drugs, a particular concern has been to accurately predict the outcome of primary treatment among patients with early stage disease, like lymph-node−negative status. Which patients can be cured by surgery alone and which require additional (adjuvant) endocrine or cytotoxic systemic drug treatment such as tamoxifen, antracyclins or taxanes? Also, many node-positive patients are at low risk—if they only could be correctly identified, they could be spared unnecessary and costly treatment.
Two recent papers in the New England Journal of Medicine describe strong correlations between biological factors of tumors and clinical outcome of breast cancer, but use quite different strategies to accomplish this important task. Keyomarsi et al.1 determine the level of the cyclin E protein and its various isoforms using traditional immunohistochemical assays or western blot analyses of tumor lysates, whereas van de Vijver et al.2 use a global approach and dense DNA microarrays. Van de Vijver et al. analyze the transcript level of thousands of genes followed by state-of-the-art statistical methods to define gene-expression signatures of low- and high-risk tumors. Both studies take advantage of five- to ten-year-old tumor banks and associated patient data.
Cyclin E is a well-known cell-cycle regulator and key executor of growth-promoting stimuli. Cyclin E is induced in late G1 phase when cells decide whether to undergo DNA replication and subsequent division, or to fall into a dormant G0 phase. Cyclin E, in complex with its kinase partner CDK2, phosphorylates the retinoblastoma protein, thereby releasing transcription factors essential for programmed DNA synthesis, and is targeted by tumor-suppressor proteins such as p53, p21 and p27. Notably, Keyomarsi et al. refine the designation of high and low cyclin E levels by allowing detection of both full-length and low-molecular-weight isoforms of the protein, using gel electrophoresis and antibodies against carboxy-terminal epitopes retained in the smaller cyclin E isoforms or cleavage products. The authors report a very strong and independent correlation of cyclin E levels with survival in both node-negative and node-positive breast cancer. This correlation is most remarkable in their set of stage I (node-negative and size 20 mm or less) tumors, where none of 102 cyclin-E−negative, but all 12 cyclin-E−positive, cases died within five years of diagnosis (Fig. 1). Unfortunately, some standard prognostic factors were not used in the analysis. These include histological grade and markers of cell proliferation (Ki-67, TLI and S-phase fraction) or invasion (urokinase plasminogen activator).
Figure 1. Prognostic performance of microarray and cyclin E analyses.
a, Tumors with good and poor signatures are compared in lymph-node−negative stage I−II patients (tumor size 50 mm or less). b, Low and high total cyclin E levels are compared in stage I patients (lymph-node−negative and tumor size 20 mm or less).
The current work by van de Vijver et al. extends their recent study3 of 78 young (<55 years) node-negative patients, selected to include both cases with early recurrence and cases with long, disease-free survival. The authors identified a set of 70 differentially expressed genes, which optimally predicted clinical outcome (Fig. 1). In their current extended study of 295 consecutive, young (<53 years), stage I−II breast cancer patients, the authors included both node-negative and node-positive cases. They used the predefined set of 70 marker genes to classify tumors according to a good- or poor-prognosis signature—and then analyzed patient outcome. Patients whose tumors had a good-prognosis signature were largely free of recurrence at the ten-year follow-up (85%) compared with patients whose tumors fell in the poor-prognosis category (50%).
The poor-prognosis signature also strongly correlated with high histological grade and negative estrogen receptor (ER) status. Nonetheless, the gene-expression signature outperformed both the NIH consensus and the St. Gallen criteria for high-risk breast cancer, which rely on more-traditional indicators.
Importantly, the prognosis signatures of van de Vijver et al. performed equally well in node-negative and node-positive patients. This indicates that the processes of dissemination of cancer cells by means of blood and lymph vessels are different; the former is dependent on critical genetic alterations that are early events in some tumors, whereas the latter is a passive process reflecting the chronologic status of the tumor4.
How do these two studies compare with each other? Could the analysis of a single factor using traditional western blot technique outweigh a state-of-the-art holistic gene expression profiling approach and sophisticated statistical efforts? The answer must await confirmatory studies and analysis of the same set of tumors with both assays. However, the Keyomarsi et al. study casts light on the significance of taking post-translational protein modifications into account when evaluating the role of key cellular regulators, information that will not be available from transcript signatures.
Cyclin E expression is tightly connected to cell proliferation and its prognostic value may to a certain extent mirror the general adverse effects of fast-growing tumors. Moreover, the low-molecular-weight cyclin E isoforms can be surrogate markers of related cellular processes, reflecting upstream gene alterations such as protease activation5 or loss of ubiquitin ligation6. More directly, high cyclin E levels, especially of the constitutively expressed isoforms, may cause chromosomal instability7 and polyploidization by endoreplication8.
Interestingly, 1 of the 70 classifiers used in the van de Vijver paper is the cyclin E2 gene, which has a pattern of expression and function that is both distinct and redundant to cyclin E(1) (ref. 9). The microarray used in the previous study by the same group3 contains a cyclin E1 clone, but its signal did not pass the tests used for measurement and was hence not part of their analysis yielding the 70 classifying genes.
A major concern in retrospective studies such as these is whether the analysis reveals true prognostic factors or a combination of factors related to the adjuvant treatment. How appropriate are these new factors in a prospective setting and for the individual patient faced with diagnosis? Both studies fall somewhat short for these real-life questions as they are based on materials from patients that underwent different types of therapy—either adjuvant anti-estrogen or chemotherapy. Obviously, future selection of patients that should be spared adjuvant treatment cannot be based entirely on studies in which treatment is confounding the results.
ER status is a weak prognostic factor per se, but a powerful predictor of response to anti-hormonal therapy10. ER status may also be related to the effect of chemotherapy-induced ovarian ablation in pre-menopausal women11. ER status has also been shown to be a major determinant of tumor phenotypes, broadly dividing breast cancer in two main subclasses with profoundly different gene-expression profiles12,
13. Accordingly, the prognostic value of any factor closely related to ER status will also be influenced by adjuvant treatment. Indeed, cyclin-E−driven breast cancers are distinct from ER-positive tumors14, and the good-prognosis signature described by van de Vijver et al. overlaps markedly with ER status. The prognostic performance of cyclin E and the microarray-based gene-expression signature must thus be critically evaluated, taking adjuvant treatment into account. However, as untreated retrospective tumor materials are scarce, an alternative solution would be to evaluate new prognostic markers on a uniformly treated patient cohort, or in hormone-dependent and -independent tumor subsets separately.
Both of the current studies have the potential of making a lasting impact on breast cancer treatment. The simplicity of the cyclin E assay is attractive and its future role will soon be confirmed or challenged. Whether the particular prognostic signatures defined in the current microarray-based study will survive, or fall into the large archives of other 'promising' factors, is somewhat less clear. In all likelihood, DNA microarray techniques will continue to have an important role in disease characterization. Breast cancer is a heterogeneous disease and thorough analysis of large sets of tumors will allow new molecular classification systems and drug targets to be defined (for instance, by activated signaling pathways, metabolic profiles and interactions with surrounding tissues). More-difficult tasks lie ahead in predicting therapy response to achieve the ultimate goal of 'individualized' treatment of cancer patients, as these algorithms must take additional parameters into account, such as variability in drug metabolism and other constitutional features.
Keyomarsi, K. et al. Cyclin E and survival in patients with breast cancer. N. Engl. J. Med.347, 1566–1575 (2002). | Article | PubMed | ISI | ChemPort |
Van de Vijver, M.J. et al. A gene-expression signature as a predictor of survival in breast cancer. N. Engl. J. Med.34, 1999–2009 (2002). | Article |
van't Veer, L.J. et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature415, 530–536 (2002). | Article | PubMed | ISI | ChemPort |
Bernards, R. & Weinberg, R.A. A progression puzzle. Nature418, 823 (2002). | Article | PubMed | ChemPort |
Porter, D.C. et al. Tumor-specific proteolytic processing of cyclin E generates hyperactive lower-molecular-weight forms. Mol. Cell. Biol.21, 6254–6269 (2001). | Article | PubMed | ISI | ChemPort |
Strohmaier, H. et al. Human F-box protein hCdc4 targets cyclin E for proteolysis and is mutated in a breast cancer cell line. Nature413, 316–322 (2001). | Article | PubMed | ISI | ChemPort |
Bermejo, R., Vilaboa, N. & Cales, C. Regulation of CDC6, Geminin, and CDT1 in human cells that undergo polyploidization. Mol. Biol. Cell13, 3989–4000 (2002). | Article | PubMed | ISI | ChemPort |
Payton, M. & Coats, S. Cyclin E2, the cycle continues. Int. J. Biochem. Cell. Biol.34, 315–320 (2002). | Article | PubMed | ISI | ChemPort |
Clark G. In Diseases of the Breast (eds. Harris, J., Lippman, M., Morrow, M. & Hellman, S.) 461–485 (Lippincott-Raven Publishers, Philadelphia, 1996).
Sjostrom, J. Predictive factors for response to chemotherapy in advanced breast cancer. Acta Oncol.41, 334–345 (2002). | Article | PubMed | ISI | ChemPort |
Perou, C.M. et al. Molecular portraits of human breast tumours. Nature406, 747–752 (2000). | Article | PubMed | ISI | ChemPort |
Gruvberger, S. et al. Estrogen receptor status in breast cancer is associated with remarkably distinct gene expression patterns. Cancer Res.61, 5979–5984 (2001). | PubMed | ISI | ChemPort |
Loden, M. et al. The cyclin D1 high and cyclin E high subgroups of breast cancer: Separate pathways in tumorigenesis based on pattern of genetic aberrations and inactivation of the pRb node. Oncogene21, 4680–4690 (2002). | Article | PubMed | ChemPort |
