In their review of multi-gene assays of breast cancer, Drs. Gökmen-Polar and Badve highlight the overall similarity of “first-generation” molecular assays that have been developed, using different strategies, to understand the relationship between gene expression within tumor samples and the outcomes of patients with breast cancer. These first-generation assays were designed to address molecular classification (intrinsic sybtypes and PAM50), prognosis of node-negative breast cancer without adjuvant systemic therapy (70-gene MammaPrint), and prognosis of node-negative breast cancer with adjuvant endocrine therapy (21-gene Recurrence Score, Oncotype DX). Overall, their relationship to outcomes is largely due to the fundamental influences of proliferation, of estrogenic stimulation, and (in a subset of patients) of the human epidermal growth factor receptor 2 (HER2) onco-protein on the transcriptional biology of breast cancer.[1,2] A recently published meta-analysis of more than 5000 breast cancer microarrays confirms that proliferation, estrogen receptor (ER) status, and HER2 status are key and interactive drivers of prognosis, such that three genes—that encode ER (ESR1) and HER2 (ERBB2), and a proliferation-related gene (aurora kinase-A [AURKA])—had the same performance as any first-generation genomic signature.[3] One must be highly cautious in the interpretation of genomic meta-analyses that pool data from various microarray methods and platforms employing profoundly different techniques. However, with regard to the identification of such robust genomic signals in breast cancer transcription, the results of this meta-analysis are probably reliable. Essentially, it is the combined information from ER, HER2, and proliferative activity that determines classification and prognosis.
The interplay between proliferation and ER-related biology is significant in the most common and transcriptionally diverse form of breast cancer, ie, ER-positive/HER2-negative disease. Interplay between these parameters is largely responsible for the performance of the Oncotype DX Recurrence Score, of the PAM50 score, and of subtypes. It appears that patients with lower-risk breast cancer benefit from endocrine therapy (vs placebo) but do not benefit from adjuvant chemotherapy, whereas patients with higher-risk breast cancers do not benefit from endocrine therapy but do benefit from chemotherapy.[2,4] This interpretation makes biological sense and is most likely correct for some patients. However, it may not be possible to generalize on the basis of the extent of benefit that was observed in node-negative disease in the National Surgical Adjuvant Breast and Bowel Project (NSABP) B-20 study. This is because the tamoxifen(Drug information on tamoxifen) treatment samples in that analysis were also used to select the 21 genes and to develop the Recurrence Score used in the Oncotype DX assay.[2] Therefore, the observed difference in outcomes between the chemotherapy and tamoxifen arms in the B-20 analysis might be exaggerated by training bias. Unfortunately, there is not another independent sample cohort to evaluate from a randomized trial of hormonal therapy vs chemo-endocrine therapy. Perhaps the less marked difference in breast cancer–specific survival between chemo-endocrine therapy and endocrine therapy for patients with a high Recurrence Score and node-positive breast cancer (Southwest Oncology Group [SWOG] 8814 study) would represent a more generalizable result if the predictive performance were independent of nodal status.[4] This illustrates some of the limitations of retrospective testing of available residual samples from past prospective trials in order to establish clinical utility.[5] Fortunately, recent and current prospective randomized clinical trials for node-negative disease (TAILORx) and node-positive disease (RxPONDER) should eventually provide unbiased estimates of the predictive accuracy of the Oncotype DX Recurrence Score as applied to current treatments—and might thus provide an opportunity to alter the thresholds at which the Recurrence Score defines different risk groups, if doing so would improve their clinical utility with current treatments. Indeed, I would argue that there is as strong a need for efficient and effective prospective validation trials to specifically test the clinical utility of tests for their intended use, as there is a need to test the efficacy of treatments.
If one were to rely solely on the lessons learned from the first generation of genomic tests, one might infer that the most chemosensitive and the most endocrine-sensitive subsets of ER-positive breast cancers are mutually exclusive. Sequential chemo-endocrine therapy is a proven component of adjuvant treatment in patients with stage II/III breast cancer, and survival rates continue to improve with each successive generation of improved chemotherapy for ER-positive breast cancer.[6-8] So, do only patients with endocrine-insensitive disease benefit from chemotherapy? In this context, we note with interest that the sensitivity to endocrine therapy (SET) index of ER-related genes (known proliferation-related genes were excluded from the SET index) identified 20% of patients with stage II/III breast cancer as having a high/intermediate SET index—and those patients had an excellent probability of survival following chemo-endocrine therapy, and one that was significantly better than that of patients whose cancer had a low SET index.[9] In addition, prior response to neoadjuvant chemotherapy was independently prognostic and appeared to be synergistic with predicted endocrine sensitivity (SET index).[9] This result appears to contradict prevailing beliefs because it supports the concept of sequential synergy of chemo-endocrine treatments, rather than the concept of mutually exclusive sensitivities to chemotherapy and endocrine therapy. It is possible that both concepts might be true—just in different subsets of patients. Clearly, further studies are needed to better understand how benefit from standard treatments can be predicted—perhaps by separate prognostic, endocrine-predictive, and chemopredictive assays.
The absence of proliferation-related genes probably explains why the SET index is predictive of survival in patients with node-negative breast cancer who receive adjuvant endocrine therapy, but is not prognostic in patients with similar disease who do not receive systemic therapy.[9] This suggests that the isolation of ER-related biology from proliferation allows more pure prediction of endocrine sensitivity. Similar results were reported from a comparison of how individual components of the Recurrence Score correlated with the effect of tamoxifen (vs placebo) in the NSABP B-14 trial samples.[10] The main implication is that baseline tumoral proliferation is informative with regard to prognosis, but it is probably not informative with regard to endocrine sensitivity.
Higher proliferation and a less differentiated state impart worse prognosis, but these factors also predict a higher probability of pathologic response to chemotherapy.[11-13] However, there is a paradoxical relationship between predicted pathologic response and significantly worse survival of the predicted responders.[12,14] The basis of this paradox is that proliferation is both a prognostic factor and the biological target of most chemotherapy treatments. For example, if the probability of a highly proliferative cancer achieving excellent pathologic response were 50%, then half of patients with highly proliferative disease who received chemotherapy would be essentially cured—but the other half would have residual disease with a poor prognosis. The latter patients would consequently be at higher risk for early relapse compared with patients whose residual cancer had low proliferation. This paradox is a major obstacle to the development of clinically useful predictors of chemosensitivity. To address this problem, we separately developed predictive signatures for SET, two definitions of chemotherapy resistance, and a definition of significant chemosensitivity.[14] Although this required a predictive algorithm that employed seven different gene signatures (to separately evaluate ER-positive/HER2-negative and ER-negative/HER2-negative breast cancer subsets), the result was accurate prediction of both response and survival in the ER-positive/HER2-negative and ER-negative/HER2-negative subsets.[14] Such results provided an interesting proof of concept that the dismantling of a complex problem into key components, followed by the development of genomic signatures for each component, and then reconstruction of those signatures into a predictive algorithm, might overcome the limitations caused by the paradoxical interactions of biology and thereby improve the clinical potential of genomic predictors. Again, further validation studies are needed.
Perhaps no single breast cancer assay should be used alone to guide treatment. This is a heterogeneous disease with a series of different treatment indications, each of which has a biological basis. Indeed, the next generation of molecular diagnostic platforms for prognostic and predictive tests of breast cancer might be far more complex than a single reverse transcriptase polymerase chain reaction (RT-PCR) or microarray signature. In order to be able to make predictions for each component of treatment, one might expect to have to use an algorithm involving specific signatures from each of the various assays. We would be wise to identify meaningful advances in this field, commit resources to appropriate validation of intended clinical use, and then add to the clinical portfolio only those tests that add useful and actionable results for patient care. The incremental advances from such an approach might save more lives and justify treatment resources more wisely than simply continuing with the status quo and hoping that an exciting “next wave” of technological advances will bring us all the answers we have been seeking. While expression profiling has already generated improved diagnostics and has more to offer in the future, we still do not know how much the incorporation of sequence information will add to predictive utility with respect to outcomes from breast cancer treatment. If the etiology and pathogenesis of breast cancers often have an endocrine or physiological basis, then clinically relevant driver mutations might be less common in breast cancers than in other forms of cancer—especially other cancers with pathogenesis that is driven by chemical or physical mutagenesis. For now, proper validation of diagnostic advances for clinical utility and monitored implementation of useful tests should be priorities in order to demonstrate incremental progress in treatment outcomes for patients. Meanwhile, researchers will continue to make progress on all fronts.
Financial Disclosure: Dr. Symmans has received research support from AstraZeneca. Dr. Symmans and the MD Anderson Cancer Center were cofounders of Nuvera Biosciences, Inc, and received equity and a royalty payment for pending and approved patents held jointly by Nuvera Biosciences and the MD Anderson Cancer Center.
