Predictive markers in breast cancer – the future
Dr J Bartlett, Endocrine Cancer Group, Edinburgh Cancer Research Centre, Edinburgh EH4 2XR, UK.
The published literature is awash with examples of new tissue biomarkers promising to predict responses to therapy in breast cancer patients. However, few, if any, of these progress from the laboratory to the clinic. In this review we discuss some of the reasons for this, illustrating our discussion with a selection of biomarkers which are in development and which may be candidates for clinical application within the next few years (topoisomerase IIα, epidermal growth factor receptor, AKT, phosphatase and tensin homologue). In particular, we explore how our ever increasing knowledge of molecular and pathway biology is facilitating hypothesis-driven biomarker discovery, and the statistical considerations which need to be addressed in order to validate new candidate biomarkers adequately.
Arimidex, Tamoxifen Alone or in Combination
Breast Cancer International Research Group
Breast International Group
epidermal growth factor receptor
mammalian target of rapamycin
National Epirubicin Adjuvant Trial
non-small cell lung cancer
phosphatase and tensin homologue
Taxotere as Adjuvant Chemotherapy Trial
Paclitaxel, Anthracycline, Gemcitabine & Cyclophosphamide
Tamoxifen and Exemestane Adjuvant Multicentre
In the last 35 years only two molecular markers, oestrogen receptor (ER) and Her2, have become standard measurements in the management of breast cancer patients.1,2 The long evolutionary cycle of predictive biomarkers seems at odds with the number of published articles on the subject, suggesting the need for a new, more rigorous approach to this area.3 At the time of writing, a simple literature search for ‘predictive markers in breast cancer’ returned over 150 papers in English in the last year alone, which represent over 20 new molecules, all of which associate with disease-free or overall survival, reflecting the bias towards identification of ‘novel’ markers and the lack of investment (both financial and intellectual) in translation of these findings into clinical settings. In spite of massive investment in biomarker discovery, only one new biomarker has been translated into UK clinical practice in the past 25 years. The aim of this review, therefore, is not to provide a comprehensive overview of candidate predictive biomarkers, but to explore the driving forces behind biomarker application and discovery, the techniques which are being used in biomarker validation, and the considerations which need to be taken into account in order for new biomarkers to be a clinical success. We will illustrate the ‘hottest’ biomarker research topics along the way, which, with appropriate and directed research effort, we believe should form part of clinical diagnostic decision-making over the next 3–5 years.
Drug-driven tissue biomarker discovery
The success of the archetypal predictive biomarkers ER and Her2 in breast cancer has been driven by the availability of pharmacological agents and at least a rudimentary knowledge of how they interact with their targets; these biomarkers are the subject of another review in this issue of Histopathology. There has been limited evidence over the past 20 years that similar mechanistically targeted biomarker discovery approaches are widely applied. However, the same rational pharmacotherapy paradigm is now gradually being applied to both existing chemotherapeutic agents, as knowledge of their mechanism of action emerges, and newer targeted ‘biological agents’, as molecular biology is factored into early phase trials during the development of novel pharmacological agents.
Existing agents—teaching old dogs new tricks
Current thinking within clinical oncology is that, although biologically targeted agents offer significant hope for the future, at least for the next 5–10 years these agents are likely to be either targeted on specific subgroups of patients, as with trastuzumab, and/or given in combination with conventional chemotherapeutic agents, as with Avastin. There remains, therefore, a significant debate as to the optimum scheduling and combination of current chemotherapeutic and endocrine agents,4 which is fuelled by two distinct thought processes. First, it is clear that many patients who receive systemic therapy for breast cancer do not, in fact, require such treatment.5 It is an inability to identify such patients prior to treatment, rather than an expectation that all patients derive benefit, which drives the treatment of significant numbers of breast cancer patients with often aggressive chemotherapy. The identification of novel prognostic markers, which is discussed elsewhere, is key to the solution of this dilemma. Second, it is becoming increasingly apparent that the wide biological diversity of breast cancer6 underpins not only potential differences in the natural history of the disease, but also differences in response to current adjuvant chemotherapy regimens. Thus it is entirely possible, and indeed highly likely, that some tumours which are insensitive to taxanes may in fact respond to anthracycline-based chemotherapies, and vice versa. Thus, the search for an optimal chemotherapy schedule and combination must in future include the expectation that this optimum will be different for different patient subgroups and that these groupings will be defined by the presence within breast cancers of molecular targets for specific chemotherapeutic agents. As a result, a key objective of research into predictive biomarkers in the near future will be the identification of markers that identify the optimal chemotherapy regimen (e.g. accelerated or not, taxane or not, etc.). This research priority was in fact the second most important goal, second only to identification of prognostic markers to identify patients not requiring treatment, identified in a recent multinational focus group document aimed at identification of the top 10 most important research questions in breast cancer (http://www.toptenresearch.org/).
This process has, in the past, been hampered by the lack of a cohesive approach to the development of predictive biomarkers3,7 and of appropriate, clinical trials-based sample banks in which such biomarkers must be validated. Recently, however, significantly more emphasis has been placed upon the collection of biological samples in the context of clinical trials. Many trials, including the UK Taxotere as Adjuvant Chemotherapy Trial (TACT), TACT2, Paclitaxel, Anthracycline, Gemcitabine & Cyclophosphamide (TANGO), Tamoxifen and Exemestane Adjuvant Multicentre (TEAM), Selective Use of Postoperative Radiotherapy after Mastectomy (SUPREMO) and Breast International Group (BIG)-1-98 trials, have included prospective tissue collection, whereas others, including Arimidex, Tamoxifen Alone or in Combination (ATAC) and National Epirubicin Adjuvant Trial (NEAT), have made significant efforts to access tumour samples retrospectively. These biorepositories represent invaluable resources for the future validation of predictive biomarker profiles. Research is ongoing in many of these studies, which either have reported or will report in the next 1–3 years.
Two key areas of biomarker research are amongst the first to be addressed using these repositories. First, is there an optimal signature for discriminating patients who may benefit from frontline treatment with aromatase inhibitors (AI) from those who may derive benefit from tamoxifen followed by an AI? Second, are there patients for whom either taxanes or anthracyclines may be omitted without reducing their benefit from adjuvant chemotherapy.
Tamoxifen resistance has for many years been recognized to be associated with gene amplification or overexpression of the HER2 oncogene or lack of expression of the progesterone receptor in ER+ breast tumours.8–11 Research in progress in the ATAC, BIG-I-98 and TEAM trials is addressing this question (see below). Coupled with further biomarker research,12–16 it is likely that we will in the coming years finally be in a position to identify de novo endocrine-resistant cancers, and possibly to identify the optimal endocrine treatment regimen for distinct biological subgroups.
For chemotherapy, key research in the TACT, TANGO and NEAT trials, within a UK-wide collaborative network, should provide insights into biomarkers relating to benefit from taxanes and anthracyclines. Although the first anthracyline was discovered in the 1950s, its mechanism of action was not elucidated until 1984.17 Amongst other activities, anthracyclines inhibit the topoisomerase IIα (TIIα) enzyme, which results in double-stranded DNA breaks and initiates apoptosis. Since anthracyclines have a low therapeutic index, and patients frequently experience treatment-related cardiomyopathy and secondary leukaemias, biomarkers which select for anthracycline responsiveness would be of benefit in reducing morbidity in patients receiving adjuvant chemotherapy. Several retrospective studies have separately analysed, on the one hand, TIIα or HER2 expression, gene amplification or deletion, and, on the other, more conventional markers of response such as proliferation.18–24 These studies suggest that TIIα aberrations, particularly amplifications, may predict for improved outcome on anthracyline-based regimens relative to tumours with a normal TIIα status. However, the water is muddied by the fact that the TIIα gene is located close to that for HER2 on chromosome 17q12-21. Amplification of HER2 is associated with TIIα gene aberrations,25 making it difficult to be clear which gene is most closely associated with anthracycline sensitivity. Previous studies have suggested that both HER2 and TIIα are potential predictive markers of anthracycline sensitivity.19,20,23,24,26,27 Our most recent data, from the BR9601 trial,28 suggest that HER1 and HER3 provide additional information when analysed with HER2. Therefore, although TIIα remains a promising candidate biomarker, the possibility remains that HER2 with HER1/HER3, or even more complex molecular panels, may be of more practical value in determining eligibility for anthracyclines (see below). Future data from the NEAT study and an ongoing meta-analysis may provide further insights in this area.
Novel agents, novel approaches—the rationale for rational biomarker-driven trials
Even although HER2 and ER have only marginally improved patient selection, by excluding non-responders, lessons have been learned, particularly by the pharmaceutical industry, about the need to decipher the biological basis of response or risk seeing agents prematurely shelved. The most notable example of this is the response of non-small cell lung cancer (NSCLC) to gefitinib (Iressa), an epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor, which has shown promising results in preclinical studies, but far more modest responses in clinical trials (10–19%).29,30 Discovery of mutations within the tyrosine kinase domain of the EGFR in NSCLC patients showing dramatic responses to gefitinib caused much excitement,31 but the initial enthusiasm was tempered when it became clear that these mutations occur at low frequency, and rarely in smokers.32 Because immunohistochemical detection of EGFR was not associated with gefitinib response,33,34 it became apparent that methods of patient selection needed to be refined and standardized, in the same way that issues surrounding the detection and criteria for treatment in Her2-overexpressing tumours have evolved since early trials.35 Debate continues as to the optimal approach for selecting tumours likely to benefit from HER-targeted therapies; high EGFR gene copy number detected by fluorescence in situ hybridization, gene mutation and protein overexpression are all candidates in NSCLC,36 suggesting that similar approaches could be explored when using EGFR-targeting agents, such as gefitinib and laptinib in the treatment of breast cancer. Although the expression levels of EGFR in breast cancer have been investigated extensively, figures for EGFR ‘positivity’ have been hampered by the methodological differences between studies, including the use of different assays targeted against DNA, RNA, protein and activated phosphorylated proteins.37 However, it appears that EGFR is positive in approximately 45% (range 14–91%) of all breast cancers,38,39 and expression is associated with high grade, hormone receptor-negative tumours and increased in vivo proliferation.40,41 Expression of EGFR is also enriched in basal-like breast cancers and breast tumours from BRCA1 mutation carriers, with up to 70% of basal-like breast cancers expressing EGFR.42,43 Furthermore, EGFR expression may be seen by immunohistochemistry in metaplastic breast carcinomas and in the stromal compartment of phyllodes tumours (80% and 19%, respectively).44,45 These data suggest that there are a number of patient groups who could benefit from EGFR-targeted therapy. However, activating mutations within the kinase domain of EGFR are not seen, and gene amplifications are detected in only up to 6% of invasive breast cancers, although this increases to 25% of those with a basal-like phenotype.46 Nevertheless, and especially due to the close association with hormone receptor status and the potential targeting of hormone-refractory breast cancer, EGFR inhibitors have been used in a number of Phase I and Phase II clinical studies.47 Responses to single-agent therapy with gefitinib in hormone-resistant breast cancer has, however, been disappointing, with low clinical benefit rates terminating one trial early, and low numbers in another making interpretation of response rates difficult. Likewise, and in spite of early data suggesting that combined therapy with gefitinib and the aromatase inhibitor anastrozole could induce disease stabilization in 20% of patients, recent reports in the neoadjuvant setting have shown that gefitinib has neither a biological nor clinical effect when added to anastrozole.48 The majority of these trials, however, did not select their patients prospectively on the basis of EGFR expression status and were small, and therefore any subgroup benefit may not have been detected. We have previously shown that other members of the HER2 gene family may influence response to therapy and clinical outcome,9,11,40,49 and results from single-agent studies with lapatinib, a small-molecule inhibitor of EGFR and Her2 tyrosine kinases and dimerization inhibitor, in which patients are treated on the basis of biomarker expression, are promising.50 Given the increasing knowledge of Her family biology, similar hypothesis-driven enrichment of patient populations to be treated with EGFR inhibitors should also be carried out, and detection and interpretation of EGFR standardized.
Looking beyond the target: pathway-driven tissue biomarker discovery
It is apparent from the examples cited above that it is naive to think we will be able to stratify accurately all patients into responders and non-responders on the basis of a single biomarker approach, even if the biomarker is known to be the drug target. Existing biomarkers are surrogates for complex tumour biology in which multiple, non-linear pathways are (abnormally) activated, within heterogeneous tissues. This means that resistance mechanisms may occur at the level of the target itself (such as presence or absence of activating mutations in EGFR), downstream of the target [such as loss of phosphatase and tensin homologue (PTEN) in Herceptin resistance], or secondary to the heterogeneity of the tissue itself (i.e. not all cells within the tumour express the target), so that the agent selects for resistant clones. In addition, other mechanisms may contribute to the effectiveness of the agent that are not directly related to the biology of the molecule that they inhibit, such as the potential contribution of inflammatory responses to Herceptin efficacy.
We have already discussed the role that specific EGFR mutations may have in gefitinib sensitivity in NSCLC. It has recently been shown, using a screening approach in vitro, that other existing RTK inhibitors may be used to successfully target imatinib (Glivec)-resistant tumours, paving the way for mutation-specific treatment,51 which would necessitate mutation-specific biomarker analysis. However, since most of these mutations are probably rare events in breast cancer, we and others are taking two approaches to predictive biomarker validation for new agents in clinical trials. The first is the ‘bottom-up’ approach, in which the targeted pathway is reconstructed from the literature, and multiple candidate biomarkers tested in neoadjuvant and adjuvant clinical trials with supporting evidence from in vitro and in vivo experiments. The second is the ‘top-down’ approach, where predictive signatures are produced from statistical analysis of gene expression array data, which is the subject of review elsewhere in this issue of Histopathology.
The precedent for the ‘bottom-up’ approach has come from the investigation of the role of PTEN in Herceptin resistance.52,53 PTEN (phosphatase and tensin homologue deleted on chromosome 10) is a tumour suppressor protein which is frequently down-regulated, although not usually mutated, in breast cancer.54 PTEN opposes the action of phosphatidylinositol 3-kinase (PI3K) and downstream Akt to attenuate signalling through this critical growth-promoting pathway, which is known to be inhibited in response to Herceptin and contribute to its antitumour action.55 Based on this knowledge of the underlying pathways, Nagata and colleagues have shown that loss of PTEN contributes to Herceptin resistance through activation of PI3K signalling, and this has since been substantiated in other studies.52,53 This raises the possibility that PTEN, and possibly AKT/PI3Kinase signalling, would be an excellent candidate biomarker for validation in the clinical trial setting, although this has yet to be tested in the context of Phase III trials. AKT amplifications are rare in breast cancer, although a number of PI3kinase mutations have been described.56 Therefore molecular profiling of the HER2/PTEN/PI3kinase pathway should be prioritized for future biomarker profiling in Herceptin resistance.
More recently, we have taken a similar approach to predictive biomarker analysis of a new agent being used in the treatment of breast cancer, RAD001, a mammalian target of rapamycin (mTOR) antagonist.57–62 mTOR is a 289-kDa serine/threonine protein kinase that belongs to the phosphatidylinsotide-3OH kinase-related kinase family. It exists in two distinct complexes within the mammalian cell, one containing mTOR, GβL and raptor (known as mTORC1) and the other containing mTOR, GβL and rictor (mTORC2). These two mTOR complexes have distinct roles in intracellular signalling, with both direct and indirect effects on Akt activity. Both inhibition and activation of Akt are seen, depending on which complex is inhibited. By reconstructing the pathway and measuring activated (phosphorylated) protein expression levels of candidate biomarkers in biopsy specimens from patients treated with RAD001 for 2 weeks prior to surgery, we have shown that pretreatment phospho-AKT (serine 473), but not total or phospho-mTOR, correlates with decreases in proliferation with treatment as measured by Ki67 proliferation indices. This counterintuitive result highlights how rational, pathway-driven biology can reveal new targets for predictive biomarker analysis beyond measurement of the drug target alone, and also how novel agents such as mTOR inhibitors could be matched, based on molecular signatures, to appropriate pharmacological partners for future polychemotherapy trials. Thus, mTOR inhibitors may have a role in breast cancer populations whose resistance to current agents, e.g. Herceptin (see above), is mediated via the PIK/AKT/mTOR signalling pathway.
Beyond discovery—biostatistics in biomarker validation
Throughout this review we have highlighted contradictory results and some failings in the journey of new biomarkers from ‘bench to bedside’. We believe that one of the key aspects of successful and speedy biomarker validation is to ensure the appropriate statistical test and appropriate statistical power are available, prior to reporting of major clinical datasets, in order to ensure robust validation and reduce misleading results. Although small sample sizes or retrospective tumour studies are invaluable for hypothesis generation, it is imperative that for biomarkers to be appropriately validated, a priori hypotheses are correctly defined and powered, prior to commencing analysis.
Early reporting or reporting of inadequately powered studies, even within the context of large Phase III clinical trials, may not inform and may, in fact, lead to inappropriate conclusions being drawn. This is particularly critical for studies reporting negative findings. As with measures of outcome, where underpowered studies in which no evidence of benefit is observed are recognized as providing relatively little value, we need to recognize that similarly rigorous interpretation must be applied to biomarker analyses.
Two key examples illustrate this dilemma. First, the potential predictive value of HER2 gene amplification or overexpression with respect to tamoxifen treatment and for HER2 to act as a predictive biomarker for response to AIs. There is extensive in vitro and in vivo evidence to support the role of HER2 as a mediator of resistance to tamoxifen,9–11 and neoadjuvant studies suggest that in fact HER2+ cancers may derive greater benefit from AIs than from tamoxifen.63 However, a recent analysis of the BIG-I-98 trial, while suggesting that HER2 was indeed predictive of early relapse, failed to substantiate the interaction between HER2 and treatment with AIs.64 Critically, the authors correctly note that, despite the large sample size of this study (7707 samples), the low rate of HER2 positivity (about 5%) and the low event rate to date (3.7%) in this low-risk ER+ population provided insufficient power to address the question of interactions between HER2 status and treatment modality. Therefore, in this respect, this study is unable to provide a definitive answer and further analysis or meta-analysis of the major adjuvant AI trials is awaited.
This example underlines the fact that biomarker analysis is a complex field and one where simplistic answers are hard to come by. For appropriate studies to be designed and correctly interpreted, statistical approaches no less rigorous, and potentially more complex, than those employed for the analysis of primary end-points of response must be employed. Treatment interactions with biomarkers, where treatment response may differ between biomarker-positive and -negative cases, require multivariate analyses, which increase the required sample size. Agreed analytical procedures, which address the potential for interactions between biomarkers and treatments and the potential for such interactions to change over time, will need to be determined for future biomarker analyses. Finally, caution must be applied to inappropriately powered studies and, where biomarkers may select small subgroups of cancers, meta-analysis of compatible trials may be required in order to provide definitive answers to some questions.
In this review we have highlighted some of the problems to be overcome before predictive biomarkers can be considered for clinical practice. Even biomarkers such as TIIα, which have been recognized as potential markers for several years and tested within large Phase II trials, remain controversial. We have shown how the ‘biomarker journey’, from bench to bedside, should be underpinned by sound biological hypotheses. We have previously discussed how the ‘evolutionary model’ of predictive biomarker development will need to change from ‘survival of the fittest’ (e.g. value based on publication frequency) to a structured, multi-tiered approach including the use of preclinical models, neoadjuvant clinical models, properly designed retrospective studies and validation within prospective clinical trials.3 Several groups have highlighted the need for early development of robust and accurate diagnostic assays for potential biomarkers.65,66 Most importantly, a structured and phased approach to biomarker validation has been suggested, which mirrors that available for the evaluation of novel pharmacological agents.65 The ‘translation’ of such proposals into a practical structure for biomarker validation would represent one of the most significant drivers for the acceleration of future biomarker discovery.
None to declare.