Molecular Profiling Assays in Breast Cancer: Are We Ready for Prime Time?
Molecular Profiling Assays in Breast Cancer: Are We Ready for Prime Time?
Breast cancer is a heterogeneous disease with diverse morphologies, molecular characteristics, and clinical behavior. The advances in molecular profiling technologies have changed our understanding of breast cancer and led to the identification of prognostic/predictive gene signatures. Despite the huge quantity of information gleaned from these profiling technologies and the increasing number of gene signatures, their incorporation into clinical decision making is a slow process and is limited in various aspects. The 70-gene assay (MammaPrint, Agendia, Netherlands) and the 21-gene assay (Oncotype DX, Genomic Health, USA) are the most widely used breast cancer multigene classifier assays. A 50-gene assay (PAM50, NanoString, USA) has shown promise but needs further independent validation. In this review, we will present the current data on commercially available molecular profiling assays in breast cancer and discuss the challenges surrounding their incorporation into routine clinical practice as prognostic and predictive tools.
Breast cancer is a heterogeneous disease. Several histological subtypes are recognized but these are of limited utility, since the vast majority of the breast cancers (65% to 80%) belong to a single subtype, invasive ductal carcinoma of no special type. Thus, the prognosis of breast cancer is mostly determined using three criteria: lymph node status, tumor size, and histological grade. The Nottingham Grading System, the modified version of the Scarff-Bloom-Richardson grading system, is of particular importance in small node-negative tumors. It is also an independent prognostic indicator in estrogen receptor (ER)-positive breast cancer, although it may not be of significant utility in ER-negative tumors, since most of these are high histologic grade. ER, although a weak prognostic marker, is a strong predictor of response to antiestrogen therapy. In addition to ER, expression of progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) are routinely used in clinical practice. HER2 testing is also critical to the selection of patients most likely to obtain benefit from therapy with the anti-HER2 antibody, trastuzumab (Herceptin). However, almost half of patients with HER2-positive disease are resistant to trastuzumab.
To improve treatment decision making for the clinician, multiparametric tools, such as TNM staging, the Nottingham prognostic index (NPI), and Adjuvant! Online (http://www.adjuvantonline.com) have been introduced. TNM staging includes tumor size, nodal status, and metastases. The NPI predicts survival in patients with breast cancer by combining tumor size, lymph node stage, and histological grade. Adjuvant! Online estimates the probability of survival and benefit from specific therapies in patients with early breast cancer. The clinical parameters included in this tool are age, menopausal status, comorbidities, ER status, tumor grade, tumor size, and the number of involved axillary nodes. Despite the utility of all these tools, they are limited in their ability to help with clinical decision making and selection of treatment options because of the heterogeneous nature of breast cancer and the relative inefficiencies of available therapies. A considerable number of patients still develop tumor recurrence and die from breast cancer in spite of having received adjuvant anti-estrogen therapy, trastuzumab, or chemotherapeutic agents. Based on current treatment algorithms, it is estimated that up to 60% of patients will only experience toxicities and will derive little or no benefit from therapy.[6,7] There is a great need for novel predictive assays that can help identify patients likely to respond to the often toxic and very expensive therapies, so patients can avoid experiencing only toxicities and no benefits from these treatments.
Key Steps for Incorporation of Molecular Profiling Assays Into Routine Clinical Practice
Molecular profiling is an emerging concept in clinical decision making that involves arrangement of biological specimens such as tumors or other tissues into groups based on multiple changes at the genomic and transcriptomic levels. In the last decade, molecular profiling technologies have advanced our knowledge of breast cancer biology. New tests utilize gene expression assays—either microarray or real-time quantitative reverse transcription polymerase chain reaction [qRT-PCR]. A number of prognostic and predictive multigene profiling assays (mainly considered as multigene breast cancer classifiers) have had a substantial impact on clinical oncology practice and have helped make individualized management of patients a possibility. While the first-generation prognostic multigene classifiers, such as the MammaPrint assay and the Oncotype DX breast cancer assay, are the closest to clinical practice, the second-generation prognostic multigene assays, which include breast cancer microenvironment or host immune response, have not been commercialized and need further external validation studies to determine their clinical utility. Despite several studies, the translation of predictive multigene classifiers into the clinic is even more challenging than that of prognostic multigene classifiers. Most of the predictive assays are derived mainly from cell lines. Also, using the microarray as the assay platform is not as quantitative as using a qRT-PCR assay. Therefore, subtle changes in gene expression may not be reflected in microarray-based assays, although these subtle differences may be sufficient to cause resistance to chemotherapeutics. Furthermore, resistance may occur due to low penetrance of the drug being administered and may be unrelated to tumor tissue.
To incorporate prognostic and/or predictive multigene classifiers into clinical practice, the following key criteria need to be fulfilled:
First, the platform on which the classifier is based should be suitable for broad clinical application and ensure that the classifier is stable under a variety of operating conditions. If not, the classifier needs to be translated to a clinically applicable platform. The assay protocols should be standardized to achieve satisfactory interlaboratory and intralaboratory reproducibility, thereby establishing analytic validity. Assay standardization includes preanalytic parameters, such as sample storage and preparation, and analytic performance parameters, such as the sensitivity and specificity of the system as well as assay reproducibility. The Clinical Laboratory Improvement Amendments of 1988 (CLIA) require laboratories to independently establish analytic validity and to improve assay standardization.
Second, it is critical to classify studies as developmental or validation studies in order to increase the clinical validity of the classifier. For assays that purport to elucidate predictive significance, this strategy needs to be applied to determine the clinical utility of the classifier.[10,11] Developmental studies need to include internal clinical validation; this can be accomplished either by splitting the study population into two populations (the training model and the testing model) or by cross-validation based on repeated model development and testing on random data partitions. These approaches will increase the accuracy of the classifier, which in turn makes its further development possible. Independent validation studies are critical to further evaluate the predictive accuracy and usefulness of the classifier in clinical practice; such studies should be prospectively designed, and should verify both clinical validity and clinical utility.
Third, does the classifier only assess prognosis, or does it help with selection of a certain type of therapy? What is the therapeutic relevance of the classifier? Prognostic multigene classifiers assess the likelihood of disease recurrence, whereas predictive multigene classifiers evaluate the potential benefit from certain types of chemotherapy or anti-estrogen therapy. However, a prognostic classifier may also exhibit predictive significance. If a classifier is a predictive classifier, the bar for utility is often quite low. For example, approximately half of patients with HER2 positivity respond to trastuzumab. However, if the assay assesses low likelihood for recurrence or metastases (a prognostic assay), patients classified as low risk need to have such a low risk that they can be spared from adjuvant therapy without affecting their long-term prognosis.
Fourth, the incorporation of the classifier into the clinic might be more beneficial if it outperforms or adds predictive power to existing prognostic methods; this would help justify the money and time invested in its external validation in a trial of a much larger scale. In other words, it is important to determine cost-effectiveness.
In this review, we discuss the advantages and limitations of the commercially available first-generation prognostic multigene classifiers based on the key criteria described above.
First-Generation Prognostic Multigene Classifiers in Breast Cancer
The “intrinsic” classification was the first assay to use modern molecular tools to classify breast cancers. The 70-gene signature of van ’t Veer et al (MammaPrint) and the 21-gene assay (Oncotype DX) were the first two breast cancer multigene classifiers to become commercially available.
Both of these have been tested in more than one validation cohort and are being tested for further clinical utility in large prospective trials in Europe (MINDACT [Microarray In node Negative Disease may Avoid ChemoTherapy]; MammaPrint assay) and in the United States (TAILORx [Trial Assigning IndividuaLized Options for treatment; Oncotype DX assay). Another assay in advanced stages of development is a 50-gene assay (PAM50), the clinically applicable platform of intrinsic subtype classification (Table). A number of other assays that are in various stages of development and validation are not discussed in this review.
The Intrinsic Subtypes of Breast Carcinoma
Using DNA microarrays, Perou et al[16,17] and Sorlie et al classified breast carcinomas into five molecular subtypes: ER-positive (luminal A), ER-positive (luminal B), ER-negative/HER2-enriched, basal-type, and normal-like subtypes. The luminal A subtype displays high expression of ER, GATA binding protein 3 (GATA3), X-box binding protein trefoil factor 3, FOXA1, and LIV-1. The luminal B subtype, on the other hand, shows moderate expression of the genes expressed by the breast luminal cells and frequently higher proliferation and lower PR levels. These subtypes are associated with differences in both disease outcome and therapeutic response.[15,18]
The major disadvantage of the original intrinsic classification is that it was based on analysis of microarrays. The data are labile in that the addition of a single new case has the potential to change the classification rules. In order to achieve stability and permit analysis of new cases, “single sample predictors” (SSPs) were introduced. These SSPs have been modified over the different publications[15,19-22]; the agreements between these definitions have been only modest (kappa value, 0.4–0.5). The classification has also suffered from lack of validation by independent investigators. In addition, the intrinsic classification raises several important issues.
First, does the “normal” subtype of breast cancer exist? It is becoming increasingly accepted that this was an artifact caused by a disproportionately high content of normal breast and stromal cells in the frozen samples used for microarray analysis.[15,21,23]
Second, and perhaps the more important question, what is the relevance of the “HER2-enriched” category? A fair number of cases included in the HER2-enriched category do not have increased HER2 mRNAs. In addition, it has been shown that a large number of clinically positive cases (by immunohistochemistry [IHC] and/or fluorescence in situ hybridization [FISH]) do not belong in the HER2-enriched category.[15,23-25] If one accepts the notion that the HER2-enriched category has no therapeutic relevance, then the existence of this category could be confusing for both patients and their treating physicians (see additional data below).
Third, can ER-positive tumors be subdivided into two distinct categories, luminal A and luminal B, or is this division artificial? It appears that this division is largely based on proliferation-related genes.[15,21,23] Other studies have suggested a continuous spectrum of proliferation in ER-positive tumors.[23,26]
The intrinsic classification is a major milestone in the molecular classification of breast cancer. However, it should be regarded as a “work in progress,” particularly since novel subtypes, such as molecular apocrine, interferon-rich, and claudin-low, are still being discovered.
The PAM50 assay is a clinically applicable updated version of the intrinsic subtyping classification based on qRT-PCR and NanoString technology. A 50-gene subtype predictor assay has been developed using a training set representing each subtype. This assay measures a risk of relapse (ROR) that is prognostic of relapse-free survival for patients who have node-negative tumors and who have not received adjuvant systemic therapy. Using an independent test cohort with ER-positive breast cancer treated with adjuvant tamoxifen, the ROR score combined with tumor size performed better than the other standard clinicopathological variables. The PAM50 assay is being further developed using the NanoString nCounter platform (NanoString Technologies, Seattle). Using this platform, samples from the ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial were analyzed to compare PAM50 ROR scores with results from the Oncotype DX assay and IHC4, a four-panel set of IHC markers (ER, PR, HER2, Ki67) used to provide information on distant recurrence-free survival. Dowsett et al reported that PAM50 assigned more women to the high-risk group and fewer women to the intermediate-risk group than did Oncotype DX. The PAM50 ROR score also significantly predicted 10-year distant recurrence-free survival and added prognostic information about distant recurrence to standard pathological variables in all patients. However, the current ATAC trial analysis of PAM50 does not validate the test’s ability to predict whether patients will derive benefit from specific chemotherapies. Results of the comparative analysis with IHC4 have not been presented. Cheang et al emphasized that the PAM50 HER2-enriched subtype gets more treatment benefit from neoadjuvant anthracyclines/taxane and trastuzumab/taxane regimens. However, this is somewhat to be expected given the low response rates in ER-positive patients. Overall, the results with PAM50 are promising, and upon validation, the test could become part of clinical decision making in the near future.