Cluster Analysis and Subtypes

Molecular studies have demonstrated the great heterogeneity of breast cancer. One of the first applications of microarray-based gene-expression profiling analysis in breast cancer was the landmark work by Perou and colleagues and Sorlie and colleagues. These studies were the first to demonstrate that estrogen receptor (ER)-positive and ER-negative breast cancers are biologically distinct diseases with respect to their molecular analysis. In addition, cluster analysis of genes revealed that there are at least 4 molecular subtypes of breast cancer, including luminal, human epidermal growth factor receptor 2 (HER2)-enriched, basal-like, and normal breast–like. Luminal A (∼40% of all breast cancers) and luminal B (∼20% of all breast cancers) are the most common subtypes, and are characterized by the expression of ER, progesterone receptor (PR), and other genes associated with ER activation. Although luminal A and luminal B constitute most ER-positive breast cancers, there are important molecular distinctions between the two. Luminal A tumors typically have high expression of ER-related genes, low expression of HER2 genes, and low expression of proliferation-related genes. In comparison with luminal A tumors, luminal B tumors have lower expression of ER-related genes, variable expression of HER2 genes, and higher expression of the proliferation-related genes. The HER2-enriched subtype (∼10%–15% of all breast cancers) is characterized by high expression of the HER2 and proliferation-related genes, and low expression of the luminal and basal-like genes. Importantly not all HER2-enriched subtypes translate to clinically HER2-positive breast cancer, and vice versa. Thus, not all HER2 mutations result in HER2 amplification and protein overexpression. Furthermore, approximately 50% of clinically HER2-positive breast cancers are not HER-enriched at a molecular level, but are characterized as HER2-positive luminal subtypes. The basal-like subtype (∼15%–20% of all breast cancers) is characterized by low expression of the luminal and HER2 genes and high expression of the proliferation cluster of genes. Although most basal-like breast cancers are triple-negative (ER-negative, PR-negative, HER2-negative), not all triple-negative breast cancers are basal-like. Finally, the normal-like subtype is characterized by gene expression similar to that of normal breast tissue. It remains unclear whether this represents a separate subtype with clinical significance or a technical artifact of the molecular analysis.

Subtype Prognostication and Treatment Recommendations

It must be borne in mind that the molecular analysis of intrinsic subtypes of breast cancer identifies relevant biology. These studies were not designed for prognostication. However, subtype analysis does correlate with prognosis in multiple large data sets. Overall, patients with luminal A breast cancer have the best prognosis, followed by patients with luminal B breast cancer. Patients with either HER2-enriched or basal-like subtypes have the worst overall survival. However, more recent advances in targeted therapies (eg, trastuzumab) hold significant promise in the ability to alter that natural history. A consensus conference was held to discuss the role of molecular subtype analysis in clinical decision making. The panel concluded that ER/PR/HER2 measurements should not be used as surrogates for assigning patients into molecular groups, and that molecular subtype analysis was insufficient at present to incorporate into the decision-making algorithm for treatment recommendations.

Cluster Analysis and Subtypes

Molecular studies have demonstrated the great heterogeneity of breast cancer. One of the first applications of microarray-based gene-expression profiling analysis in breast cancer was the landmark work by Perou and colleagues and Sorlie and colleagues. These studies were the first to demonstrate that estrogen receptor (ER)-positive and ER-negative breast cancers are biologically distinct diseases with respect to their molecular analysis. In addition, cluster analysis of genes revealed that there are at least 4 molecular subtypes of breast cancer, including luminal, human epidermal growth factor receptor 2 (HER2)-enriched, basal-like, and normal breast–like. Luminal A (∼40% of all breast cancers) and luminal B (∼20% of all breast cancers) are the most common subtypes, and are characterized by the expression of ER, progesterone receptor (PR), and other genes associated with ER activation. Although luminal A and luminal B constitute most ER-positive breast cancers, there are important molecular distinctions between the two. Luminal A tumors typically have high expression of ER-related genes, low expression of HER2 genes, and low expression of proliferation-related genes. In comparison with luminal A tumors, luminal B tumors have lower expression of ER-related genes, variable expression of HER2 genes, and higher expression of the proliferation-related genes. The HER2-enriched subtype (∼10%–15% of all breast cancers) is characterized by high expression of the HER2 and proliferation-related genes, and low expression of the luminal and basal-like genes. Importantly not all HER2-enriched subtypes translate to clinically HER2-positive breast cancer, and vice versa. Thus, not all HER2 mutations result in HER2 amplification and protein overexpression. Furthermore, approximately 50% of clinically HER2-positive breast cancers are not HER-enriched at a molecular level, but are characterized as HER2-positive luminal subtypes. The basal-like subtype (∼15%–20% of all breast cancers) is characterized by low expression of the luminal and HER2 genes and high expression of the proliferation cluster of genes. Although most basal-like breast cancers are triple-negative (ER-negative, PR-negative, HER2-negative), not all triple-negative breast cancers are basal-like. Finally, the normal-like subtype is characterized by gene expression similar to that of normal breast tissue. It remains unclear whether this represents a separate subtype with clinical significance or a technical artifact of the molecular analysis.

Subtype Prognostication and Treatment Recommendations

It must be borne in mind that the molecular analysis of intrinsic subtypes of breast cancer identifies relevant biology. These studies were not designed for prognostication. However, subtype analysis does correlate with prognosis in multiple large data sets. Overall, patients with luminal A breast cancer have the best prognosis, followed by patients with luminal B breast cancer. Patients with either HER2-enriched or basal-like subtypes have the worst overall survival. However, more recent advances in targeted therapies (eg, trastuzumab) hold significant promise in the ability to alter that natural history. A consensus conference was held to discuss the role of molecular subtype analysis in clinical decision making. The panel concluded that ER/PR/HER2 measurements should not be used as surrogates for assigning patients into molecular groups, and that molecular subtype analysis was insufficient at present to incorporate into the decision-making algorithm for treatment recommendations.

Overview

Concurrent with the evolution of molecular subtype classification of breast cancers, several researchers and industry sponsors have developed multiple gene prognostic signatures, several of which have been validated and are in clinical use. The 3 most commonly used molecular prognostic profiles are the Recurrence Score (RS), derived from Oncotype Dx, the Amsterdam 70-gene signature (Mammaprint), and the Risk of Recurrence Score (ROR), derived from PAM50. The RS was validated in an independent data set from samples collected from node-negative, ER-positive patients treated with tamoxifen in the large multicenter National Surgical Adjuvant Breast and Bowel Project (NSABP) B-14 trial. A recurrence score from zero to 100 is reported, with 3 categories of risk: low (scored 0–18), medium (scored 19–30), and high (scored 31–100). The test also reports ER, PR, and HER2 status. The 70-gene signature was developed using a supervised DNA microarray analysis of gene-expression arrays on frozen tissue from 98 primary breast tumors, followed by validation in multiple studies. It is approved for use in both ER-positive and ER-negative tumors, and reports either a low-risk or high-risk result. The PAM50 ROR is a 50-gene test that characterizes an individual tumor by intrinsic subtype and generates a ROR; it has also been validated in several studies. Although several other genomic signatures are available, the remainder of this review refers only to published data related to these 3 validated signatures. Furthermore, because ER/PR-negative tumors of all sizes are candidates for adjuvant chemotherapy, with or without the addition of trastuzumab depending on HER2 expression, this article focuses on the evidence surrounding the use of molecular prognostic tools for patients with ductal carcinoma in situ (DCIS) and ER-positive invasive breast cancer.

Ductal Carcinoma in Situ

Breast-conserving therapy, which includes wide local excision of the tumor followed by irradiation, has become a standard treatment option for women with DCIS. Local recurrence has been shown to be affected by several patient and tumor characteristics, including patient’s age, extent of disease, nuclear grade, margin status, presence of comedonecrosis, and utilization of adjuvant radiation. The National Comprehensive Cancer Network (NCCN) included excision alone as an acceptable treatment option for patients with DCIS in the 2008 practice guidelines, but they did not define which subgroup of patients for which excision alone is appropriate. In 2003, Silverstein updated his Van Nuys Prognostic Index, which describes the use of nuclear grade, necrosis, size, margin width, and patient’s age to predict recurrence following excision of DCIS. Excision alone is recommended for those with scores of 4 to 6; excision plus adjuvant radiation therapy is recommended for those with scores of 7 to 9; and mastectomy is recommended for those with scores of 10 to 12. This scoring translated to a less than 20% local recurrence rate at 12 years when these criteria were followed. However, the primary limitation for the Van Nuys Prognostic Index is the lack of an ability to account for the wide heterogeneity of DCIS.

Molecular and biological markers that provide prognostic and predictive information hold the most promise for tailoring therapy on an individual level. Expression of p16, cyclooxygenase-2, and Ki-67, which indicate an abrogated response to cellular stress, have been shown to delineate which DCIS lesions are more likely to confer a high risk for recurrence. The 21-gene RS for ER-positive invasive breast cancer was used as a benchmark for the development of a 12-gene subset that has been used to develop and validate a DCIS score that divides patients into low risk, intermediate risk, and high risk for 10-year local in-breast recurrence. The DCIS score was prospectively evaluated using archived tumor samples from the Eastern Cooperative Oncology Group E5194 study. This prospective, nonrandomized study investigated the risk of local recurrence in 670 patients with DCIS following wide local excision alone. Patients were required to have negative margin widths of 3 mm or more and were divided into 2 treatment arms consisting of grade 1 or 2, size 2.5 cm or smaller, or grade 3, size 1.0 cm or smaller. Radiation was not allowed, but approximately 30% of patients did receive optional adjuvant tamoxifen.

The relationship between the DCIS score risk group and 10-year risk of an ipsilateral breast event (IBE), whether in situ or invasive, was highly statistically significant ( P = .006 for any IBE and P = .003 for an invasive IBE). Multivariable modeling of risk for IBE, both excluding and including the DCIS score, were also performed. When the DCIS score was excluded, tumor size and postmenopausal status were the only factors significantly associated with risk for IBE (hazard ratio [HR] 1.54, P = .006 and HR 0.49, P = .02, respectively). However, when the DCIS score was included in the model, the 12-gene score was statistically significant (HR 2.37, P = .02), in addition to tumor size and menopausal status. Thus, patients who have undergone wide local excision for DCIS who have a low-risk score could reasonably elect to omit radiation therapy, whereas patients in the intermediate or high-risk groups should consider adjuvant radiation therapy and tamoxifen following wide local excision.

One significant limitation in the application of the DCIS score to clinical practice is that the patients have to fit the specific profile of the patients in the E5194 study; this is challenging with regard to margin status, whereby many clinicians commonly consider margins of less than 3 mm sufficient. Patients with close or positive margins following wide local excision of DCIS are not appropriate candidates for molecular profiling at present. One must also consider a patient’s age, comorbidities, tumor size, tamoxifen therapy eligibility, and other clinical factors when making adjuvant radiation decisions.

ER-Positive, Node-Negative Invasive Breast Cancer

Adjuvant chemotherapy has been in widespread use for ER-positive, lymph node–negative breast cancers, even for small Stage I tumors, since the 1988 National Institutes of Health guidelines were published. Patients have been assigned to high-risk and low-risk groups based on clinical and pathologic characteristics in an attempt to personalize the treatment approach. The emergence of genomic techniques and the ability to measure the expression of many disparate genes have extended our ability to tailor therapy to the individual patient beyond that offered by standard clinical and pathologic parameters. Furthermore, molecular assays can provide both prognostic information (projected clinical outcome at the time of diagnosis independent of therapy) and predictive information (likelihood of response to a given therapeutic modality).

In a landmark study, Paik and colleagues validated the 21-gene RS as a predictor of the prospectively defined primary end point of distant recurrence-free survival in a large cohort of ER-positive, node-negative, tamoxifen-treated patients with breast cancer enrolled in the NSABP B-14 trial. The RS was calculated for each patient; 51% of the patient population fell into the low-risk group (n = 338), 22% fell into the intermediate-risk group (n = 149), and 27% fell into the high-risk group (n = 181). The rates of distant recurrence at 10 years were 6.8% for the low-risk group, 14.3% for the intermediate-risk group, and 30.5% for the high-risk group ( P <.001 low-risk vs high-risk).

In a subsequent validation study, the 21-gene RS was strongly associated with the risk of death from breast cancer in a similar cohort and also in patients who did not receive adjuvant systemic therapy. The objective of this additional study of the NSABP B-20 patients was to determine the magnitude of the chemotherapy benefit with methotrexate and fluorouracil (MF) or cyclophosphamide, methotrexate, and fluorouracil (CMF) as a function of the RS. The primary analysis was predetermined to compare the tamoxifen-treated patients with both chemotherapy arms combined. Patients with a low RS showed no evidence of benefit from CMF-like chemotherapy, and patients with an intermediate RS also did not show clear benefit from adjuvant CMF-like chemotherapy. By contrast, those patients with a high recurrence score treated with adjuvant CMF-like chemotherapy, in addition to endocrine therapy, experienced a significant improvement in distant relapse-free survival, with a 28% absolute benefit for adjuvant chemotherapy and tamoxifen ( P <.001). Furthermore, this association with RS and survival were independent from standard clinical and pathologic variables, supporting the utility of this molecular analysis beyond ER and HER2 status, tumor grade, and stage. One interesting subanalysis from the NSABP B-20 validation study is that 44% of the patients younger than 40 years had low-risk RS results. Although younger patients do worse overall and are probably more likely to benefit from chemotherapy, there is a large fraction of younger patients for whom the RS is low and benefits of chemotherapy may be minimal.

The 21-gene RS has been endorsed by American Society of Clinical Oncology and 2 expert panels (NCCN Breast Cancer Clinical Practice Guidelines and the 2011 St Gallen International Expert Consensus). These panels consider RS to be useful for patients with ER-positive, node-negative breast cancer as an aid to decision-making for administering adjuvant chemotherapy. The management of patients with intermediate-risk RS is being studied in the TAILORx (Trial Assigning IndividuaLized Options for Treatment) trial, where patients with ER-positive, node-negative breast cancer and intermediate-risk RS were randomized to either endocrine therapy alone or endocrine therapy with adjuvant chemotherapy. The study has completed accrual, but the survival data will not be available for several years.

The RS assay has also been shown to predict the likelihood of locoregional recurrence. In an analysis of patients from the NSABP B-14 and B-20 trials, there was a statistically significant association between the RS and locoregional recurrence regardless of the treatment group. In patients treated with tamoxifen alone, the rate of locoregional recurrence at 10 years was only 4.3% for those with a low-risk RS compared with 7.2% for those with an intermediate-risk RS and 15.8% for those with a high-risk RS ( P <.001). There were also significant associations between the RS and locoregional recurrence in placebo-treated patients from NSABP B-14 ( P = .022) and in chemotherapy plus tamoxifen-treated patients from NSABP B-20 ( P = .028). The RS was an independent predictor of locoregional recurrence in multivariate analysis, in addition to patient’s age and type of initial treatment. What remains unanswered is whether the recommendations for adjuvant radiation therapy should be altered based on these molecular data. The omission of adjuvant radiation for patients with a low-risk RS is a potential future area of research interest.

The 70-gene signature has been approved by the Food and Drug Administration for the prognostication of patients with Stage I or II, node-negative, invasive breast cancer of tumor size less than 5 cm. This profile was based on an empirical microarray analysis of 78 breast cancers from patients who were younger than 55 years, who had tumors that were 5 cm or less and node-negative, and who had not received any adjuvant systemic therapy. The patients were stratified into poor-prognosis disease if distant metastasis occurred within 5 years and good prognosis if no metastasis occurred within 5 years. The optimum threshold between these disparate outcomes was identified and applied to a cohort of 295 retrospectively accrued invasive breast cancers, of which 61 were node-negative in the initial study. This analysis revealed that the 70-gene signature was an independent prognostic marker of outcome. Subsequent studies have shown that the 70-gene signature also correlates with chemotherapy sensitivity, whereby patients with poor prognosis signatures derive the greatest benefit from adjuvant chemotherapy. There is also Level II evidence that the 70-gene signature more accurately predicts outcome in cases where there is disagreement between the 70-gene signature and standard clinical and pathologic features. The MINDACT (Microarray In Node-Negative and 1 to 3 positive lymph node Disease may Avoid Chemo Therapy) is a prospective, randomized phase 3 trial that was designed to answer this question, comparing the 70-gene signature with the common clinicopathologic characteristics in selecting patients with breast cancer with 0 to 3 positive nodes for adjuvant chemotherapy.

The PAM50 assay provides a ROR prognostic for relapse-free survival in patients with node-negative tumors who did not receive adjuvant systemic therapy. The ability of the ROR to predict prognosis was confirmed in a subsequent study of 786 patients with ER-positive disease treated only with adjuvant tamoxifen. When combined with tumor size, the ROR was more predictive of outcome when compared with standard clinicopathologic variables, including tumor grade, PR status, and Ki-67 proliferation marker. Dowsett and colleagues showed that the ROR outperformed the RS in its ability to stratify patients with ER-positive, node-negative breast cancer. More patients were scored as high risk and fewer as intermediate risk by ROR in comparison with RS, suggesting that the ROR may be a better prognostic indicator than the RS among this population.

ER-Positive, Node-Positive Invasive Breast Cancer

Node positivity continues to be an indication for cytotoxic therapy in patients with ER-positive, node-positive breast cancer, and this recommendation is included in the current NCCN guidelines. It is well known that an increasing amount of lymph node disease is associated with poorer outcomes. However, there is a recognized subset of node-positive women who never develop metastatic disease despite not being treated with cytotoxic therapy. Therefore, within this population of node-positive patients, multigene assays are being used to further stratify patients in terms of prognosis; similarly to patients with node-negative disease, those with higher risk of distant recurrence could be treated more aggressively.

Multiple studies have evaluated the prognostic capability of gene-expression signatures in the setting of node-positive breast cancer, although many of these studies are small and have limited follow-up. A subset analysis of the Southwestern Oncology Group (SWOG) 8814 trial was performed to determine whether the RS could provide prognostic information for patients with node-positive disease as it does for node-negative disease. In the original SWOG trial, postmenopausal women with hormone receptor–positive, node-positive breast cancer were randomized to tamoxifen alone, cyclophosphamide, doxorubicin, fluorouracil (CAF) chemotherapy with sequential tamoxifen, or CAF with concurrent tamoxifen. The addition of CAF to tamoxifen improved disease-free and overall survival for patients with 1 to 3 positive nodes and for patients with 4 or more positive nodes. Oncotype DX was performed on banked tissue samples from the trial (148 in the tamoxifen-alone arm and 219 in the sequential CAF-tamoxifen arm). The RS significantly predicted 10-year disease-free and overall survival for patients in the tamoxifen-alone arm after adjusting for the amount of nodal disease: 10-year disease-free survival was 60%, 49%, and 43%, respectively, for patients with low-, intermediate-, and high-risk RS ( P = .017). Overall survival was 77%, 68%, and 51% ( P = .003) for the same groups.

Dowsett and colleagues analyzed tissue samples from the ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial to determine whether the RS could predict distant recurrence in both node-negative and node-positive patients. For 872 node-negative patients, 9-year distant recurrence rates were 4%, 12%, and 25% for the low-, intermediate-, and high-risk RS groups, respectively. A similar trend was seen for the 306 node-positive patients, for whom distant recurrence rates were 17%, 28%, and 49% ( P = .002). These rates reflect adjustment for tumor size, grade, patient’s age, and treatment (tamoxifen vs anastrozole). Tumor size and amount of lymph node disease remained significant predictors of time to distant recurrence on multivariate analysis. The HR between low and high RS was more pronounced in the node-negative group (5.2) when compared with the node-positive group (2.7). Overall survival for node-negative patients with low-, intermediate-, and high-risk RS were 88%, 84%, and 73%, respectively. For node-positive patients, overall survival was 74%, 69%, and 54%. It is worth noting that this is one of the only studies evaluating a gene assay on patients treated with aromatase inhibitors (rather than tamoxifen, as was used in most trials).

Distant disease and overall survival are also predicted by the Mammaprint 70-gene signature. Van de Vijver and colleagues reported a consecutive series of 295 patients with Stage I and II breast cancer, all younger than 53 years, and half node-positive. One hundred twenty of the 144 node-positive patients included received systemic therapy: chemotherapy, endocrine therapy, or both. Distant metastatic events and overall survival were significantly predicted by the 70-gene signature among all patients and among the node-positive subgroup (all P <.001). On multivariate analysis, poor prognosis signature, tumor size, and nonuse of chemotherapy remained the most significant predictors of distant metastasis (all P <.001). Similarly, Mook and colleagues applied the 70-gene signature to 241 patients with 1 to 3 positive nodes. Ten-year distant metastasis–free survival and breast cancer–specific survival were 91% and 96%, respectively, for women with good prognosis signatures, and 76% and 76% for women with poor prognosis signatures. On multivariate analysis, the gene signature predicted disease-specific survival better than traditional prognostic factors ( P = .005).

Genetic signatures can also provide prognostic information regarding local-regional disease. Mamounas and colleagues reported that RS predicts local-regional recurrence in node-positive patients. Node-positive, ER-positive patients included in the NSABP B-28 trial were treated with both chemotherapy and tamoxifen. RS significantly predicted local-regional recurrence for patients undergoing mastectomy ( P = .004) or lumpectomy with radiation ( P = .022), and for patients with 4 or more positive nodes ( P = .001); there was a nonsignificant trend for patients with 1 to 3 positive nodes ( P = .12).

Several molecular diagnostic tests, including Oncotype DX, the Mammaprint 70-gene signature, and PAM50, were compared by Prat and colleagues. A data set was created of ER-positive patients, both node-negative and node-positive, treated with tamoxifen. The investigators concluded that multiple gene-expression signatures were prognostic for node-negative, ER-positive cancers, but that this prognostic capability was significantly worse for node-positive cancers. In addition, a paucity of node-positive tumors, specifically those with low-risk and intermediate-risk RS, had at least 90% distant recurrence-free survival at 8.5 years’ follow-up, and therefore only a small minority of these patients would receive little benefit from chemotherapy.

In addition to providing this prognostic information in the setting of node-positive disease, molecular diagnostics can also provide predictive information. In the SWOG subanalysis performed by Albain and colleagues, RS was seen to predict survival benefit with chemotherapy over tamoxifen alone; no significant disease-free or overall survival improvement was seen with CAF for patients with low-risk and intermediate-risk RS, but CAF did improve these survival rates for patients with high-risk RS.

To provide more robust data on the predictive power of these molecular assays in the setting of node-positive, ER-positive disease, several large prospective trials are under way. MINDACT includes multiple treatment arms; clinical and pathologic factors and the 70-gene signature are used to stratify patients. Poor-prognosis patients are given chemotherapy and, if they are ER-positive, endocrine therapy. Good-prognosis patients are treated with endocrine therapy if they are ER-positive. Patients whose clinicopathologic features and gene signature are discordant are randomized; either the clinicopathologic feature or the gene signature is used to determine treatment recommendations. SWOG S1007 (the RxPonder [Rx for POsitive NoDe Endocrine Responsive] Trial) includes women with hormone receptor–positive, HER2-negative breast cancer and 1 to 3 positive nodes. Patients with an RS less than or equal to 25 are randomized to chemotherapy plus endocrine therapy or to endocrine therapy alone. Patients with an RS greater than 25 are not randomized, but treatment is determined by the treating physician.