Breast cancer is a clinically heterogeneous disease that can affect individuals with seemingly identical clinicopathologic parameters differently. This clinical heterogeneity is driven to a large extent by abnormal gene expression within tumors. Investigators now have the ability to identify the gene-expression fingerprint of an individual’s tumor. This information may be used to rationally design therapeutic targets in the future, and also to predict the clinical course of an individual’s disease, including response to a specific treatment.

Genetic profiles of tumors are now being correlated with clinical outcome, and several prognostic and predictive indicators have emerged based on this research. There are at least four commercially available predictive or prognostic tests, and several more are looming on the horizon. The data gathered from these tests augment standard diagnostic and prognostic information obtained from traditional clinical pathological variables. The advent of gene-profiling technologies started to change the conduct of clinical trials. In the not too distant future, prospective tissue collection for molecular analysis may become routine in order to stratify patients for treatment arms and to optimize treatment strategies based on molecular features of the cancer. Coordinated efforts among oncologists, pathologists, surgeons, laboratory scientists, statisticians, and regulators will be essential in the quest to incorporate genetic profiling and molecular hypotheses into clinical trial planning and conduct.

Breast cancer is a clinically heterogeneous disease; individuals with the same stage of disease and similar pathological diagnoses can experience very different clinical courses ^[1]. This clinical heterogeneity is driven by the genetic variability of patients and tumors. It is widely acknowledged that a continuum of abnormal gene expression predicts the tumorigenic phenotype and the sensitivity of tumors to treatment (Fig. 1) ^[2]. Clinical investigators now have the capability to create a genetic blueprint of individual tumors; the genetic abnormalities identified within these tumors offer a hope to rationally select therapeutic targets for the treatment of patients with cancer ^[1]. Ultimately, researchers aim to use the molecular data gathered from an individual tumor for prognostication and customization of therapy for each patient. Gene-expression profiling has shown promise to distinguish between patients at low and high risk for developing distant metastases and identify those who are likely to benefit from adjuvant therapy ^[3]. Prognostic genetic tests for patients with breast cancer are now commercially available, and additional tests will be available in the very near future. This article reviews recent developments in molecular diagnostics and discusses the impact these developments may have on the future of breast cancer therapy.

The current standard for prognostic stratification includes Adjuvant! Online, the Nottingham Prognostic Index, and the American Joint Committee on Cancer staging system, which form the basis of treatment guidelines issued by the National Institutes of Health (NIH) Consensus Statement on Adjuvant Therapy in Breast Cancer and the St. Gallen Consensus Statement ^[4–8]. These tools integrate clinicopathologic factors into multivariate prediction models. Although these tools allow clinicians to estimate the relative risks for recurrence and mortality and estimate the potential benefits of chemotherapy for groups of patients with given disease characteristics ^[7], they do not address the fundamental question oncologists and patients struggle with: who as an individual (rather than as a group) will benefit from adjuvant therapy? Up to 40%–50% of patients with a poor prognosis as defined by conventional clinicopathological parameters may remain disease free without adjuvant therapy ^[3]. Likewise, benefit from systemic adjuvant chemotherapy for patients with lymph node–negative (LNN) disease is not uniform; some patients relapse despite therapy and others may already be cured by locoregional treatment ^[1]. More accurate molecular prognostic and response prediction tools could assist in minimizing overtreatment of low-risk patients and reduce undertreatment of high-risk patients, who are also sensitive to existing systemic treatment modalities.

Intrinsic Subtype Predictor
A novel molecular classification of breast cancer was proposed based on large-scale gene-expression analyses of breast cancer ^[9]. Four major molecular classes of breast cancer emerged from several studies: luminal-A, luminal-B, basal-like, and human epidermal growth factor receptor (HER)-positive cancers ^[9–11]. The overall survival and chemotherapy sensitivity of the different molecular subgroups vary. Luminal-type cancers are mostly estrogen receptor (ER) positive, and patients with luminal-A cancers have the most favorable long-term survival (with endocrine therapy) compared with the other types, whereas basal-like and HER-2–positive tumors are more sensitive to chemotherapy ^{[10, 12, 13]}. The Intrinsic subtype predictor was developed to assign molecular classes to newly diagnosed breast cancers ^[14].

Rotterdam 76-Gene Set
The Rotterdam gene set was developed to predict the prognosis of patients with LNN breast cancer ^[15]. Two hundred eighty-six patients who had locoregional therapy only were included in the initial marker development and validation study. Markers were selected separately from ER-negative and ER-positive tumors and were combined into a single 76-gene prognostic signature (VDX2; Veridex, LLC, Warren, NJ) that was able to predict distant metastatic recurrence with a sensitivity of 93% and a specificity of 48% ^[15]. This prognostic indicator performed better than standard, clinical variables in a multivariate analysis (hazard ratio ^[HR], 5.55; 95% confidence interval ^[CI], 2.46–12.5). Subsequently, this test was also evaluated on two other cohorts of patients who were not included in the original study. The first cohort included 180 patients with stage I–II breast cancer and showed 5- and 10-year distant metastasis-free survival rates of 96% (95% CI, 89%–99%) and 94% (95% CI, 83%–98%), respectively, for the good prognosis group; the corresponding rates were 74% (95% CI, 64%–81%) and 65% (95% CI, 53%–74%) for the poor prognosis group ^[16]. The sensitivity for 5-year metastasis-free survival was 90%, and the specificity was 50%, with positive and negative predictive values of 38% and 94%, respectively. The second validation cohort included 198 LNN cases and demonstrated similarly good 5- and 10-year distant metastasis-free survival rates: 98% (95% CI, 88%–100%) and 94% (95% CI, 83%–89%), respectively, for the genomic low-risk group ^[17]. The recurrence rates were significantly worse for the poor prognosis group: 76% (95% CI, 68%–82%) and 73% (95% CI, 65%–79%) at 5 and 10 years, respectively ^[17]. Importantly, the 76-gene signature could restratify patients within the clinical risk categories defined by the Adjuvant! Online program and the recurrence HRs remained similar after adjustments for tumor grade, size, and ER status.

Invasive Gene Signature
The putative breast cancer stem cell, which can be identified by low expression of CD24 and high expression of CD44, is highly tumorigenic in experimental models ^[18–20]. When normal breast epithelial cells were compared with breast CD44+/CD24– cells, 186 genes that are associated with tumorigenic breast “stem” cells were discovered ^[21]. These genes became known as the invasive gene signature (IGS), and their presence was significantly associated with shorter overall survival and metastasis-free survival times (p

< .001). The IGS was combined with prognostic criteria from the NIH and used to stratify patients with early-stage breast cancer, who are at high risk for metastasis or death, into good or poor prognostic groups. The 10-year metastasis-free survival rate was 81% among patients in the good prognosis group, but only 57% for patients in the poor prognosis group. Moreover, there was an association between the IGS and clinical outcome in patients with tumors that showed intermediate-grade differentiation.

Wound Response Indicator
Tumors have been compared to nonhealing wounds ^[22]. Because the response of serum-stimulated fibroblasts includes many of the processes involved in wound healing ^[23], the wound response indicator (WRI) was developed from genes whose expression changed following the activation of cultured fibroblasts with serum ^[24]. This signature was validated in patients with early-stage breast cancer (n = 295) ^[25]. Patients whose tumors expressed the WRI had significantly shorter overall survival and distant metastasis-free survival times relative to patients whose tumors did not express this gene signature. Moreover, the WRI signature was an independent predictor of death in a multivariate analysis of metastasis and death.

The oncotype DX^TM Recurrence Score^TM
The oncotype DX^TM (Genomic Health, Inc; Redwood City, CA) is a 21-gene indicator. Two hundred fifty candidate genes were chosen from gene-expression profiling experiments, published literature, and genomic databases ^{[9, 13, 26, 27]}; these genes were correlated with breast cancer recurrence in 447 patients ^[28]. Sixteen cancer-related genes and five reference genes were selected from the candidate genes. The 16 cancer-related genes were then used to develop an algorithm based on the expression levels of these genes, thus allowing a Recurrence Score^TM (RS) to be computed for each specimen. This RS correlated with the rate of distant recurrence at 10 years. This assay uses fixed tumor specimens, rather than frozen tissue.

The oncotype DX^TM assay was externally validated in the National Surgical Adjuvant Breast and Bowel Project (NSABP) clinical trial B-14, which examined the effect of adjuvant tamoxifen in patients with hormone receptor–positive LNN breast cancer ^[28]. Approximately 25% of the tumors from the tamoxifen arm were analyzed, and their RSs were compared with patient outcomes after >10 years of follow-up. The results showed that 7% of low-risk patients (RS <18) relapsed, whereas 31% of high-risk patients (RS >31) relapsed. Subsequent studies have shown that the RS is independently associated with sensitivity to chemotherapy ^{[29, 30]} and mortality ^[31].

The RS is now being prospectively validated in the Trial Assigning Individualized Options for Treatment (TAILORx) clinical trial ^[32]. In this phase III trial, women who have undergone surgery for ER-positive LNN breast cancer are being assigned to one of three groups based on their RS: group 1, RS <11; group 2, RS ≥11–25; group 3, RS >25. It is important to note that these cutoff values are different from those used in all previous studies (i.e., low RS group, < 18; high RS group, ≥31). These more conservative thresholds to categorize patients into good, intermediate, and high risk were selected to err on the safe side and not to exclude from adjuvant chemotherapy ER-positive patients who may conceivably have a small chance of benefiting from it. In the TAILORx study, patients in group 1 receive hormone therapy alone. Patients in group 3 receive combination chemotherapy and hormone therapy. Patients in group 2 are randomized to receive either chemotherapy and hormone therapy or hormone therapy alone. The primary endpoints of this study are disease-free survival, distant recurrence-free interval, recurrence-free interval, and overall survival in group 2.

MammaPrint® 70-Gene Profile
The MammaPrint® (Agendia BV; Amsterdam, The Netherlands) 70-gene profile was developed from patients with LNN disease who were ?55 years of age. The frozen tumors were separated into two groups: (a) those from patients who developed distant metastases within 5 years of completing treatment and (b) those from patients who remained disease free for at least 5 years ^[27]. When the gene-expression profiles of these two groups were compared, a 70-gene profile correlating with clinical outcome was identified. This gene set was validated internally ^[33] and externally ^[34]. The first validation used a retrospective analysis of relatively young patients (age <53 years; n = 295) with LNN and lymph node–positive disease. Within this initial group, 115 patients with a mean 5-year survival rate of 97% were classified as having a good prognosis, and 180 patients with a mean 5-year survival rate of 74% were classified as having a poor prognosis ^[33]. The second validation was the Translating Molecular Knowledge Into Early Breast Cancer Management: Building on the Breast International Group Network for Improved Treatment Tailoring (TRANSBIG) study ^[34]. This study included women (n = 307) with LNN, T1–2 breast cancer who were <61 years of age and had not previously received treatment with systemic adjuvant therapy. The duration of follow-up for patients in this study was >10 years. The MammaPrint® gene set more accurately predicted time to distant metastases (HR=2.32; 95% CI, 1.35–4.00) and overall survival (HR, 2.79; 95% CI, 1.60–4.87) than Adjuvant! Online (HR, 1.68; 95% CI, 0.92–3.07). These validation studies led to the clearance of this test by the U.S. Food and Drug Administration (FDA), allowing the test to be marketed as a prognostic marker to be used with other clinicopathologic factors.

The ongoing Microarray in Node Negative Disease May Avoid Chemotherapy (MINDACT) clinical trial is currently evaluating the usefulness of the MammaPrint® gene set in determining systemic adjuvant therapy for patients with LNN breast cancer ^{[35, 36]}. That trial will compare the prognostic information provided by the 70-gene set with prognostic information provided by Adjuvant! Online ^{[37, 38]}. Investigators plan to enroll 6,000 LNN breast cancer patients; each will have their risk assessed through both Adjuvant! Online and the 70-gene profile. If both methods classify the patient’s risk for relapse as low (an estimated 13% of patients), adjuvant chemotherapy will be withheld; if both methods classify the patient’s risk for relapse as high (an estimated 55% of patients), then chemotherapy will be proposed; if the methods give discordant results (an estimated 32% of patients), the patient will be randomized to either follow the clinicopathological method or follow the genomic results. It is expected that 10%–20% of women who would normally receive adjuvant chemotherapy based on their clinicopathological factors will be spared this therapy, without having any negative impact on their survival ^{[35, 36]}.