Introducing Molecular Subtyping of Breast Cancer Into the Clinic?

Breast cancer is a collection of diseases demonstrating heterogeneity at the molecular, histopathologic, and clinical level. The diversity at the molecular level is manifested in differences in gene expression patterns (both of mRNA and microRNA), different frequencies and magnitudes of genomic aberrations, and differential protein expression across breast tumors, even among those of similar histopathologic type. In addition to this, and influencing this diversity in tumor cells, is the diversity in the microenvironment, reflecting different degrees of involvement of various biologic processes. The molecular heterogeneity is reflected in the clinical course of the diseases and responses to treatment. During the last 10 years, whole-genome analyses using microarrays have revolutionized cancer research. Such studies of breast tumors have led to the identification of five molecular subtypes associated with differences in patient survival, as well as numerous gene expression signatures associated with different clinical parameters. Some of these have formed the basis for commercialized tests that are now being assessed in large clinical trials; however, none has taken into account the heterogeneity represented by the molecular subtypes. In this issue of Journal of Clinical Oncology, Parker et al report on a risk prediction model for breast cancer developed from expression data of 50 genes representing the five intrinsic molecular subtypes. They show that the intrinsic subtypes are present in several tumor cohorts, both those positive or negative for estrogen receptor (ER) and human epidermal growth factor receptor 2 (HER-2), and are associated with significant differences in relapse-free survival in several published breast cancer microarray data sets. Importantly, this illustrates that the molecular subtypes are not just recapitulating the classic clinical markers for classification; for example, basal-like tumors cannot simply be substituted by triple-negative (ER negative/ progesterone receptor negative/HER-2 negative) status. In addition, they show in a multivariable analysis that the subtypes are able to predict nonresponse to a neoadjuvant paclitaxel plus fluorouracil, doxorubicin, and cyclophosphamide regimen with high sensitivity and high negative predictive value, although at the expense of specificity. The predictor was built from a microarray training set consisting of a published subcohort of patients with node-negative untreated breast cancer. A model that incorporated both the intrinsic subtypes and tumor size resulted in improved risk prediction for relapse in untreated patient cohorts compared with either subtypes or clinical markers alone. A final validation step by quantitative reverse transcriptase polymerase chain reaction test using formalin-fixed paraffinembedded material showed similar subtype distribution and prognostic capability of the predictor. A strength of the study is the ability to assign subtypes with a limited gene set using archived tumor tissue with extensive follow-up information. The predictor classified patients into three risk groups (high, medium, and low), two of which contained each of the subtypes; the low-risk class consisted only of luminal A type tumors. Conversely, the risk predictor divided patients with luminal A type tumors into both lowand intermediate-risk groups, which is of significant value if this may help to identify patients who might be spared aggressive treatment. Although patients with luminal A type tumors are associated with relatively good prognosis, a significant portion experience relapse, indicating heterogeneity within this group. It remains to be seen if this three-category risk predictor is superior to any of the other published prognostic predictors based on gene expression signatures. Another strength of this model is that it provides prognostic value for all types of breast cancer patients, irrespective of ER and node status as well as disease stage. Whether this same model may robustly predict nonresponse to treatment is still unclear, given that this study predicted residual disease in a preoperative cohort designed to predict pathologic complete response, and in which patients with residual disease comprised the major part. These five molecular subtypes, which were validated extensively, are profoundly different in their gene expression patterns (including microRNA expression), genetic alterations, and distribution of mutations and normal variants. This suggests that they are developing along distinct pathways and are different mechanistically. This is especially true for the luminal A and basal-like subtypes. These are inversely correlated and protrude in any genome-wide study at the genomic, transcriptomic, and proteomic level. They probably originate from different lineages of progenitor cells (or stem cells) at different stages of differentiation. The distinctiveness of the three other subtypes is more unclear. The ERBB2-positive group is characterized by strong expression of the ERBB2 oncogene and a few other genes located in the same region on chromosome 17q as a result of amplification. However, amplification of this region is also seen in luminal B tumors, a subtype of tumors expressing ER and other estrogen-related genes, but which also share some expression characteristics with basallike tumors. The normal-like tumor subtype, originally defined by its similarity in expression with normal tissue and benign tumors, is not just a reflection of poor tissue sampling, but is nevertheless not JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L VOLUME 27 NUMBER 8 MARCH 1