Genomic testing for sensitivity of breast cancer to hormonal therapy.

Optimal selection of endocrine therapy for women with estrogen receptor–positive breast cancer will require biomarkers with better positive predictive value than the current immunohistochemical assay. There has been great interest to employ genomic technology to identify the subset of women with estrogen receptor–positive breast cancers who benefit from hormonal agents, and from those studies we consistently realize the importance of study design for discovery and validation. Translation of a predictive biomarker into a diagnostic test requires well-designed clinical trials for proof of benefit. The most common design is to empirically discover biomarkers from archival tumor samples with annotated survival after monotherapy and to then measure the predictive accuracy in an independent cohort. This conveniently uses existing samples and known clinical outcome but also has obstacles relating to sample size, unseen bias, availability and usability of archival samples, and translation of technology from a broad discovery platform to a focused test. The clinical context of treatment, however, presents a potentially greater challenge. Adjuvant hormonal therapy is administered over a period of years and may then be switched to a different hormonal agent. Treatment often follows adjuvant chemotherapy, making it difficult to determine the survival benefit from different treatments. Those who receive only hormonal therapy usually have early-stage breast cancer; therefore, adjuvant studies cannot distinguish treatment responders from those who would have survived from surgery alone. Finally, because survival differences between estrogen receptor–positive and estrogen receptor– negative breast cancers are time dependent, the duration of study follow-up affects the results. Whereas neoadjuvant studies do circumvent the uncertainties of follow-up, the primary end point of pathologic complete response is infrequent and it is difficult to translate a prediction of clinical response after a few months into the adjuvant setting. The development of a HOXB13/IL17BR gene expression ratio illustrates the complexity of developing a predictive biomarker for hormonal therapy. The ratio of HOXB13/IL17BR gene expression was first derived from microarrays of 60 frozen samples of estrogen receptor–positive breast cancer based on their association with distant relapse after adjuvant tamoxifen therapy. In the original microarray data, this two-gene ratio was independently associated with distant relapse-free survival (odds ratio, 7.3; 95% confidence interval, 2.1-26.3) in a multivariate model that included other prognostic factors (1). The investigators also developed a real-time reverse transcription-PCR assay to measure the expression of both genes relative to a housekeeping gene and showed that the reverse transcription-PCR assay provided equivalent results in both frozen and formalin-fixed paraffin-embedded tissue samples. A pilot validation was then conducted in 20 selected formalin-fixed paraffin-embedded samples (10 patients relapsed, 10 did not) and yielded promising results (1). That set the scene for a larger study [presented in part by Sgroi et al. at American Society of Clinical Oncology 2004 (2) and published in full by Goetz et al. (3) in this issue] in which the two genes were measured using reverse transcription-PCR in 206 archival primary tumor samples from a completed multicenter clinical trial. The ratio of HOXB13/ IL17BR expression was compared with survival after adjuvant tamoxifen treatment (3). In the interim, there has been a dialogue of published letters and reports about this two-gene ratio in estrogen receptor– positive breast cancer. A letter from the authors clarified that 19 of the 20 formalin-fixed paraffin-embedded samples were from lymph node–negative breast cancers (4). Meanwhile, a different group tested the ratio using reverse transcription-PCR of HOXB13, IL17BR, and a housekeeping gene, but in 58 frozen samples and using primers that recognize a different region of each gene transcript (5). The ratio of HOXB13/ IL17BR expression had no significant association with distant relapse in this second study (5). Seventy-seven percent of the patients in the second study, however, had node-positive breast cancer. The accompanying editorial discussed in detail that sample size, study bias, different assay methods, and lack of independent validation limit the ability to reliably interpret such phase II diagnostic studies (6). A third study from Jansen et al. (7, 8) tested the ratio of HOXB13/IL17BR gene expression from microarrays of 112 frozen primary estrogen receptor– positive breast cancer samples (nodal status not stated) and reported a significant association with objective response (versus progressive disease) when the women had relapsed and were treated with first-line tamoxifen therapy (area under the curve, 0.612; P = 0.04). These studies were small, used different assays, and addressed different stages of breast cancer. In addition, the biological relevance of this two-gene expression ratio from the original primary tumor sample in patients with relapsed disease is unknown (1, 7–9). The current study is not seen as an independent validation of the original two-gene ratio test because the assay and the methods of interpretation were changed substantially to accommodate archival formalin-fixed paraffin-embedded Editorial