Circulating tumor cell detection and transcriptomic profiles in early breast cancer patients.

In the early 2000s, studies using high-throughput transcriptomic analyses revealed different aspects of breast cancer biology. Several outcome-based predictors and biologybased classifiers have been proposed: invasiveness gene signature (IGS), intrinsic molecular subtype, wound-response signature, etc. Recently, circulating tumor cell (CTC) cytological detection has been associated with metastasis-free and overall survival in the REMAGUS02 neoadjuvant trial [1]. In this article, we report, for the first time, the transcriptomic analysis of 60 nonmetastatic breast cancers according to CTC detection. The REMAGUS02 trial included breast cancer patients with locally advanced or large breast cancer. CTC detection was carried out at baseline and at the end of chemotherapy using the CellSearch system (Veridex, Raritan, NJ). Total RNA extracted from the pretreatment cancer biopsy was hybridized on the GeneChip Human Genome U1331 2.0 Array (Affymetrix, Santa Clara, CA). Conditions of the sampling procedures and RNA quality control are detailed elsewhere [2]. The genechip robust multi-array average procedure [3] was used to normalize the gene expression data. A hierarchical clustering was then carried out using the 10 000 probe sets that showed the highest values in the interquartile range (IQR). We applied the intrinsic gene set on the complete dataset of 60 samples to define the molecular subtypes and the IGS. We used the defined and validated centroids of 306 genes to discriminate between previously identified molecular breast cancer subtypes. We matched the probe list UniGene ID (Build#204) to the GeneChip Human Genome U133A 2.0 Array, resulting in a list of 294 unique probe sets. Each sample was assigned to the nearest subtype/centroid as determined by the highest Spearman rank order correlation between the gene expression values of the 294 probe sets and the 5 subtype centroids. The IGS score was determined by calculating the Pearson correlation between the probe set expression values of each sample and the 110 reference expression levels of the same genes defining the IGS signature. Single probe set analyses were carried out using the Wilcoxon rank sum test. Finally, because half of the probe sets showed the highest IQR values, a significance analysis of microarray (SAM) was proposed to detect differentially associated genes (DEG) between CTCpositive and -negative patients. At first, unsupervised clustering revealed three major tumor clusters; they unsurprisingly corresponded to triplenegative, human epithelial growth factor receptor 2(HER2)positive/estrogen receptor (ER)-negative, and ER-positive breast cancer immunohistological phenotypes. CTC detection was not statistically different among these three clusters (Figure 1). Breast cancers were then classified according to selected biology-based signatures: the intrinsic subtype classifier and IGS, which is stemness related. CTC detection was not statistically different among subgroups: basal, n = 6/18 (33%); HER2, n = 3/10 (33%); luminal A, n = 2/15 (13%); luminal B, n = 2/10 (20%); normal like, n = 2/7 (29%); low-risk IGS, n = 9/27 (33%); and high-risk IGS, n = 6/33 (18%). Single probe set analyses were then carried out on candidate genes that are directly involved in CTC detection by the CellSearch system (cytokeratin 8, 18, 19, and EpCAM) or that are surrogate markers for breast cancer stem cells (CD24, CD44, ALDH1A1): messenger RNA levels were not correlated with CTC positivity. In addition, at the high false discovery rate of 30%, only 18 DEG were found using a SAM procedure. This is the first study, to date, to correlate CTC detection in nonmetastatic breast cancer with the gene expression profile of the primary tumor. We did not confirm previously published in vitro experiments that suggested that normal-like