Transcriptional expression of 8 genes predicts pathological response to first-line docetaxel + trastuzumab-based neoadjuvant chemotherapy

In order to predict the response to treatment, we calculated the correlation coefficient of each tumor with each cell line. The correlation coefficient nearest to 1 corresponds to the prediction profile. As the tumor response was known, we eliminated genes one by one until we obtained the best prediction performances. This was achieved (Table 2) with the association of the expression of only 8 genes: CTNS, DERL-1, FAM114A2, KIAA1549, P2RX1, PITPNA, PSMD11, and WEE1. As an example, for patient A with a pCR and patient B with no pCR, the correlation between patient A’s tumor cells and the reference cell lines showed r = 0.93 with the sensitive HCC2218 cells, whereas we obtained r = −0.36 with the resistant HCC1419 cells, classifying it as a sensitive tumor (Figure 1A). In contrast, patient B’s tumor cells correlated positively with the resistant HCC1954 cell line (r = 0.85) and negatively with the sensitive BT474 cells (r = −0.25), classifying it as a resistant tumor (Figure 1B). All correlation coefficients are presented in Additional file 2. Surprisingly, the expression level of these 8 genes individually was not significantly different between pCR patients and non-pCR patients (Figure 1C), suggesting that the combination of the expressions more than the expression of each gene individually was responsible for the prediction capacity. In parallel, we also studied corresponding FFPE samples for 34 patients of our population for expression of the 8 genes. To determine whether our signature could also be assessed on FFPE samples, we calculated correlation coefficients of each gene between frozen and FFPE paired samples. Except for DERL-1 (r = −0.19), the 7 genes CTNS (r = 0.77), FAM114A2 (r = 0.19), KIAA1549 (r = 0.61), P2RX1 (r = 0.44), PITPNA (r = 0.44), PSMD11 (r = 0.39), and WEE1 (r = 0.13) correlated positively suggesting that paraffin treatment of samples did not alter the expression of the genes. The absence of a positive correlation for DERL-1 expression between frozen and FFPE samples could have been due to the smaller number of FFPE samples.

To test the prediction capacity of the combined expression of the 8 genes in different conditions, we used microarray datasets available on the GEO website. We first correlated the expression of our 8 genes with the expression of ERBB2. The 8 genes correlated significantly and positively with ERBB2 expression (Additional file 3). As our first gene expression signature was obtained from tumors treated with trastuzumab + docetaxel-based chemotherapy, we tested the prediction in patients treated with first-line neoadjuvant - docetaxel-based regimen (GSE22358), first-line neoadjuvant - trastuzumab monotherapy (GSE37946), or first-line neoadjuvant - trastuzumab + docetaxel-based chemotherapy (GSE22358). When the regimen contained docetaxel without trastuzumab (Figure 2A) or trastuzumab alone (Figure 2B), our profile was not able to dichotomize tumor response. In contrast, the response of patients treated with first-line trastuzumab + docetaxel-based chemotherapy was well classified by our profile (Figure 2C). Indeed, the accuracy of the classification was 85% (23/27), with a sensitivity of 92% (11/12) and a specificity of 80% (12/15). Finally, it appeared that the profile was not usable for a second-line neoadjuvant trastuzumab + docetaxel-based regimen, at least after first-line neoadjuvant 5-fluorouracile + epirubicin + cyclophosphamide (GSE42822) (Figure 2D). This could be explained by modifications in tumor cell gene expression induced by the first treatment, which may have influenced the subsequent response of tumor cells to docetaxel + trastuzumab.