Main

We used a variety of human metastatic cancer lines and tumours to discover which microRNA, if any, was correlated with metastatic behaviour, rather than to find a prognostic clinical marker. Our functional analysis of miR-10b suggested that it is mechanistically important2, but we have no idea whether its expression in early-stage whole-tumour specimens can predict clinical progression. The discovery by Gee et al.1 that miR-10b is not a useful prognostic marker is therefore not surprising.

Whereas Gee et al.1 studied early-stage cancers, we analysed 23 primary tumour samples obtained at the time of mastectomy, when both lymph node and distant metastases had already developed. On the basis of these findings2, we concluded that miR-10b expression increased in metastatic breast tumours—not that its expression in unfractionated bulk populations of tumours could predict metastatic relapse2. MiR-10b is also significantly upregulated in human glioblastoma multiforme3 and in pancreatic adenocarcinomas4, two types of highly invasive and/or metastatic cancers.

Our results indicate that induction of miR-10b expression is probably not an early event during tumour progression, but occurs after activation of the Twist transcription factor at a later stage2. We propose that expression of Twist and miR-10b in many primary tumours is likely to be only local and transient, occurring in a minority of cells such as those in the invasive front of a breast tumour that are responding to contextual signals from nearby stromal cells to undergo an epithelial–mesenchymal transition. However, it is technically difficult to capture this minority population of cancer cells. Almost all microRNA expression analyses performed so far have been on bulk individual tumours consisting of neoplastic cells and recruited non-neoplastic stromal cells. These are therefore liable to underestimate or even misinterpret the range of messenger RNAs and microRNAs expressed by a small minority of the cells within such tumours.

Meaningful clinical data may eventually come from determination of which types of primary tumour cell do and do not express Twist and miR-10b when exposed to certain heterotypic contextual signals of stromal origin. Tumours that behave badly clinically might be those that upregulate Twist and miR-10b in response to such signals, but this is still years away from miR-10b as a prognostic marker. We agree with Gee et al.1 that sufficient numbers of patients with long-term follow-up are needed for clinical correlative studies of molecular markers in breast cancer and other diseases.