Metastatic growth in distant organs is the major cause of cancer mortality. The development of metastasis is a multistage process with several rate-limiting steps 1 . Although dissemination of tumour cells seems to be an early and frequent event 2 , the successful initiation of metastatic growth, a process termed 'metastatic colonization', is inefficient for many cancer types and is accomplished only by a minority of cancer cells that reach distant sites 3,4 . Prevalent target sites are characteristic of many tumour entities 5 , suggesting that inadequate support by distant tissues contributes to the inefficiency of the metastatic process. Here we show that a small population of cancer stem cells is critical for metastatic colonization, that is, the initial expansion of cancer cells at the secondary site, and that stromal niche signals are crucial to this expansion process. We find that periostin (POSTN), a component of the extracellular matrix, is expressed by fibroblasts in the normal tissue and in the stroma of the primary tumour. Infiltrating tumour cells need to induce stromal POSTN expression in the secondary target organ (in this case lung) to initiate colonization. POSTN is required to allow cancer stem cell maintenance, and blocking its function prevents metastasis. POSTN recruits Wnt ligands and thereby increases Wnt signalling in cancer stem cells. We suggest that the education of stromal cells by infiltrating tumour cells is an important step in metastatic colonization and that preventing de novo niche formation may be a novel strategy for the treatment of metastatic disease.We aimed to explore limiting factors that determine metastatic success using the MMTV-PyMT mouse breast cancer model, which spontaneously metastasizes to the lungs 6 . We reasoned that the recently identified cancer stem cells (CSCs, also called tumour-initiating cells), a subset of cancer cells that allow long-term tumour growth and are thought to be responsible for remissions 7,8 , might also be relevant to the development of metastatic disease ( Supplementary Fig. 1). We measured the relative size of the population of CSCs from primary MMTV-PyMT tumours and their pulmonary metastases using the previously established markers CD90 and CD24, which label a subset of the CD24 1
CD29hi or CD241 CD49f hi population used earlier to isolate CSCs and normal mammary gland stem cells 9-13 (Supplementary Fig. 2). This CSC subset accounts for 3 6 2.1% (s.d.) of all tumour cells from both primary tumours and metastases (Fig. 1a). When CD90 1
CD241 CSCs or CD90 1 CD24 1 -depleted non-CSCs are separately isolated from GFP 1 tumours and directly introduced into mice through tail vein injection (GFP, green fluorescent protein), only the CSC population is able to produce lung metastases (Fig. 1b). Moreover, CD90
1
CD241 cells isolated subsequently from pulmonary metastases are again the only tumour cell population that efficiently initiates secondary metastases (Fig. 1c). This is not due to differences in the extravasation capabilities of CSCs an...