Microenvironment-modulated metastatic CD133+/CXCR4+/EpCAM-lung cancer-initiating cells sustain tumor dissemination and correlate with poor prognosis

Metastasis is the main reason for lung cancer-related mortality, but little is known about specific determinants of successful dissemination from primary tumors and metastasis initiation. Here, we show that CD133+/CXCR4+ cancer-initiating cells (CIC) directly isolated from patient-derived xenografts (PDX) of non-small cell lung cancer are endowed with superior ability to seed and initiate metastasis at distant organs. We additionally report that CXCR4 inhibition successfully prevents the increase of cisplatin-resistant CD133+/CXCR4+ cells in residual tumors and their metastatization. Immunophenotypic analysis of lung tumor cells intravenously injected or spontaneously disseminated to murine lungs demonstrated the survival advantage and increased colonization ability of a specific subset of CD133+/CXCR4+ with reduced expression of epithelial cell adhesion molecule (EpCAM-), which also shows the greatest in vitro invasive potential. We next prove that recovered disseminated cells from lungs of PDX-bearing mice enriched for CD133+/CXCR4+/EpCAM-CICs are highly tumorigenic and metastatic. Importantly, microenvironment stimuli eliciting epithelial-to-mesenchymal transition, including signals from cancer-associated fibroblasts, are able to increase the dissemination potential of lung cancer cells through the generation of the CD133+/CXCR4+/EpCAM-subset. These findings also have correlates in patient samples where disseminating CICs are enriched in metastatic lymph nodes (20-fold, P=0.006) and their detection in primary tumors is correlated with poor clinical outcome (diseasefree survival: P=0.03; overall survival: P=0.05). Overall, these results highlight the importance of specific cellular subsets in the metastatic process, the need for in-depth characterization of disseminating tumor cells, and the potential of therapeutic strategies targeting both primary tumor and tumor-microenvironment interactions.