Recently, we demonstrated that inducible T-cell costimulator (ICOS) shares its unique ligand (ICOSL) with osteopontin (OPN), and OPN/ICOSL binding promotes tumor metastasis and angiogenesis in the 4T1 breast cancer model. Literature showed that OPN promotes melanoma metastasis by suppressing T-cell activation and recruiting myeloid suppressor cells (MDSC). On the opposite, ICOS/ICOSL interaction usually sustains an antitumor response. Here, we engineered murine B16F10 melanoma cells, by transfecting or silencing ICOSL. In vitro data showed that loss of ICOSL favors anchorage-independent growth and induces more metastases in vivo, compared to ICOSL expressing cells. To dissect individual roles of the three molecules, we compared data from C57BL/6 with those from OPN-KO, ICOS-KO, and ICOSL-KO mice, missing one partner at a time. We found that OPN produced by the tumor microenvironment (TME) favors the metastasis by interacting with stromal ICOSL. This activity is dominantly inhibited by ICOS expressed on TME by promoting Treg expansion. Importantly, we also show that OPN and ICOSL highly interact in human melanoma metastases compared to primary tumors. Interfering with this binding may be explored in immunotherapy either for nonresponding or patients resistant to conventional therapies.