There is growing evidence that the metastatic spread of melanoma is driven not by a linear increase in tumorigenic
aggressiveness, but rather by switching back and forth between two different phenotypes of metastatic potential. In vitro these phenotypes are respectively defined by the characteristics of strong proliferation ⁄ weak invasiveness and weak proliferation ⁄ strong invasiveness. Melanoma cell phenotype is tightly linked to gene expression. Taking advantage of this, we have developed a gene expression–based tool for predicting phenotype called Heuristic Online Phenotype Prediction. We demonstrate the predictive utility of this tool by comparing phenotype-specific signatures with measurements of characteristics of melanoma phenotype-specific biology in different melanoma cell lines and short-term cultures. We further show that 86% of 536 tested
melanoma lines and short-term cultures are significantly associated with the phenotypes we describe. These findings reinforce the concept that a two-state system, as described by the phenotype switching model, underlies melanoma progression.