Résumé : Cellular plasticity endows cancer cells with capacity towards a phenotypic state. Except from the genetic origin of cellular plasticity, cancer cells can transit to reversible phenotypic states as a result of microenvironmental cues and are often driven by stochastic epigenetic and/or transcriptional fluctuations. Melanoma is a prime example of heterogeneous and plastic tumors where the origin and magnitude of cell state diversity remains poorly understood. Combining lineage tracing, 3D imaging and quantitative mathematical modelling in a clinically-relevant mouse model of melanoma, we demonstrated that tumors follow a hierarchical model of growth supported by a population of Melanoma Stem-like Cells (MSCs) that exhibit a transcriptomic signature of pre-migratory neural crest cells established transiently during embryonic development. ScRNA seq data in mouse and drug naïve human biopsies revealed that MSCs are evolutionarily conserved and independent of the “genetic makeup” of the tumors. By deploying spatial transcriptomic approaches and multiplex imaging, we developed a spatially and temporally resolved map of the diversity of melanoma showing a non-random tumor organization. Multimodal analysis unraveled unique cell type and state interactions and importantly demonstrated that MSCs reside in perivascular niches favoring tumor growth. Endothelial Cells were found to have a key role in the acquisition of stemness properties. Co-culture assays led to melanoma dedifferentiation and proliferation advantage. Of note, supplementing melanoma cells with ECs accelerated the growth of tumors in vivo, where MSCs pool was increased. Taken together, these results will pave the way for the development of strategies that exploit the “chameleonic” nature of cancer cells and, ultimately, target MSC niche-dependent specification mechanisms