Tumor analyses have shown that DNA mutations can transform cells in some situations but not in others. In melanoma, BRAF^V600E [serine/threonine-protein kinase B-raf (BRAF) in which valine at position 600 is replaced with glutamic acid] expression coincides with the aberrant expression of a developmental signature typical of the neural crest, suggesting that the neural crest state might be linked to oncogenic competence. The mechanisms by which this state is acquired and gives rise to oncogenic competence are unknown.

During development, neural crest cells give rise to melanoblasts, which then further differentiate into melanocytes. Using a combination of zebrafish transgenesis and human pluripotent stem cell models, we assessed oncogenic response to BRAF^V600E in each of these three cell states. We used these models to determine why certain cell states but not others were preferentially competent to form tumors.

We engineered zebrafish to drive BRAF^V600E expression under the sox10 promoter, the mitfa promoter, or the tyrp1 promoter in the p53^−/− background. This allows for oncogene activation at the neural crest, at the melanoblast, or melanocyte stage, respectively. Fish that received the oncogenic mutation during the neural crest and melanoblast stages efficiently developed tumors, whereas melanocytes were relatively resistant. To extend these findings to human cells, we developed a human pluripotent stem cell (hPSC)–derived tumor model. hPSCs can differentiate into neural crest cells, melanoblasts, and melanocytes. We used gene targeting to introduce oncogenic BRAF^V600E and to inactivate the tumor suppressors RB1, TP53, and P16 in these cells. Analogous to what we found in the zebrafish, upon subcutaneous transplant into NSG mice, human melanocytes were relatively resistant to malignant transformation, whereas the neural crest cells and melanoblasts readily developed tumors. RNA-sequencing of these cells with or without BRAF^V600E revealed a robust transcriptional response to the oncogene in the neural crest cells and melanoblasts, whereas melanocytes had little change despite equal activation of phosphorylated extracellular signal–regulated kinase (pERK). This suggested that there were intrinsic differences across these cells that determined how they responded to oncogenic insult. Pathway analysis revealed significant up-regulation of several chromatin-modifying enzymes in the more competent neural crest and melanoblast cells, including readers, writers, and erasers (ATAD2, BPTF, BAZ1A, EZH2, and others). This raised the hypothesis that the neural crest cells and melanoblasts had higher intrinsic transcriptional plasticity compared with that of melanocytes because of these chromatin modifiers. Further investigation demonstrated that one of these factors, ATAD2, is commonly amplified or overexpressed in melanoma patients and is correlated with a worse prognosis. In vitro, expression of ATAD2 in hPSC-derived melanocytes led to acquisition of a more progenitor-like phenotype, and inhibition of ATAD2 impaired neural crest differentiation. In vivo, ATAD2 is sufficient to endow oncogenic competence to the melanocytes and allows for formation of melanoma in these otherwise more resistant cells. We show that ATAD2 forms a protein complex with SOX10, a key transcription factor required for neural crest development. Analysis with ATAC-seq, RNA-seq, and Cut&Run demonstrate that ATAD2 facilitates expression of SOX10 developmental target genes as well as genes related to the mitogen-activated protein kinase (MAPK) pathway. Together, these allow …