Epigenetic mechanisms are essential to establish and safeguard cellular identities in mammals. They dynamically regulate the expression of genes, transposable elements and higher-order chromatin structures. Consequently, these chromatin marks are indispensable for mammalian development and alterations often lead to disease, such as cancer. Bivalent promoters are especially important during differentiation and development. Here we used a genetic screen to identify new regulators of a bivalent repressed gene. We identify BEND3 as a regulator of hundreds of bivalent promoters, some of which it represses, and some of which it activates. We show that BEND3 is recruited to a CpG-containg consensus site that is present in multiple copies in many bivalent promoters. Besides having direct effect on the promoters it binds, the loss of BEND3 leads to genome-wide gains of DNA methylation, which are especially marked at regions normally protected by the TET enzymes. DNA hydroxymethylation is reduced in Bend3 mutant cells, possibly as consequence of altered gene expression leading to diminished alpha-ketoglutarate production, thus lowering TET activity. Our results clarify the direct and indirect roles of an important chromatin regulator, BEND3, and, more broadly, they shed light on the regulation of bivalent promoters.

The proper tissue-specific regulation of gene expression is essential for development and homeostasis in metazoans. However, the illegitimate expression of normally tissue-restricted genes-like testis- or placenta-specific genes-is frequently observed in tumors; this promotes transformation, but also allows immunotherapy. Two important questions are: how is the expression of these genes controlled in healthy cells? And how is this altered in cancer? To address these questions, we used an unbiased approach to test the ability of 350 distinct genetic or epigenetic perturbations to induce the illegitimate expression of over 40 tissue-restricted genes in primary human cells. We find that almost all of these genes are remarkably resistant to reactivation by a single alteration in signaling pathways or chromatin regulation. However, a few genes differ and are more readily activated; one is the placenta-expressed gene ADAM12, which promotes invasion. Using cellular systems, an animal model, and bioinformatics, we find that a non-canonical but druggable TGF-β/KAT2A/TAK1 axis controls ADAM12 induction in normal and cancer cells. More broadly, our data show that illegitimate gene expression in cancer is an heterogeneous phenomenon, with a few genes activatable by simple events, and most genes likely requiring a combination of events to become reactivated.