Multi-omics and machine learning reveal context-specific gene regulatory activities of PML::RARA in acute promyelocytic leukemia.

ABSTRACT

The PMLRARA fusion protein is the hallmark driver of Acute Promyelocytic Leukemia (APL) and disrupts retinoic acid signaling, leading to wide-scale gene expression changes and uncontrolled proliferation of myeloid precursor cells. While known to be recruited to binding sites across the genome, its impact on gene regulation and expression is under-explored. Using integrated multi-omics datasets, we characterize the influence of PMLRARA binding on gene expression and regulation in an inducible PMLRARA cell line model and APL patient ex vivo samples. We find that genes whose regulatory elements recruit PMLRARA are not uniformly transcriptionally repressed, as commonly suggested, but also may be upregulated or remain unchanged. We develop a computational machine learning implementation called Regulatory Element Behavior Extraction Learning to deconvolute the complex, local transcription factor binding site environment at PMLRARA bound positions to reveal distinct signatures that modulate how PMLRARA directs the transcriptional response.