Systematic elucidation and in vivo validation of sequences enriched in hindbrain transcriptional control

Abstract

Illuminating the primary sequence encryption of enhancers is central to understanding the regulatory architecture of genomes. We have developed a machine learning approach to decipher motif patterns of hindbrain enhancers and identify40,000 sequences in the human genome that we predict display regulatory control that includes the hindbrain. Consistent with their roles in hindbrain patterning, MEIS1, NKX6-1, as well as HOX and POU family binding motifs contributed strongly to this enhancer model. Predicted hindbrain enhancers are over-represented at genes expressed in hindbrain and associated with nervous system development, and primarily reside in the areas of open chromatin.In addition, 77 (0.2%) of these predictions are identifiedas hindbrain enhancerson the VISTA enhancer browser, and 26,000 (60%) overlap enhancer marks (H3K4me1 or H3K27ac). To validate these putative hindbrain enhancers, we selected 55 elementsdistributed throughout our predictions and 6 low scoring controls for evaluationin a zebrafish transgenic assay. When assayed in mosaic transgenic embryos, 51/55 elementsdirected expression in the CNS. Furthermore, 30/34 (88%) predicted enhancersanalyzed in stable zebrafish transgenic lines directed expression in the larval zebrafish hindbrain. Subsequent analysisof sequence fragments selected based upon motif clustering further confirmed the critical role of the motifs contributing to the classifier. Our results demonstrate the existence of a primary sequence code characteristic to hindbrain enhancers. This code can be accurately extracted using machine-learning approaches and applied successfully for de novo identification of hindbrain enhancers. This study represents a critical step towards the dissection of regulatory control in specific neuronal subtypes.