Genome-wide studies reveal the function and evolution of DNA shape

2014-11-03Gene regulation is orchestrated by the binding of transcription factors (TFs) to their specific target DNA sequence. An increasing number of structural biology and genomics studies associate protein-DNA binding with the recognition of three-dimensional DNA structure, or “DNA shape”. Retrieval of local nuances in DNA shape is therefore a fundamental step to understand the general mechanisms of gene regulation. ❧ Due to the limitation of X-ray crystallography and NMR spectroscopy, molecular simulations have long been the only available approach to deduce atomic information on intrinsic DNA structure. Despite significant recent advances, the bottleneck for any molecular simulation is that only a limited number of DNA sequences can be pursued. To overcome this limitation, we have developed a high-throughput (HT) method for DNA shape prediction based on the mining of atomic resolution data obtained from Monte Carlo (MC) simulations of numerous DNA fragments. ❧ Applying the HT method to predict the DNA shape of many SELEX-seq-identified Hox binding sites, we revealed that closely related Hox paralogs achieve their distinct function in vivo through differential binding preference for DNA sequences with distinct minor groove topographies. Analyzing millions of DNA backbone cleavage events on naked genomic DNA conducted by the endonuclease DNase I, we showed that CpG methylation modulates DNase I-DNA interaction strengths via the remodeling of DNA shape. Examining the effect of millions of single nucleotide variants (SNVs) on local DNA minor groove width (MGW) pattern, we proved that purifying selection acts on DNA shape as well as nucleotide sequence. ❧ To gain a comprehensive, systematic and quantitative view of DNA shape effects on protein-DNA binding, we integrated intuitive DNA shape features in the modeling of TF binding specificities using statistical machine learning approaches. We compared the quantitative predictions of DNA binding specificities for 68 mouse and human TFs using shape-augmented models with those using sequence-based models. Based on our model evaluation, which included cross-validation on specific PBM array designs, testing across different PBM array designs, and using PBM-trained models to predict relative binding affinities derived from SELEX-seq data, we showed that shape-augmented models compared favorably to sequence-based models. Besides, whereas adding either k-mer or DNA shape features to sequence-only models improves the modeling of DNA binding specificities, using DNA shape that represents interdependencies implicitly with smaller numbers of features yields more efficient models. ❧ In summary, my work on the DNA shape prediction and its application adds a new dimension to genome-wide studies of gene regulation and evolution