DNA sequence and shape are predictive for meiotic crossovers throughout the plant kingdom

• Published in: The Plant journal : for cell and molecular biology Vol. 95; no. 4; pp. 686 - 699
• Main Authors: Demirci, Sevgin, Peters, Sander A., Ridder, Dick, Dijk, Aalt D.J.
• Format: Journal Article
• Language: English
• Published: England Blackwell Publishing Ltd 01.08.2018
• Subjects: Annotations; Arabidopsis; Arabidopsis thaliana; Corn; crossover; Crossovers; Deoxyribonucleic acid; DNA; DNA shape; Flowers & plants; genome accessibility; Genomes; Learning algorithms; machine learning; maize; Mathematical models; Meiosis; meiotic recombination; Model accuracy; Nucleotide sequence; Oryza sativa; Plant breeding; prediction; rice; Solanum lycopersicum; Species; tomato; Tomatoes; Zea mays; Arabidopsis thaliana; meiotic recombination; crossover; DNA shape; genome accessibility; prediction; tomato; machine learning; rice; maize
• ISSN: 0960-7412; 1365-313X
• DOI: 10.1111/tpj.13979

Summary A better understanding of genomic features influencing the location of meiotic crossovers (COs) in plant species is both of fundamental importance and of practical relevance for plant breeding. Using CO positions with sufficiently high resolution from four plant species [Arabidopsis thaliana, Solanum lycopersicum (tomato), Zea mays (maize) and Oryza sativa (rice)] we have trained machine‐learning models to predict the susceptibility to CO formation. Our results show that CO occurrence within various plant genomes can be predicted by DNA sequence and shape features. Several features related to genome content and to genomic accessibility were consistently either positively or negatively related to COs in all four species. Other features were found as predictive only in specific species. Gene annotation‐related features were especially predictive for maize, whereas in tomato and Arabidopsis propeller twist and helical twist (DNA shape features) and AT/TA dinucleotides were found to be the most important. In rice, high roll (another DNA shape feature) and low CA dinucleotide frequency in particular were found to be associated with CO occurrence. The accuracy of our models was sufficient for Arabidopsis and rice (area under receiver operating characteristic curve, AUROC > 0.5), and was high for tomato and maize (AUROC ≫ 0.5), demonstrating that DNA sequence and shape are predictive for meiotic COs throughout the plant kingdom. Significance Statement Understanding the genomic features influencing the location of crossovers is of fundamental importance for many areas of plant biology, but a consistent, simultaneous analysis of multiple plant species in order to compare the genomic determinants of crossovers is lacking. We apply machine learning to crossover datasets from four different plant species in order to develop predictive models for the occurrence of crossovers, and to learn about relevant and important features correlated with CO formation.