Training Population Optimization for Prediction of Cassava Brown Streak Disease Resistance in West African Clones

• Main Authors: Alfred Ozimati, Robert Kawuki, Williams Esuma, Ismail Siraj Kayondo, Marnin Wolfe, Roberto Lozano, Ismail Rabbi, Peter Kulakow, Jean-Luc Jannink
• Format: Article
• Language: English
• Published: Oxford University Press 2018-12-01
• Series: G3: Genes, Genomes, Genetics
• Subjects: Cassava; genomic selection; training population; and cassava brown streak disease; Genomic Prediction; GenPred; Shared Data Resources
• Online Access: http://g3journal.org/lookup/doi/10.1534/g3.118.200710

Cassava production in the central, southern and eastern parts of Africa is under threat by cassava brown streak virus (CBSV). Yield losses of up to 100% occur in cases of severe infections of edible roots. Easy illegal movement of planting materials across African countries, and long-range movement of the virus vector (Bemisia tabaci) may facilitate spread of CBSV to West Africa. Thus, effort to pre-emptively breed for CBSD resistance in W. Africa is critical. Genomic selection (GS) has become the main approach for cassava breeding, as costs of genotyping per sample have declined. Using phenotypic and genotypic data (genotyping-by-sequencing), followed by imputation to whole genome sequence (WGS) for 922 clones from National Crops Resources Research Institute, Namulonge, Uganda as a training population (TP), we predicted CBSD symptoms for 35 genotyped W. African clones, evaluated in Uganda. The highest prediction accuracy (r = 0.44) was observed for cassava brown streak disease severity scored at three months (CBSD3s) in the W. African clones using WGS-imputed markers. Optimized TPs gave higher prediction accuracies for CBSD3s and CBSD6s than random TPs of the same size. Inclusion of CBSD QTL chromosome markers as kernels, increased prediction accuracies for CBSD3s and CBSD6s. Similarly, WGS imputation of markers increased prediction accuracies for CBSD3s and for cassava brown streak disease root severity (CBSDRs), but not for CBSD6s. Based on these results we recommend TP optimization, inclusion of CBSD QTL markers in genomic prediction models, and the use of high-density (WGS-imputed) markers for CBSD predictions across population.