Genetic gains and genomic prediction of selected cassava traits in Uganda

Abstract

Variety replacement tailored towards addressing the urgent and futuristic society needs is an important function of plant breeding. To ensure that superior varieties are developed fast enough to match the growing demand for food and industry, routine assessment of genetic gains and adoption of breeding methods that can fast-track variety replacement is critical. Accordingly, a study was conducted to determine (i) the annual rate of genetic gain for cassava traits selected between 1940 to 2019 in Uganda, and (ii) the prediction accuracy of genomic selection for resistance to virus diseases in cassava. To estimate genetic gains, thirty-two varieties developed between 1940 to 2019, were evaluated simultaneously in three major cassava production zones; central (Namulonge), eastern (Serere), and northern (Loro). Best linear unbiased predictors (BLUPs) of the genotypic value for each clone were obtained across environments and regressed on order of released year to estimate annual genetic gains. Results from genetic gain assessment showed that most genetic trends were quadratic. On average, dry matter content and resistance to cassava mosaic disease (CMD) increased by 0.1% and 1.9% per year, respectively, while annual genetic improvements in harvest index (0.0%) and fresh root yield (-5kg per ha or -0.03% per ha) were not substantial. For cassava brown streak disease (CBSD) resistance breeding which was only initiated in 2003, average annual genetic gains for resistances to CBSD foliar symptoms and CBSD root necrosis were 2.3% and 1.5% respectively. In order to determine whether genomic prediction would accurately predict CMD and CBSD severity, four genomic prediction models were evaluated using three breeding populations: cycle zero (C0), cycle one (C1) and the pre-breeding population that was developed independently. These models were; genomic best linear unbiased prediction (GBLUP), Bayesian ridge regression (BRR), Bayesian least absolute shrinkage and selection operator (BL) and reproducing kernel Hilbert spaces (RKHS). Cross validation prediction accuracies were performed using the four genomic prediction models in each of the three populations. For independent validation, C0 and C1 were used singly and in combination to predict resistance to CMD and CBSD in the pre-breeding population. Results showed that cross validation prediction accuracies were generally modest, ranging from 0.13 to 0.41 for resistance to CMD and 0.27 to 0.58 for resistance to CBSD. Generally low prediction accuracies were observed during independent validation, ranging from -0.07 to 0.24 for CBSD traits, and -0.05 to 0.34 for CMD traits. Overall, RKHS and GBLUP performed similarly and were appreciably superior to BRR and BL for prediction of resistance to CMD and CBSD. Based on the generated datasets, it was concluded that substantial genetic improvement for resistance to CMD and CBSD plus dry matter content was registered between 1940 to 2019. The results also demonstrated that the annual rate of genetic gain for cassava yield in Uganda was not sufficient to achieve the desired output necessary to reach the cassava production demand predicted for 2050. On the other hand, given that the observed prediction accuracies of genomic selection based on independent validations exceeded 0.30, results indicated that genomic selection could exceed phenotypic selection in genetic gain per unit time.