Genetic Basis of Texture and Associated Traits in Cassava

Abstract

Cassava breeding efforts in Uganda have in the past largely focused on improving agronomic traits such as yield and disease resistance but little has been done to improve consumer traits. This has limited adoption of improved varieties among consumers. In Uganda, cassava is predominantly consumed in boiled form and varieties that have a soft texture are preferred. To boost integration of texture traits in routine breeding operations, this study was conducted with specific objectives to (1) assess the phenotypic variability for texture traits (softness, toughness, stiffness) and water absorption (WAB) in a cycle 2 (C2) genomic selection diverse panel, (2) determine the prediction accuracy of Near Infrared Spectroscopy (NIRS) for phenotyping tex ture traits and WAB30 and (3) identify genomic loci and polymorphisms linked to variation for texture traits and further determine genomic prediction accuracy for these traits. From study one, significant differences (p<0.001) were observed among 250 accessions for all texture traits and WAB30, though environment effects also significantly influenced (p<0.001) the variation observed across all traits. Broad sense heritability ranged from low (Soft_p; H2 = 0.38) to moderate (Gradient; H2 = 0.62) indicating substantial genetic control of these traits. Near infrared spectroscopy (NIRS) quantitative predictions for texture traits and WAB30 based on partial least squares regression (PLSR) were all low. Using qualitative machine learning models based on binary classification with support vector machines (SVM), WAB had the high est prediction accuracy and reliability (r2cv = 0.78, ĸcv = 0.46) when spectra were pre- treated with standard normal variate (SNV) in combination with gap segment derivatization (GAP). This finding is promising for the integration of NIRS in routine phenotyping activities for consumer preferred cooking traits in cassava breeding. Through marker-trait association mapping, 24 SNPs that explained a substantial proportion of phenotypic variation were found significantly associated with textural traits and WAB in boiled cassava. Most SNPs were found on chromosomes 4, 8, 17, 18, and a survey of the cassava ge nome v7.1 positioned these SNPs in the vicinity of several genes coding for cell wall modifying proteins including Manes.04G139200 (Soft_T; Betagalactosidase 1) and Manes.18G044401 (WAB30; Glycine rich ell wall structural protein). Assessment of genomic prediction accuracy for texture traits and WAB30 found that all pre dictions across traits remained low (r2cv ≤ 0.27). Also, the Bayes A, Bayesian Ridge Regression (BRR) and Reproducing Kernel Hilbert Spaces (RKHS) models generally gave better predic tion accuracies than GBLUP model. These findings though promising, point at the need for optimal training population size and composition to achieve higher prediction accura cies. Overall, this work demonstrated that tools for improving cassava for consumer preferred traits in Uganda are available and with improvements, could be directly integrated into breeding op erations. This initiative is crucial for developing relevant cassava varieties combining end-user preferences and superior agronomic performances for the millions of actors in the cassava value chain in Uganda