Genetic Variation and Trait Correlations in an East African Cassava Breeding Population for Genomic Selection

Abstract

Cassava (Manihot esculenta Crantz) is a major source of dietary carbohydrates for >700 million people globally. However, its long breeding cycle has slowed the rate of genetic gain for target traits. This study aimed to asses genetic variation, the level of inbreeding, and trait correlations in genomic selection breeding cycles. We used phenotypic and genotypic data from the National Crops Resources Research Institute (NaCRRI) foundation population (Cycle 0, C0) and the progeny (Cycle 1, C1) derived from crosses of 100 selected C0 clones as progenitors, both to evaluate and optimize genomic selection. The highest broad‐sense heritability (H2 = 0.95) and narrow‐sense heritability (h2 = 0.81) were recorded for cassava mosaic disease severity and the lowest for root weight per plot (H2 = 0.06 and h2 = 0.00). We observed the highest genetic correlation (rg= 0.80) between cassava brown streak disease root incidence measured at seedling and clonal stages of evaluation, suggesting the usefulness of seedling data in predicting clonal performance for cassava brown streak root necrosis. Similarly, high genetic correlations were observed between cassava brown streak disease severity (rg= 0.83) scored at 3 and 6 mo after planting (MAP) and cassava mosaic disease, scored at 3 and 6 MAP (rg= 0.95), indicating that data obtained on these two diseases at 6 MAP would suffice. Population differentiation between C0 and C1 was not well defined, implying that the 100 selected progenitors of C1 captured the diversity in the C0. Overall, genetic gain for most traits were observed from C0 to C1.

Genomic prediction and quantitative trait locus discovery in a cassava training population constructed from multiple breeding stages

“…Generally, the positive correlations of FYLD, RTNO, and SHTWT could be exploited for simultaneous trait improvements. We also observed a slight positive genetic correlation between MCMDS and MCBSDS (r = 0.3) but a very low association between MCBSDS and CBSDRS, which agrees with the findings in other studies (Ozimati et al, 2019;Rwegasira & Rey, 2012). The positive relationship between these two foliar diseases provides an opportunity to increase resistance against both concurrently.…”

“…Cassava brown streak disease severity, SHTWT, FYLD, and DM in the CETUKG trials and MCBSDS and DM in the Kibaha trials all had medium heritability estimates. Similar estimates were reported in other cassava populations (Kayondo et al, 2018;Ozimati et al, 2019;Wolfe et al, 2017). We observed lower FYLD heritability in AYTKIB than in any other trials.…”

On to the Next Chapter for Crop Breeding: Convergence with Data Science

“…There are many recent published examples of selection on relatively small scale data sets in crop and animal breeding (Gorjanc et al, 2018;Dimitrijevic and Horn, 2018;Ozimati et al, 2019;Rio et al, 2019;Sweeney et al, 2019). Linear methods such as GBLUP typically work quite well for building predictions for almost all applications.…”

Incorporating selfing to purge deleterious alleles in a cassava genomic selection program

“…The change in genotypic value was of course affected by the simulation parameters we 354 assumed: relatively high error variance among seedlings and low selection intensity during purging 355 selection. Empirical estimates of error variance or broad sense heritability for yield traits in cassava 356 seedlings have not been reported (e.g., Ozimati et al 2019). Such estimates would be difficult to 357 obtain given that all seedlings are genetically unique and cannot be replicated.…”

“…With GS breeding schemes being adopted in cassava breeding (Wolfe et al, 2017), there is a 69 higher risk of inbreeding because cycles of selection are more frequent and each cycle entails a 70 bottleneck (Ozimati et al, 2019). Ozimati et al (2019) compared inbreeding between C0 and C1 71 populations using the average of diagonal elements of a kinship matrix, as a measure of the 72 inbreeding coefficient. While they found less inbreeding in the C1 than in the C0 population, 73 they attributed that to a purposeful crossing design that prevented genomically similar 74 individuals from being crossed.…”