Modeling first order additive × additive epistasis improves accuracy of genomic prediction for sclerotinia stem rot resistance in canola

Abstract

The fungus Sclerotinia sclerotiorum infects hundreds of plant species including many crops. Resistance to this pathogen in canola (Brassica napus L. subsp. napus) is controlled by numerous quantitative trait loci (QTL). For such polygenic traits, genomic prediction may be useful for breeding as it can capture many QTL at once while also considering nonadditive genetic effects. Here, we test application of common regression models to genomic prediction of S. sclerotiorum resistance in canola in a diverse panel of 218 plants genotyped at 24,634 loci. Disease resistance was scored by infection with an aggressive isolate and monitoring over 3 wk. We found that including first-order additive × additive epistasis in linear mixed models (LMMs) improved accuracy of breeding value estimation between 3 and 40%, depending on method of assessment, and correlation between phenotypes and predicted total genetic values by 14%. Bayesian models performed similarly to or worse than genomic relationship matrix-based models for estimating breeding values or overall phenotypes from genetic values. Bayesian ridge regression, which is most similar to the genomic relationship matrix-based approach in the amount of shrinkage it applies to marker effects, was the most accurate of this family of models. This confirms several studies Abbreviations: AGG, Australian Grains Genebank; ASSYST, Associative expression and systems analysis of complex traits in oilseed rape/canola; AUDPC, area under the disease progress curve; BLUP, best linear unbiased predictor; DPI, days post inoculation; G-BLUP, genomic best linear unbiased predictor; GEBV, genomic estimated breeding value; h, square root of narrow-sense heritability; H 2 , broad-sense heritability; h 2 , narrow-sense heritability; LMM, linear mixed model; MCMC, Markov chain Monte Carlo; PSCL, proportion of soft or collapsed lesions; QTL, quantitative trait loci; RR-BLUP, ridge regression best linear unbiased predictor; SNP, single nucleotide polymorphism. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Genome-wide association mapping and genomic prediction for adult stage sclerotinia stem rot resistance in Brassica napus (L) under field environments

“…These results clearly demonstrate that genome-wide markers are efficient in predicting SSR resistance. None of the models outperform than the others, with an exception for CARR_20 environment, consistent with results obtained by other researchers 41 , 86 , 87 . The observed differences in the predictive abilities among the used models for the combENV traits were mostly 1 to 2 units, which were likely due to GS model’s underlying assumptions.…”

“…BRR produces homogeneous shrinkage of all marker effects towards zero and yields a normal distribution of the marker effects 89 . The consistent results were also reported by Derbyshire et al 41 ,where Bayesian models perform similar or worse than G-BLUP model to predict S. sclerotiorum resistance in B. napus . The predictive ability in this study ranging from medium to high, were comparable or higher than the estimated predictive ability of SSR resistance in two previous studies 18 , 41 .…”

“…However, true breeding values for this trait are typically unknown. Therefore, the prediction accuracy was indirectly estimated by dividing the correlation between GEBV and observed phenotypic value by an estimate of square root of heritability ( ) 41 , 62 . Thus, the prediction accuracy of each model was estimated by dividing the mean predictive ability by square root of heritability ( h 2 ).…”

“…The potential of GS in rapeseed/canola was investigated for various agronomic traits including resistance to blackleg disease 39 , 40 and concluded as a promising tool for rapeseed breeding. However, the application of genomic prediction for SSR resistance in rapeseed/canola has been limited to date 18 , 41 . Recently, Derbyshire et al 41 implemented GP on adult plant SSR resistance in B. napus and reported that the GS can be used for the improvement of S. sclerotiorum resistance.…”

“…However, the application of genomic prediction for SSR resistance in rapeseed/canola has been limited to date 18 , 41 . Recently, Derbyshire et al 41 implemented GP on adult plant SSR resistance in B. napus and reported that the GS can be used for the improvement of S. sclerotiorum resistance.…”