Location: Plant Science ResearchTitle: Genomic prediction for resistance to fusarium ear rot and fumonisin contamination in maize
|Holland, Jim - Jim|
|MARINO, T - North Carolina State University|
|MANCHING, H - University Of Delaware|
|WISSER, R - University Of Delaware|
Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/31/2020
Publication Date: N/A
Interpretive Summary: Fusarium ear rot disease of corn reduces yields and the fungus produces a toxin that can contaminate corn grain. We are breeding for resistance to this disease, but evaluating large numbers of corn varieties in a breeding program for this disease is difficult and requires multiple years to obtain useful data for selection. In this study, we evaluate the potential usefulness of genomic selection to improve the efficiency of selection for disease resistance. To do this, we obtained thousands of genetic markers from sequencing many corn varieties that were also evaluated in replicated multi-environment trials for Fusarium ear rot disease. We compared different statistical and machine learning modelling techniques for their ability to predict the resistance level of other varieties. We found that we had reasonably good prediction accuracy from standard genomic selection models, and there results suggest that we can perform genomic selection for this disease resistance trait faster than we can evaluate new lines in the field, so this may help us improve our selections for improving ear rot resistance.
Technical Abstract: Fusarium ear rot (FER) disease of maize (Zea mays L.) is caused by Fusarium verticillioides (Sacc.) Nirenberg, which produces fumonisin (FUM), a mycotoxin linked to human and animal health risks. Extensive field trials, laborious inoculations and ear evaluations, and expensive antibody assays are required to reliably assess resistances to FER and FUM contamination in breeding populations. To evaluate the potential utility of genomic selection to improve FER and FUM in maize, we genotyped 5049 SNPs on 451 S0:1 lines from a recurrent selection population. Two different partitions of the S0:1 evaluation data were made to test the ability of models trained on 252 or 202 lines evaluated at three locations in 2014-15 to predict FER and FUM of 199 or 249 different lines evaluated at three locations in 2016. Single-stage univariate and multivariate GBLUP models and two-stage GBLUP, Bayes C pi, Bayesian LASSO, and extreme gradient boosting models were compared for prediction. Optimal prediction accuracy of prediction for untested lines in a new year was 0.47 for FER and 0.66 for FUM. Bayesian models were best for predicting fumonisin in one set, despite no evidence for significant individual SNP-trait associations from genome-wide association study in the training sets; otherwise GBLUP models were best. These results suggest that genomic selection can help improve resistance to Fusarium ear rot and fumonisin contamination in an applied breeding program.