Classical and genomic prediction of synthetic open‐pollinated sweet corn performance in organic environments

Abstract

AbstractOpen‐pollinated cultivars provide a number of benefits for organic and smallholder farmers, allowing them to save seed, conduct on‐farm selection, and maintain on‐farm crop genetic diversity. Synthetic open‐pollinated cultivars provide advantages over traditional open‐pollinated cultivars for such farmers. The objective of this research was to determine the usefulness of marker‐based prediction models relative to structured mating design‐based predictions for selecting untested sweet corn synthetic cultivars for organic production systems. This study used marker data and phenotypic data collected in 2015 and 2016 in 11 organic trials across six locations on 40 sweet corn (Zea mays L.) inbreds and 100 hybrid progeny formed from four disconnected North Carolina Design II (NC DII) mating blocks to predict performance of untested synthetic open‐pollinated sweet corn populations. In 2017, validation trials of 26 previously untested synthetic populations were grown in five organic environments to assess correlations between actual performance and performance predicted by genomic best linear unbiased prediction (GBLUP) or NC DII general combining abilities (GCAs). Correlations between values measured in 2017 validation trials and values predicted from the complete dataset ranged from 0.28 to 0.68 for GCA‐based predictions, from 0.25 to 0.67 for the additive GBLUP model, and from 0.28 to 0.68 for the additive plus dominance GBLUP model. In general, neither the genomic prediction model with solely additive effects nor genomic prediction with both additive and dominance effects demonstrated consistent increases in accuracy of predictions of synthetic population performance above predictions based solely on general combining ability. However, genomic prediction could be used to predict synthetic populations that shared alleles with the training set, even if the parents were never included in the training set, which is not possible with traditional general combining ability models.