Due to the importance of glucosinolates and their hydrolysis products in human nutrition and plant defense, optimizing the content of these compounds is a frequent breeding objective for Brassica crops. Toward this goal, we investigated the feasibility of using models built from relative transcript abundance data for the prediction of glucosinolate and hydrolysis product concentrations in broccoli. We report that predictive models explaining at least 50% of the variation for a number of glucosinolates and their hydrolysis products can be built for prediction within the same season, but prediction accuracy decreased when using models built from one season's data for prediction of an opposing season. This method of phytochemical profile prediction could potentially allow for lower phytochemical phenotyping costs and larger breeding populations. This, in turn, could improve selection efficiency for phase II induction potential, a type of chemopreventive bioactivity, by allowing for the quick and relatively cheap content estimation of phytochemicals known to influence the trait.
Keywords: Brassica; broccoli; glucosinolate; predictive modeling; qRT-PCR.