A simple approach to predict growth stages in winter wheat (Triticum aestivum L.) combining prediction of a crop model and marker based prediction of the deviation to a reference cultivar: A case study in France

Predicting wheat growth stages using ecophysiological models is of particular interest as it allows anticipating important agricultural managements. Numerous ecophysiological models exist but they need cultivar-specific parameterization, which is often costly and time consuming. The work presented here proposes a simple approach to predict wheat growth stages using the allelic composition of wheat cultivars. It relies on using the prediction of a modified version of the ARCWHEAT model for a well parameterized reference cultivar (Soissons) and the marker-based predicted deviation in days to the reference cultivar. First, the deviations to the reference cultivar Soissons for the beginning of stem elongation (δZ30) and heading date (δZ55) were calculated for a large panel of cultivars. Analysis of variance showed prominent genetic effects for δZ30 and δZ55 and possible genotype × environment interactions (G × E) for δZ30. Genotypic means δZ30 and δZ55 were used in association genetics revealing 90 and 83 genetic markers associated to these traits, respectively. Multiple linear regression models predicting δZ30 using 11 genetic markers (R2 = 76%) or δZ55 using 17 markers (R2 = 85%) were obtained by a stepwise procedure. Marker PPD-D1 had the largest impact in both models. Finally, marker-based deviations added to the prediction for the reference cultivar Soissons allowed predicting Z30 or Z55 for a large independent validation dataset. The root mean square error of prediction for Z30 and Z55 using the approach proposed in this paper (6.8 and 4.7 days, respectively) was comparable to the one obtained using the conventional approach with cultivar-specific parameters values (6.5 and 4.1, respectively). The models proposed in this paper appeared sufficient in order to predict growth stages of cultivars which cannot be parameterized such as new cultivars coming out on the market. Moreover, genetic markers involved in the multiple linear regression models predicting δZ30 and δZ55 may provide interesting candidates to unravel new genes determining earliness in winter wheat.